Datasets:

---

bigcode

/

jupyter-parsed

like 4

Follow

BigCode 1.35k

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Visualizing and Comparing LIS Output", "_____no_output_____" ], [ "```{figure} ./images/nasa-lis-combined-logos.png\n---\nwidth: 300px\n---\n```", "_____no_output_____" ], [ "## LIS Output Primer\n\nLIS writes model state variables to disk at a frequency selected by the user (e.g., 6-hourly, daily, monthly). The LIS output we will be exploring was originally generated as *daily* NetCDF files, meaning one NetCDF was written per simulated day. We have converted these NetCDF files into a [Zarr](https://zarr.readthedocs.io/en/stable/) store for improved performance in the cloud.", "_____no_output_____" ], [ "## Import Libraries", "_____no_output_____" ] ], [ [ "# interface to Amazon S3 filesystem\nimport s3fs\n\n# interact with n-d arrays\nimport numpy as np\nimport xarray as xr\n\n# interact with tabular data (incl. spatial)\nimport pandas as pd\nimport geopandas as gpd\n\n# interactive plots\nimport holoviews as hv\nimport geoviews as gv\nimport hvplot.pandas\nimport hvplot.xarray\n\n# used to find nearest grid cell to a given location\nfrom scipy.spatial import distance\n\n# set bokeh as the holoviews plotting backend\nhv.extension('bokeh')", "_____no_output_____" ] ], [ [ "## Load the LIS Output\n\nThe `xarray` library makes working with labelled n-dimensional arrays easy and efficient. If you're familiar with the `pandas` library it should feel pretty familiar.\n\nHere we load the LIS output into an `xarray.Dataset` object:", "_____no_output_____" ] ], [ [ "# create S3 filesystem object\ns3 = s3fs.S3FileSystem(anon=False)\n\n# define the name of our S3 bucket\nbucket_name = 'eis-dh-hydro/SNOWEX-HACKWEEK'\n\n# define path to store on S3\nlis_output_s3_path = f's3://{bucket_name}/DA_SNODAS/SURFACEMODEL/LIS_HIST.d01.zarr/'\n\n# create key-value mapper for S3 object (required to read data stored on S3)\nlis_output_mapper = s3.get_mapper(lis_output_s3_path)\n\n# open the dataset\nlis_output_ds = xr.open_zarr(lis_output_mapper, consolidated=True)\n\n# drop some unneeded variables\nlis_output_ds = lis_output_ds.drop_vars(['_history', '_eis_source_path'])", "_____no_output_____" ] ], [ [ "## Explore the Data\n\nDisplay an interactive widget for inspecting the dataset by running a cell containing the variable name. Expand the dropdown menus and click on the document and database icons to inspect the variables and attributes.", "_____no_output_____" ] ], [ [ "lis_output_ds", "_____no_output_____" ] ], [ [ "### Accessing Attributes", "_____no_output_____" ], [ "Dataset attributes (metadata) are accessible via the `attrs` attribute:", "_____no_output_____" ] ], [ [ "lis_output_ds.attrs", "_____no_output_____" ] ], [ [ "### Accessing Variables\n\nVariables can be accessed using either **dot notation** or **square bracket notation**:", "_____no_output_____" ] ], [ [ "# dot notation\nlis_output_ds.SnowDepth_tavg", "_____no_output_____" ], [ "# square bracket notation\nlis_output_ds['SnowDepth_tavg']", "_____no_output_____" ] ], [ [ "#### Which syntax should I use?\nWhile both syntaxes perform the same function, the square-bracket syntax is useful when interacting with a dataset programmatically. For example, we can define a variable `varname` that stores the name of the variable in the dataset we want to access and then use that with the square-brackets notation:", "_____no_output_____" ] ], [ [ "varname = 'SnowDepth_tavg'\n\nlis_output_ds[varname]", "_____no_output_____" ] ], [ [ "The dot notation syntax will not work this way because `xarray` tries to find a variable in the dataset named `varname` instead of the value of the `varname` variable. When `xarray` can't find this variable, it throws an error:", "_____no_output_____" ] ], [ [ "# uncomment and run the code below to see the error\n\n# varname = 'SnowDepth_tavg'\n\n# lis_output_ds.varname", "_____no_output_____" ] ], [ [ "### Dimensions and Coordinate Variables\n\nThe dimensions and coordinate variable fields put the \"*labelled*\" in \"labelled n-dimensional arrays\":\n\n* **Dimensions:** labels for each dimension in the dataset (e.g., `time`)\n* **Coordinates:** labels for indexing along dimensions (e.g., `'2019-01-01'`)\n\nWe can use these labels to select, slice, and aggregate the dataset.", "_____no_output_____" ], [ "#### Selecting/Subsetting\n\n`xarray` provides two methods for selecting or subsetting along coordinate variables:\n\n* index selection: `ds.isel(time=0)`\n* value selection `ds.sel(time='2019-01-01')`\n\nFor example, we can select the first timestep from our dataset using index selection by passing the dimension name as a keyword argument:", "_____no_output_____" ] ], [ [ "# remember: python indexes start at 0\nlis_output_ds.isel(time=0)", "_____no_output_____" ] ], [ [ "Or we can use value selection to select based on the coordinate(s) (think \"labels\") of a given dimension:", "_____no_output_____" ] ], [ [ "lis_output_ds.sel(time='2018-01-01')", "_____no_output_____" ] ], [ [ "The `.sel()` approach also allows the use of shortcuts in some cases. For example, here we select all timesteps in the month of January 2018:", "_____no_output_____" ] ], [ [ "lis_output_ds.sel(time='2018-01')", "_____no_output_____" ] ], [ [ "Select a custom range of dates using Python's built-in `slice()` method:", "_____no_output_____" ] ], [ [ "lis_output_ds.sel(time=slice('2018-01-01', '2018-01-15'))", "_____no_output_____" ] ], [ [ "#### Latitude and Longitude\n\nYou may have noticed that latitude (`lat`) and longitude (`lon`) are listed as data variables, not coordinate variables. This dataset would be easier to work with if `lat` and `lon` were coordinate variables and dimensions. Here we define a helper function that reads the spatial information from the dataset attributes, generates arrays containing the `lat` and `lon` values, and appends them to the dataset:", "_____no_output_____" ] ], [ [ "def add_latlon_coords(dataset: xr.Dataset)->xr.Dataset:\n \"\"\"Adds lat/lon as dimensions and coordinates to an xarray.Dataset object.\"\"\"\n \n # get attributes from dataset\n attrs = dataset.attrs\n \n # get x, y resolutions\n dx = round(float(attrs['DX']), 3)\n dy = round(float(attrs['DY']), 3)\n \n # get grid cells in x, y dimensions\n ew_len = len(dataset['east_west'])\n ns_len = len(dataset['north_south'])\n \n # get lower-left lat and lon\n ll_lat = round(float(attrs['SOUTH_WEST_CORNER_LAT']), 3)\n ll_lon = round(float(attrs['SOUTH_WEST_CORNER_LON']), 3)\n \n # calculate upper-right lat and lon\n ur_lat = ll_lat + (dy * ns_len)\n ur_lon = ll_lon + (dx * ew_len)\n \n # define the new coordinates\n coords = {\n # create an arrays containing the lat/lon at each gridcell\n 'lat': np.linspace(ll_lat, ur_lat, ns_len, dtype=np.float32, endpoint=False),\n 'lon': np.linspace(ll_lon, ur_lon, ew_len, dtype=np.float32, endpoint=False)\n }\n \n lon_attrs = dataset.lon.attrs\n lat_attrs = dataset.lat.attrs\n \n # rename the original lat and lon variables\n dataset = dataset.rename({'lon':'orig_lon', 'lat':'orig_lat'})\n # rename the grid dimensions to lat and lon\n dataset = dataset.rename({'north_south': 'lat', 'east_west': 'lon'})\n # assign the coords above as coordinates\n dataset = dataset.assign_coords(coords)\n dataset.lon.attrs = lon_attrs\n dataset.lat.attrs = lat_attrs\n \n return dataset", "_____no_output_____" ] ], [ [ "Now that the function is defined, let's use it to append `lat` and `lon` coordinates to the LIS output:", "_____no_output_____" ] ], [ [ "lis_output_ds = add_latlon_coords(lis_output_ds)", "_____no_output_____" ] ], [ [ "Inspect the dataset:", "_____no_output_____" ] ], [ [ "lis_output_ds", "_____no_output_____" ] ], [ [ "Now `lat` and `lon` are listed as coordinate variables and have replaced the `north_south` and `east_west` dimensions. This will make it easier to spatially subset the dataset!", "_____no_output_____" ], [ "#### Basic Spatial Subsetting", "_____no_output_____" ], [ "We can use the `slice()` function we used above on the `lat` and `lon` dimensions to select data between a range of latitudes and longitudes:", "_____no_output_____" ] ], [ [ "lis_output_ds.sel(lat=slice(37, 41), lon=slice(-110, -101))", "_____no_output_____" ] ], [ [ "Notice how the sizes of the `lat` and `lon` dimensions have decreased.", "_____no_output_____" ], [ "#### Subset Across Multiple Dimensions", "_____no_output_____" ], [ "Select snow depth for Jan 2017 within a range of lat/lon:", "_____no_output_____" ] ], [ [ "# define a range of dates to select\nwy_2018_slice = slice('2017-10-01', '2018-09-30')\nlat_slice = slice(37, 41)\nlon_slice = slice(-109, -102)\n\n# select the snow depth and subset to wy_2018_slice\nsnd_CO_wy2018_ds = lis_output_ds['SnowDepth_tavg'].sel(time=wy_2018_slice, lat=lat_slice, lon=lon_slice)\n\n# inspect resulting dataset\nsnd_CO_wy2018_ds", "_____no_output_____" ] ], [ [ "### Plotting\n\nWe've imported two plotting libraries:\n\n* `matplotlib`: static plots\n* `hvplot`: interactive plots\n\nWe can make a quick `matplotlib`-based plot for the subsetted data using the `.plot()` function supplied by `xarray.Dataset` objects. For this example, we'll select one day and plot it:", "_____no_output_____" ] ], [ [ "# simple matplotlilb plot\nsnd_CO_wy2018_ds.sel(time='2018-01-01').plot()", "_____no_output_____" ] ], [ [ "Similarly we can make an interactive plot using the `hvplot` accessor and specifying a `quadmesh` plot type:", "_____no_output_____" ] ], [ [ "# hvplot based map\nsnd_CO_20180101_plot = snd_CO_wy2018_ds.sel(time='2018-01-01').hvplot.quadmesh(geo=True, rasterize=True, project=True,\n xlabel='lon', ylabel='lat', cmap='viridis',\n tiles='EsriImagery')\n\nsnd_CO_20180101_plot", "_____no_output_____" ] ], [ [ "Pan, zoom, and scroll around the map. Hover over the LIS data to see the data values.", "_____no_output_____" ], [ "If we try to plot more than one time-step `hvplot` will also provide a time-slider we can use to scrub back and forth in time:", "_____no_output_____" ] ], [ [ "snd_CO_wy2018_ds.sel(time='2018-01').hvplot.quadmesh(geo=True, rasterize=True, project=True,\n xlabel='lon', ylabel='lat', cmap='viridis',\n tiles='EsriImagery')", "_____no_output_____" ] ], [ [ "From here on out we will stick with `hvplot` for plotting.", "_____no_output_____" ], [ "#### Timeseries Plots\n\nWe can generate a timeseries for a given grid cell by selecting and calling the plot function:", "_____no_output_____" ] ], [ [ "# define point to take timeseries (note: must be present in coordinates of dataset)\nts_lon, ts_lat = (-105.65, 40.35)\n\n# plot timeseries (hvplot knows how to plot based on dataset's dimensionality!)\nsnd_CO_wy2018_ds.sel(lat=ts_lat, lon=ts_lon).hvplot(title=f'Snow Depth Timeseries @ Lon: {ts_lon}, Lat: {ts_lat}',\n xlabel='Date', ylabel='Snow Depth (m)') + \\\n snd_CO_20180101_plot * gv.Points([(ts_lon, ts_lat)]).opts(size=10, color='red')\n \n", "_____no_output_____" ] ], [ [ "In the next section we'll learn how to create a timeseries over a broader area.", "_____no_output_____" ], [ "## Aggregation\n\nWe can perform aggregation operations on the dataset such as `min()`, `max()`, `mean()`, and `sum()` by specifying the dimensions along which to perform the calculation.\n\nFor example we can calculate the mean and maximum snow depth at each grid cell over water year 2018 as follows:", "_____no_output_____" ] ], [ [ "# calculate the mean at each grid cell over the time dimension\nmean_snd_CO_wy2018_ds = snd_CO_wy2018_ds.mean(dim='time')\nmax_snd_CO_wy2018_ds = snd_CO_wy2018_ds.max(dim='time')\n\n# plot the mean and max snow depth\nmean_snd_CO_wy2018_ds.hvplot.quadmesh(geo=True, rasterize=True, project=True,\n xlabel='lon', ylabel='lat', cmap='viridis',\n tiles='EsriImagery', title='Mean Snow Depth - WY2018') + \\\n max_snd_CO_wy2018_ds.hvplot.quadmesh(geo=True, rasterize=True, project=True,\n xlabel='lon', ylabel='lat', cmap='viridis',\n tiles='EsriImagery', title='Max Snow Depth - WY2018')", "_____no_output_____" ] ], [ [ "### Area Average", "_____no_output_____" ] ], [ [ "# take area-averaged mean at each timestep\nmean_snd_CO_wy2018_ds = snd_CO_wy2018_ds.mean(['lat', 'lon'])\n\n# inspect the dataset\nmean_snd_CO_wy2018_ds", "_____no_output_____" ], [ "# plot timeseries (hvplot knows how to plot based on dataset's dimensionality!)\nmean_snd_CO_wy2018_ds.hvplot(title='Mean LIS Snow Depth for Colorado', xlabel='Date', ylabel='Snow Depth (m)')", "_____no_output_____" ] ], [ [ "## Comparing LIS Output\n\nNow that we're familiar with the LIS output, let's compare it to two other datasets: SNODAS (raster) and SNOTEL (point).", "_____no_output_____" ], [ "### LIS (raster) vs. SNODAS (raster)", "_____no_output_____" ], [ "First, we'll load the SNODAS dataset which we also have hosted on S3 as a Zarr store:", "_____no_output_____" ] ], [ [ "# load SNODAS dataset\n\n#snodas depth\nkey = \"SNODAS/snodas_snowdepth_20161001_20200930.zarr\" \nsnodas_depth_ds = xr.open_zarr(s3.get_mapper(f\"{bucket_name}/{key}\"), consolidated=True)\n\n# apply scale factor to convert to meters (0.001 per SNODAS user guide)\nsnodas_depth_ds = snodas_depth_ds * 0.001", "_____no_output_____" ] ], [ [ "Next we define a helper function to extract the (lon, lat) of the nearest grid cell to a given point:", "_____no_output_____" ] ], [ [ "def nearest_grid(ds, pt):\n \n \"\"\"\n Returns the nearest lon and lat to pt in a given Dataset (ds).\n \n pt : input point, tuple (longitude, latitude)\n output:\n lon, lat\n \"\"\"\n \n if all(coord in list(ds.coords) for coord in ['lat', 'lon']):\n df_loc = ds[['lon', 'lat']].to_dataframe().reset_index()\n else:\n df_loc = ds[['orig_lon', 'orig_lat']].isel(time=0).to_dataframe().reset_index()\n \n loc_valid = df_loc.dropna()\n pts = loc_valid[['lon', 'lat']].to_numpy()\n idx = distance.cdist([pt], pts).argmin()\n \n return loc_valid['lon'].iloc[idx], loc_valid['lat'].iloc[idx]", "_____no_output_____" ] ], [ [ "The next cell will look pretty similar to what we did earlier to plot a timeseries of a single point in the LIS data. The general steps are:\n\n* Extract the coordinates of the SNODAS grid cell nearest to our LIS grid cell (`ts_lon` and `ts_lat` from earlier)\n* Subset the SNODAS and LIS data to the grid cells and date ranges of interest\n* Create the plots!", "_____no_output_____" ] ], [ [ "# get lon, lat of snodas grid cell nearest to the LIS coordinates we used earlier\nsnodas_ts_lon, snodas_ts_lat = nearest_grid(snodas_depth_ds, (ts_lon, ts_lat))\n\n# define a date range to plot (shorter = quicker for demo)\nstart_date, end_date = ('2018-01-01', '2018-03-01')\nplot_daterange = slice(start_date, end_date)\n\n# select SNODAS grid cell and subset to plot_daterange\nsnodas_snd_subset_ds = snodas_depth_ds.sel(lon=snodas_ts_lon,\n lat=snodas_ts_lat,\n time=plot_daterange)\n\n# select LIS grid cell and subset to plot_daterange\nlis_snd_subset_ds = lis_output_ds['SnowDepth_tavg'].sel(lat=ts_lat,\n lon=ts_lon,\n time=plot_daterange)\n\n# create SNODAS snow depth plot\nsnodas_snd_plot = snodas_snd_subset_ds.hvplot(label='SNODAS')\n\n# create LIS snow depth plot\nlis_snd_plot = lis_snd_subset_ds.hvplot(label='LIS')\n\n# create SNODAS vs LIS snow depth plot\nlis_vs_snodas_snd_plot = (lis_snd_plot * snodas_snd_plot)\n\n# display the plot\nlis_vs_snodas_snd_plot.opts(title=f'Snow Depth @ Lon: {ts_lon}, Lat: {ts_lat}',\n legend_position='right',\n xlabel='Date',\n ylabel='Snow Depth (m)')", "_____no_output_____" ] ], [ [ "### LIS (raster) vs. SNODAS (raster) vs. SNOTEL (point)\n\nNow let's add SNOTEL point data to our plot.\n\nFirst, we're going to define some helper functions to load the SNOTEL data:", "_____no_output_____" ] ], [ [ "# load csv containing metadata for SNOTEL sites in a given state (e.g,. 'colorado')\ndef load_site(state):\n \n # define the path to the file\n key = f\"SNOTEL/snotel_{state}.csv\"\n \n # load the csv into a pandas DataFrame\n df = pd.read_csv(s3.open(f's3://{bucket_name}/{key}', mode='r'))\n \n return df\n\n# load SNOTEL data for a specific site\ndef load_snotel_txt(state, var):\n \n # define the path to the file\n key = f\"SNOTEL/snotel_{state}{var}_20162020.txt\"\n \n # determine how many lines to skip in the file (they start with #)\n fh = s3.open(f\"{bucket_name}/{key}\")\n lines = fh.readlines()\n skips = sum(1 for ln in lines if ln.decode('ascii').startswith('#'))\n \n # load the data into a pandas DataFrame\n df = pd.read_csv(s3.open(f\"s3://{bucket_name}/{key}\"), skiprows=skips)\n \n # convert the Date column from strings to datetime objects\n df['Date'] = pd.to_datetime(df['Date'])\n return df", "_____no_output_____" ] ], [ [ "For the purposes of this tutorial let's load the SNOTEL data for sites in Colorado. We'll pick one site to plot in a few cells.", "_____no_output_____" ] ], [ [ "# load SNOTEL snow depth for Colorado into a dictionary\nsnotel_depth = {'CO': load_snotel_txt('CO', 'depth')}", "_____no_output_____" ] ], [ [ "We'll need another helper function to load the depth data:", "_____no_output_____" ] ], [ [ "# get snotel depth\ndef get_depth(state, site, start_date, end_date):\n \n # grab the depth for the given state (e.g., CO)\n df = snotel_depth[state]\n \n # define a date range mask\n mask = (df['Date'] >= start_date) & (df['Date'] <= end_date)\n \n # use mask to subset between time range\n df = df.loc[mask]\n \n # extract timeseries for the given site\n return pd.concat([df.Date, df.filter(like=site)], axis=1).set_index('Date')", "_____no_output_____" ] ], [ [ "Load the site metadata for Colorado:", "_____no_output_____" ] ], [ [ "co_sites = load_site('colorado')\n\n# peek at the first 5 rows\nco_sites.head()", "_____no_output_____" ] ], [ [ "The point we've been using so far in the tutorial actually corresponds to the coordinates for the Bear Lake SNOTEL site! Let's extract the site data for that point:", "_____no_output_____" ] ], [ [ "# get the depth data by passing the site name to the get_depth() function\nbear_lake_snd_df = get_depth('CO', 'Bear Lake (322)', start_date, end_date)\n\n# convert from cm to m\nbear_lake_snd_df = bear_lake_snd_df / 100", "_____no_output_____" ] ], [ [ "Now we're ready to plot:", "_____no_output_____" ] ], [ [ "# create SNOTEL plot\nbear_lake_plot = bear_lake_snd_df.hvplot(label='SNOTEL')\n\n# combine the SNOTEl plot with the LIS vs SNODAS plot\n(bear_lake_plot * lis_vs_snodas_snd_plot).opts(title=f'Snow Depth @ Lon: {ts_lon}, Lat: {ts_lat}', legend_position='right')", "_____no_output_____" ] ], [ [ "# Conclusion\n\nYou should now be more familiar with LIS data and how to interact with it in Python. The code in this notebook is a great jumping off point for developing more advanced comparisons and interactive widgets. For an example of what is possible, open the next notebook and run all the cells (Run > Run All Cells). After a few minutes, two interactive widgets will appear that allow you to explore and compare LIS output with SNODAS and SNOTEL data.", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown" ]

[ [ "markdown", "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "import os\n\nos.chdir('../code/yolov5/')\n\nos.getcwd()", "_____no_output_____" ] ], [ [ "``` python\n!python train.py --img 160 --batch 4 --epochs 5 --data ./data/train_imgs_sliced_160.yaml --cfg ./models/yolov5s.yaml --weights '' \n```", "_____no_output_____" ], [ "``` python\n!python train.py --img 640 --batch 10 --epochs 100 --data ./data/train_imgs_sliced_640_val.yaml --cfg ./models/yolov5_tile.yaml --weights '' \n``` ", "_____no_output_____" ], [ "``` python\n%run train.py --img 320 --batch 5 --epochs 5 --data ./data/train_imgs_sliced_320.yaml --cfg ./models/yolov5s.yaml --weights '' \n```", "_____no_output_____" ], [ "%run train.py --img 320 --batch 5 --epochs 60 --data ./data/train_imgs_sliced_320_val.yaml --cfg ./models/yolov5st.yaml --weights ./runs/train/exp40/weights/last.pt", "_____no_output_____" ] ], [ [ "%run train.py --img 320 --batch 5 --epochs 60 --data ./data/train_imgs_sliced_320_val.yaml --weights ./runs/train/exp40/weights/last.pt", "Using torch 1.6.0+cu101 CUDA:0 (GeForce RTX 2060, 6144MB)\n\nNamespace(adam=False, batch_size=5, bucket='', cache_images=False, cfg='', data='./data/train_imgs_sliced_320_val.yaml', device='', epochs=60, evolve=False, exist_ok=False, global_rank=-1, hyp='data/hyp.scratch.yaml', image_weights=False, img_size=[320, 320], local_rank=-1, log_imgs=16, multi_scale=False, name='exp', noautoanchor=False, nosave=False, notest=False, project='runs/train', rect=False, resume=False, save_dir='runs\\\\train\\\\exp41', single_cls=False, sync_bn=False, total_batch_size=5, weights='./runs/train/exp40/weights/last.pt', workers=8, world_size=1)\nStart Tensorboard with \"tensorboard --logdir runs/train\", view at http://localhost:6006/\nHyperparameters {'lr0': 0.01, 'lrf': 0.2, 'momentum': 0.937, 'weight_decay': 0.0005, 'warmup_epochs': 3.0, 'warmup_momentum': 0.8, 'warmup_bias_lr': 0.1, 'box': 0.05, 'cls': 0.5, 'cls_pw': 1.0, 'obj': 1.0, 'obj_pw': 1.0, 'iou_t': 0.2, 'anchor_t': 4.0, 'fl_gamma': 0.0, 'hsv_h': 0.015, 'hsv_s': 0.7, 'hsv_v': 0.4, 'degrees': 0.0, 'translate': 0.1, 'scale': 0.5, 'shear': 0.0, 'perspective': 0.0, 'flipud': 0.0, 'fliplr': 0.5, 'mosaic': 1.0, 'mixup': 0.0}\n\n from n params module arguments \n 0 -1 1 3520 models.common.Focus [3, 32, 3] \n 1 -1 1 18560 models.common.Conv [32, 64, 3, 2] \n 2 -1 1 19904 models.common.BottleneckCSP [64, 64, 1] \n 3 -1 1 73984 models.common.Conv [64, 128, 3, 2] \n 4 -1 1 161152 models.common.BottleneckCSP [128, 128, 3] \n 5 -1 1 295424 models.common.Conv [128, 256, 3, 2] \n 6 -1 1 641792 models.common.BottleneckCSP [256, 256, 3] \n 7 -1 1 1180672 models.common.Conv [256, 512, 3, 2] \n 8 -1 1 656896 models.common.SPP [512, 512, [5, 9, 13]] \n 9 -1 1 1248768 models.common.BottleneckCSP [512, 512, 1, False] \n 10 -1 1 131584 models.common.Conv [512, 256, 1, 1] \n 11 -1 1 0 torch.nn.modules.upsampling.Upsample [None, 2, 'nearest'] \n 12 [-1, 6] 1 0 models.common.Concat [1] \n 13 -1 1 378624 models.common.BottleneckCSP [512, 256, 1, False] \n 14 -1 1 33024 models.common.Conv [256, 128, 1, 1] \n 15 -1 1 0 torch.nn.modules.upsampling.Upsample [None, 2, 'nearest'] \n 16 [-1, 4] 1 0 models.common.Concat [1] \n 17 -1 1 95104 models.common.BottleneckCSP [256, 128, 1, False] \n 18 -1 1 147712 models.common.Conv [128, 128, 3, 2] \n 19 [-1, 14] 1 0 models.common.Concat [1] \n 20 -1 1 313088 models.common.BottleneckCSP [256, 256, 1, False] \n 21 -1 1 590336 models.common.Conv [256, 256, 3, 2] \n 22 [-1, 10] 1 0 models.common.Concat [1] \n 23 -1 1 1248768 models.common.BottleneckCSP [512, 512, 1, False] \n 24 [17, 20, 23] 1 32364 models.yolo.Detect [7, [[9, 9, 14, 14, 23, 20], [61, 25, 34, 50, 55, 108], [124, 49, 237, 85, 111, 224]], [128, 256, 512]]\nModel Summary: 283 layers, 7271276 parameters, 7271276 gradients\n\nTransferred 370/370 items from ./runs/train/exp40/weights/last.pt\nOptimizer groups: 62 .bias, 70 conv.weight, 59 other\nScanning '..\\train_imgs_sliced_all_320_val6\\labels.cache' for images and labels... 28267 found, 0 missing, 0 empty, 0 c\nScanning '..\\train_imgs_sliced_all_320_val\\labels.cache' for images and labels... 14548 found, 0 missing, 0 empty, 0 co\nImage sizes 320 train, 320 test\nUsing 5 dataloader workers\nLogging results to runs\\train\\exp41\nStarting training for 60 epochs...\n\n Epoch gpu_mem box obj cls total targets img_size\n 0%| | 0/5654 [00:00<?, ?it/s]" ], [ "%run train.py --img 320 --batch 5 --epochs 60 --data ./data/train_imgs_sliced_320_val.yaml --cfg ./models/yolov5s_se.yaml --weights ''", "Using torch 1.6.0+cu101 CUDA:0 (GeForce RTX 2060, 6144MB)\n\nNamespace(adam=False, batch_size=5, bucket='', cache_images=False, cfg='./models/yolov5s_se.yaml', data='./data/train_imgs_sliced_320_val.yaml', device='', epochs=60, evolve=False, exist_ok=False, global_rank=-1, hyp='data/hyp.scratch.yaml', image_weights=False, img_size=[320, 320], local_rank=-1, log_imgs=16, multi_scale=False, name='exp', noautoanchor=False, nosave=False, notest=False, project='runs/train', rect=False, resume=False, save_dir='runs\\\\train\\\\exp48', single_cls=False, sync_bn=False, total_batch_size=5, weights=\"''\", workers=8, world_size=1)\nStart Tensorboard with \"tensorboard --logdir runs/train\", view at http://localhost:6006/\nHyperparameters {'lr0': 0.01, 'lrf': 0.2, 'momentum': 0.937, 'weight_decay': 0.0005, 'warmup_epochs': 3.0, 'warmup_momentum': 0.8, 'warmup_bias_lr': 0.1, 'box': 0.05, 'cls': 0.5, 'cls_pw': 1.0, 'obj': 1.0, 'obj_pw': 1.0, 'iou_t': 0.2, 'anchor_t': 4.0, 'fl_gamma': 0.0, 'hsv_h': 0.015, 'hsv_s': 0.7, 'hsv_v': 0.4, 'degrees': 0.0, 'translate': 0.1, 'scale': 0.5, 'shear': 0.0, 'perspective': 0.0, 'flipud': 0.0, 'fliplr': 0.5, 'mosaic': 1.0, 'mixup': 0.0}\n\n from n params module arguments \n 0 -1 1 3520 models.common.Focus [3, 32, 3] \n 1 -1 1 18560 models.common.Conv [32, 64, 3, 2] \n 2 -1 1 19904 models.common.BottleneckCSP [64, 64, 1] \n 3 -1 1 73984 models.common.Conv [64, 128, 3, 2] \n 4 -1 1 161152 models.common.BottleneckCSP [128, 128, 3] \n 5 -1 1 2048 models.common.SELayer [128, 16] \n 6 -1 1 295424 models.common.Conv [128, 256, 3, 2] \n 7 -1 1 641792 models.common.BottleneckCSP [256, 256, 3] \n 8 -1 1 8192 models.common.SELayer [256, 16] \n 9 -1 1 1180672 models.common.Conv [256, 512, 3, 2] \n 10 -1 1 656896 models.common.SPP [512, 512, [5, 9, 13]] \n 11 -1 1 1248768 models.common.BottleneckCSP [512, 512, 1, False] \n 12 -1 1 131584 models.common.Conv [512, 256, 1, 1] \n 13 -1 1 0 torch.nn.modules.upsampling.Upsample [None, 2, 'nearest'] \n 14 [-1, 8] 1 0 models.common.Concat [1] \n 15 -1 1 378624 models.common.BottleneckCSP [512, 256, 1, False] \n 16 -1 1 33024 models.common.Conv [256, 128, 1, 1] \n 17 -1 1 0 torch.nn.modules.upsampling.Upsample [None, 2, 'nearest'] \n 18 [-1, 5] 1 0 models.common.Concat [1] \n 19 -1 1 95104 models.common.BottleneckCSP [256, 128, 1, False] \n 20 -1 1 147712 models.common.Conv [128, 128, 3, 2] \n 21 [-1, 16] 1 0 models.common.Concat [1] \n 22 -1 1 313088 models.common.BottleneckCSP [256, 256, 1, False] \n 23 -1 1 590336 models.common.Conv [256, 256, 3, 2] \n 24 [-1, 12] 1 0 models.common.Concat [1] \n 25 -1 1 1248768 models.common.BottleneckCSP [512, 512, 1, False] \n 26 [19, 22, 25] 1 32364 models.yolo.Detect [7, [[9, 9, 14, 14, 23, 20], [61, 25, 34, 50, 55, 108], [124, 49, 237, 85, 111, 224]], [128, 256, 512]]\nModel Summary: 297 layers, 7281516 parameters, 7281516 gradients\n\nOptimizer groups: 62 .bias, 74 conv.weight, 59 other\nScanning '..\\train_imgs_sliced_all_320_val6\\labels.cache' for images and labels... 28267 found, 0 missing, 0 empty, 0 c\nScanning '..\\train_imgs_sliced_all_320_val\\labels.cache' for images and labels... 14548 found, 0 missing, 0 empty, 0 co\nImage sizes 320 train, 320 test\nUsing 5 dataloader workers\nLogging results to runs\\train\\exp48\nStarting training for 60 epochs...\n\n Epoch gpu_mem box obj cls total targets img_size\n 0%| | 0/5654 [00:00<?, ?it/s]" ], [ "from utils.plots import plot_results \nplot_results(save_dir='./runs/train/exp21') ", "_____no_output_____" ] ] ]

[ "code", "markdown", "code" ]

[ [ "code" ], [ "markdown", "markdown", "markdown", "markdown" ], [ "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# bert4sentiment - an easy implementation with BERT with Hugggingface for sentiment analysis\n\nLet's build a Sentiment Classifier using the amazing Transformers library by Hugging Face!\n\nLoad the ber4sentiment environment. Type from the project folder type \n`conda env create -f configuration.yml` \n\nthis will create a conda _bert4sentiment_ environment. then type \n`conda activate bert4sentiment`\n\nand run the notebook\n\n`jupyter notebook`\n", "_____no_output_____" ] ], [ [ "%reload_ext watermark\n%watermark -v -p numpy,pandas,torch,transformers", "Python implementation: CPython\nPython version : 3.9.6\nIPython version : 7.25.0\n\nnumpy : 1.21.0\npandas : 1.3.0\ntorch : 1.9.0\ntransformers: 4.8.2\n\n" ], [ "#@title Setup & Config\nimport transformers\nfrom transformers import BertModel, BertTokenizer, AdamW, get_linear_schedule_with_warmup\nimport torch\n\nimport numpy as np\nimport pandas as pd\nimport seaborn as sns\nfrom pylab import rcParams\nimport matplotlib.pyplot as plt\nfrom matplotlib import rc\nfrom sklearn.model_selection import train_test_split\nfrom sklearn.metrics import confusion_matrix, classification_report\nfrom collections import defaultdict\nfrom textwrap import wrap\n\nfrom torch import nn, optim\nfrom torch.utils.data import Dataset, DataLoader\nimport torch.nn.functional as F\n\n%matplotlib inline\n%config InlineBackend.figure_format='retina'\n\nsns.set(style='whitegrid', palette='muted', font_scale=1.2)\n\nHAPPY_COLORS_PALETTE = [\"#01BEFE\", \"#FFDD00\", \"#FF7D00\", \"#FF006D\", \"#ADFF02\", \"#8F00FF\"]\n\nsns.set_palette(sns.color_palette(HAPPY_COLORS_PALETTE))\n\nrcParams['figure.figsize'] = 12, 8\n\nRANDOM_SEED = 42\nnp.random.seed(RANDOM_SEED)\ntorch.manual_seed(RANDOM_SEED)\n\ndevice = torch.device(\"cuda:0\" if torch.cuda.is_available() else \"cpu\")\ndevice", "_____no_output_____" ] ], [ [ "## Data Exploration\n\nWe'll load the Google Play app reviews dataset, that we've put together in the previous part:", "_____no_output_____" ] ], [ [ "df = pd.read_csv(\"reviews.csv\")\ndf.head()", "_____no_output_____" ], [ "df.shape", "_____no_output_____" ] ], [ [ "We have about 16k examples. Let's check for missing values:", "_____no_output_____" ] ], [ [ "df.info()", "<class 'pandas.core.frame.DataFrame'>\nRangeIndex: 15746 entries, 0 to 15745\nData columns (total 11 columns):\n # Column Non-Null Count Dtype \n--- ------ -------------- ----- \n 0 userName 15746 non-null object\n 1 userImage 15746 non-null object\n 2 content 15746 non-null object\n 3 score 15746 non-null int64 \n 4 thumbsUpCount 15746 non-null int64 \n 5 reviewCreatedVersion 13533 non-null object\n 6 at 15746 non-null object\n 7 replyContent 7367 non-null object\n 8 repliedAt 7367 non-null object\n 9 sortOrder 15746 non-null object\n 10 appId 15746 non-null object\ndtypes: int64(2), object(9)\nmemory usage: 1.3+ MB\n" ] ], [ [ "Great, no missing values in the score and review texts! Do we have class imbalance?", "_____no_output_____" ] ], [ [ "sns.countplot(x=df.score)\nplt.xlabel('review score');", "_____no_output_____" ] ], [ [ "That's hugely imbalanced, but it's okay. We're going to convert the dataset into negative, neutral and positive sentiment:", "_____no_output_____" ] ], [ [ "def to_sentiment(rating):\n rating = int(rating)\n if rating <= 2:\n return 0\n elif rating == 3:\n return 1\n else: \n return 2\n\ndf['sentiment'] = df.score.apply(to_sentiment)", "_____no_output_____" ], [ "class_names = ['negative', 'neutral', 'positive']", "_____no_output_____" ], [ "ax = sns.countplot(x=df.sentiment)\nplt.xlabel('review sentiment')\nax.set_xticklabels(class_names);", "_____no_output_____" ] ], [ [ "The balance was (mostly) restored.", "_____no_output_____" ], [ "## Data Preprocessing\n\nWe have to prepare the data for the Transformers that means to: \n\n- Add special tokens to separate sentences and do classification\n- Pass sequences of constant length (introduce padding)\n- Create array of 0s (pad token) and 1s (real token) called *attention mask*\n", "_____no_output_____" ] ], [ [ "PRE_TRAINED_MODEL_NAME = 'bert-base-cased'", "_____no_output_____" ] ], [ [ "Let's load a pre-trained [BertTokenizer](https://huggingface.co/transformers/model_doc/bert.html#berttokenizer):", "_____no_output_____" ] ], [ [ "tokenizer = BertTokenizer.from_pretrained(PRE_TRAINED_MODEL_NAME)", "_____no_output_____" ] ], [ [ "We'll use this text to understand the tokenization process:", "_____no_output_____" ] ], [ [ "sample_txt = 'When was I last outside? I am stuck at home for 2 weeks.'", "_____no_output_____" ] ], [ [ "Some basic operations can convert the text to tokens and tokens to unique integers (ids):", "_____no_output_____" ] ], [ [ "tokens = tokenizer.tokenize(sample_txt)\ntoken_ids = tokenizer.convert_tokens_to_ids(tokens)\n\nprint(f' Sentence: {sample_txt}')\nprint(f' Tokens: {tokens}')\nprint(f'Token IDs: {token_ids}')", " Sentence: When was I last outside? I am stuck at home for 2 weeks.\n Tokens: ['When', 'was', 'I', 'last', 'outside', '?', 'I', 'am', 'stuck', 'at', 'home', 'for', '2', 'weeks', '.']\nToken IDs: [1332, 1108, 146, 1314, 1796, 136, 146, 1821, 5342, 1120, 1313, 1111, 123, 2277, 119]\n" ] ], [ [ "### Special Tokens\n\n`[SEP]` - marker for ending of a sentence\n", "_____no_output_____" ] ], [ [ "tokenizer.sep_token, tokenizer.sep_token_id", "_____no_output_____" ] ], [ [ "`[CLS]` - we must add this token to the start of each sentence, so BERT knows we're doing classification", "_____no_output_____" ] ], [ [ "tokenizer.cls_token, tokenizer.cls_token_id", "_____no_output_____" ] ], [ [ "There is also a special token for padding:", "_____no_output_____" ] ], [ [ "tokenizer.pad_token, tokenizer.pad_token_id", "_____no_output_____" ] ], [ [ "BERT understands tokens that were in the training set. Everything else can be encoded using the `[UNK]` (unknown) token:", "_____no_output_____" ] ], [ [ "tokenizer.unk_token, tokenizer.unk_token_id", "_____no_output_____" ] ], [ [ "All of that work can be done using the [`encode_plus()`](https://huggingface.co/transformers/main_classes/tokenizer.html#transformers.PreTrainedTokenizer.encode_plus) method:", "_____no_output_____" ] ], [ [ "encoding = tokenizer.encode_plus(\n sample_txt,\n max_length=32,\n add_special_tokens=True, # Add '[CLS]' and '[SEP]'\n return_token_type_ids=False,\n pad_to_max_length=True,\n return_attention_mask=True,\n return_tensors='pt', # Return PyTorch tensors\n truncation=True,\n)\n\nencoding.keys()", "/home/test/anaconda3/envs/bert4sentiment/lib/python3.9/site-packages/transformers/tokenization_utils_base.py:2126: FutureWarning: The `pad_to_max_length` argument is deprecated and will be removed in a future version, use `padding=True` or `padding='longest'` to pad to the longest sequence in the batch, or use `padding='max_length'` to pad to a max length. In this case, you can give a specific length with `max_length` (e.g. `max_length=45`) or leave max_length to None to pad to the maximal input size of the model (e.g. 512 for Bert).\n warnings.warn(\n" ] ], [ [ "The token ids are now stored in a Tensor and padded to a length of 32:", "_____no_output_____" ] ], [ [ "print(len(encoding['input_ids'][0]))\nencoding['input_ids'][0]", "32\n" ] ], [ [ "The attention mask has the same length:", "_____no_output_____" ] ], [ [ "print(len(encoding['attention_mask'][0]))\nencoding['attention_mask']", "32\n" ] ], [ [ "We can inverse the tokenization to have a look at the special tokens:", "_____no_output_____" ] ], [ [ "tokenizer.convert_ids_to_tokens(encoding['input_ids'][0])", "_____no_output_____" ] ], [ [ "### Choosing Sequence Length\n\nBERT works with fixed-length sequences. We'll use a simple strategy to choose the max length. Let's store the token length of each review:", "_____no_output_____" ] ], [ [ "token_lens = []\n\nfor txt in df.content:\n tokens = tokenizer.encode(txt, max_length=512)\n token_lens.append(len(tokens))", "Truncation was not explicitly activated but `max_length` is provided a specific value, please use `truncation=True` to explicitly truncate examples to max length. Defaulting to 'longest_first' truncation strategy. If you encode pairs of sequences (GLUE-style) with the tokenizer you can select this strategy more precisely by providing a specific strategy to `truncation`.\n" ] ], [ [ "and plot the distribution:", "_____no_output_____" ] ], [ [ "sns.histplot(x=token_lens)\nplt.xlim([0, 256]);\nplt.xlabel('Token count');", "_____no_output_____" ] ], [ [ "Most of the reviews seem to contain less than 128 tokens, but we'll be on the safe side and choose a maximum length of 160.", "_____no_output_____" ] ], [ [ "MAX_LEN = 160", "_____no_output_____" ] ], [ [ "We have all building blocks required to create a PyTorch dataset. Let's do it:", "_____no_output_____" ] ], [ [ "class GPReviewDataset(Dataset):\n\n def __init__(self, reviews, targets, tokenizer, max_len):\n self.reviews = reviews\n self.targets = targets\n self.tokenizer = tokenizer\n self.max_len = max_len\n \n def __len__(self):\n return len(self.reviews)\n \n def __getitem__(self, item):\n review = str(self.reviews[item])\n target = self.targets[item]\n\n encoding = self.tokenizer.encode_plus(\n review,\n add_special_tokens=True,\n max_length=self.max_len,\n return_token_type_ids=False,\n pad_to_max_length=True,\n return_attention_mask=True,\n return_tensors='pt',\n )\n\n return {\n 'review_text': review,\n 'input_ids': encoding['input_ids'].flatten(),\n 'attention_mask': encoding['attention_mask'].flatten(),\n 'targets': torch.tensor(target, dtype=torch.long)\n }", "_____no_output_____" ] ], [ [ "The tokenizer is doing most of the heavy lifting for us. We also return the review texts, so it'll be easier to evaluate the predictions from our model. Let's split the data:", "_____no_output_____" ] ], [ [ "df_train, df_test = train_test_split(df, test_size=0.1, random_state=RANDOM_SEED)\ndf_val, df_test = train_test_split(df_test, test_size=0.5, random_state=RANDOM_SEED)", "_____no_output_____" ], [ "df_train.shape, df_val.shape, df_test.shape", "_____no_output_____" ] ], [ [ "We also need to create a couple of data loaders. Here's a helper function to do it:", "_____no_output_____" ] ], [ [ "def create_data_loader(df, tokenizer, max_len, batch_size):\n ds = GPReviewDataset(\n reviews=df.content.to_numpy(),\n targets=df.sentiment.to_numpy(),\n tokenizer=tokenizer,\n max_len=max_len\n )\n\n return DataLoader(\n ds,\n batch_size=batch_size,\n num_workers=4\n )", "_____no_output_____" ], [ "BATCH_SIZE = 16\n\ntrain_data_loader = create_data_loader(df_train, tokenizer, MAX_LEN, BATCH_SIZE)\nval_data_loader = create_data_loader(df_val, tokenizer, MAX_LEN, BATCH_SIZE)\ntest_data_loader = create_data_loader(df_test, tokenizer, MAX_LEN, BATCH_SIZE)", "_____no_output_____" ] ], [ [ "Let's have a look at an example batch from our training data loader:", "_____no_output_____" ] ], [ [ "data = next(iter(train_data_loader))\ndata.keys()", "/home/test/anaconda3/envs/bert4sentiment/lib/python3.9/site-packages/transformers/tokenization_utils_base.py:2126: FutureWarning: The `pad_to_max_length` argument is deprecated and will be removed in a future version, use `padding=True` or `padding='longest'` to pad to the longest sequence in the batch, or use `padding='max_length'` to pad to a max length. In this case, you can give a specific length with `max_length` (e.g. `max_length=45`) or leave max_length to None to pad to the maximal input size of the model (e.g. 512 for Bert).\n warnings.warn(\n/home/test/anaconda3/envs/bert4sentiment/lib/python3.9/site-packages/transformers/tokenization_utils_base.py:2126: FutureWarning: The `pad_to_max_length` argument is deprecated and will be removed in a future version, use `padding=True` or `padding='longest'` to pad to the longest sequence in the batch, or use `padding='max_length'` to pad to a max length. In this case, you can give a specific length with `max_length` (e.g. `max_length=45`) or leave max_length to None to pad to the maximal input size of the model (e.g. 512 for Bert).\n warnings.warn(\n/home/test/anaconda3/envs/bert4sentiment/lib/python3.9/site-packages/transformers/tokenization_utils_base.py:2126: FutureWarning: The `pad_to_max_length` argument is deprecated and will be removed in a future version, use `padding=True` or `padding='longest'` to pad to the longest sequence in the batch, or use `padding='max_length'` to pad to a max length. In this case, you can give a specific length with `max_length` (e.g. `max_length=45`) or leave max_length to None to pad to the maximal input size of the model (e.g. 512 for Bert).\n warnings.warn(\n/home/test/anaconda3/envs/bert4sentiment/lib/python3.9/site-packages/transformers/tokenization_utils_base.py:2126: FutureWarning: The `pad_to_max_length` argument is deprecated and will be removed in a future version, use `padding=True` or `padding='longest'` to pad to the longest sequence in the batch, or use `padding='max_length'` to pad to a max length. In this case, you can give a specific length with `max_length` (e.g. `max_length=45`) or leave max_length to None to pad to the maximal input size of the model (e.g. 512 for Bert).\n warnings.warn(\n" ], [ "print(data['input_ids'].shape)\nprint(data['attention_mask'].shape)\nprint(data['targets'].shape)", "torch.Size([16, 160])\ntorch.Size([16, 160])\ntorch.Size([16])\n" ] ], [ [ "## Sentiment Classification with BERT and Hugging Face", "_____no_output_____" ], [ "We'll use the basic [BertModel](https://huggingface.co/transformers/model_doc/bert.html#bertmodel) and build our sentiment classifier on top of it. Let's load the model:", "_____no_output_____" ] ], [ [ "bert_model = BertModel.from_pretrained(PRE_TRAINED_MODEL_NAME, return_dict = False)\n", "Some weights of the model checkpoint at bert-base-cased were not used when initializing BertModel: ['cls.predictions.decoder.weight', 'cls.seq_relationship.weight', 'cls.predictions.bias', 'cls.seq_relationship.bias', 'cls.predictions.transform.dense.bias', 'cls.predictions.transform.LayerNorm.bias', 'cls.predictions.transform.dense.weight', 'cls.predictions.transform.LayerNorm.weight']\n- This IS expected if you are initializing BertModel from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).\n- This IS NOT expected if you are initializing BertModel from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).\n" ] ], [ [ "And try to use it on the encoding of our sample text:", "_____no_output_____" ] ], [ [ "last_hidden_state, pooled_output = bert_model(\n input_ids=encoding['input_ids'], \n attention_mask=encoding['attention_mask'], return_dict=False\n)", "_____no_output_____" ] ], [ [ "The `last_hidden_state` is a sequence of hidden states of the last layer of the model. Obtaining the `pooled_output` is done by applying the [BertPooler](https://github.com/huggingface/transformers/blob/edf0582c0be87b60f94f41c659ea779876efc7be/src/transformers/modeling_bert.py#L426) on `last_hidden_state`:", "_____no_output_____" ] ], [ [ "last_hidden_state.shape", "_____no_output_____" ] ], [ [ "We have the hidden state for each of our 32 tokens (the length of our example sequence). But why 768? This is the number of hidden units in the feedforward-networks. We can verify that by checking the config:", "_____no_output_____" ] ], [ [ "bert_model.config.hidden_size", "_____no_output_____" ] ], [ [ "\n\nYou can think of the `pooled_output` as a summary of the content, according to BERT. Albeit, you might try and do better. Let's look at the shape of the output:", "_____no_output_____" ] ], [ [ "pooled_output.shape", "_____no_output_____" ] ], [ [ "We can use all of this knowledge to create a classifier that uses the BERT model:", "_____no_output_____" ] ], [ [ "class SentimentClassifier(nn.Module):\n def __init__(self, n_classes):\n super(SentimentClassifier, self).__init__()\n self.bert = BertModel.from_pretrained(PRE_TRAINED_MODEL_NAME)\n self.drop = nn.Dropout(p=0.3)\n self.out = nn.Linear(self.bert.config.hidden_size, n_classes)\n\n def forward(self, input_ids, attention_mask):\n bertOutput = self.bert(\n input_ids=input_ids,\n attention_mask=attention_mask\n )\n output = self.drop(bertOutput['pooler_output'])\n\n return self.out(output)", "_____no_output_____" ] ], [ [ "Our classifier delegates most of the heavy lifting to the BertModel. We use a dropout layer for some regularization and a fully-connected layer for our output. Note that we're returning the raw output of the last layer since that is required for the cross-entropy loss function in PyTorch to work.\n\nThis should work like any other PyTorch model. Let's create an instance and move it to the GPU:", "_____no_output_____" ] ], [ [ "model = SentimentClassifier(len(class_names))\nmodel = model.to(device)", "Some weights of the model checkpoint at bert-base-cased were not used when initializing BertModel: ['cls.predictions.decoder.weight', 'cls.seq_relationship.weight', 'cls.predictions.bias', 'cls.seq_relationship.bias', 'cls.predictions.transform.dense.bias', 'cls.predictions.transform.LayerNorm.bias', 'cls.predictions.transform.dense.weight', 'cls.predictions.transform.LayerNorm.weight']\n- This IS expected if you are initializing BertModel from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).\n- This IS NOT expected if you are initializing BertModel from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).\n" ] ], [ [ "We'll move the example batch of our training data to the GPU:", "_____no_output_____" ] ], [ [ "input_ids = data['input_ids'].to(device)\nattention_mask = data['attention_mask'].to(device)\n\nprint(input_ids.shape) # batch size x seq length\nprint(attention_mask.shape) # batch size x seq length", "torch.Size([16, 160])\ntorch.Size([16, 160])\n" ] ], [ [ "To get the predicted probabilities from our trained model, we'll apply the softmax function to the outputs:", "_____no_output_____" ] ], [ [ "F.softmax(model(input_ids, attention_mask), dim=1)", "_____no_output_____" ] ], [ [ "### Training", "_____no_output_____" ], [ " we'll use the [AdamW](https://huggingface.co/transformers/main_classes/optimizer_schedules.html#adamw) optimizer provided by Hugging Face that corrects weight decay.", "_____no_output_____" ] ], [ [ "EPOCHS = 100\n\noptimizer = AdamW(model.parameters(), lr=2e-5, correct_bias=False)\ntotal_steps = len(train_data_loader) * EPOCHS\n\nscheduler = get_linear_schedule_with_warmup(\n optimizer,\n num_warmup_steps=0,\n num_training_steps=total_steps\n)\n\nloss_fn = nn.CrossEntropyLoss().to(device)", "_____no_output_____" ] ], [ [ "How do we come up with all hyperparameters? The BERT authors have some recommendations for fine-tuning:\n\n- Batch size: 16, 32\n- Learning rate (Adam): 5e-5, 3e-5, 2e-5\n- Number of epochs: 2, 3, 4\n\nWe're going to ignore the number of epochs recommendation but stick with the rest. Note that increasing the batch size reduces the training time significantly, but gives you lower accuracy.\n\nLet's continue with writing a helper function for training our model for one epoch:", "_____no_output_____" ] ], [ [ "def train_epoch(\n model, \n data_loader, \n loss_fn, \n optimizer, \n device, \n scheduler, \n n_examples\n):\n model = model.train()\n\n losses = []\n correct_predictions = 0\n \n for d in data_loader:\n input_ids = d[\"input_ids\"].to(device)\n attention_mask = d[\"attention_mask\"].to(device)\n targets = d[\"targets\"].to(device)\n\n outputs = model(\n input_ids=input_ids,\n attention_mask=attention_mask\n )\n\n _, preds = torch.max(outputs, dim=1)\n loss = loss_fn(outputs, targets)\n\n correct_predictions += torch.sum(preds == targets)\n losses.append(loss.item())\n\n loss.backward()\n nn.utils.clip_grad_norm_(model.parameters(), max_norm=1.0)\n optimizer.step()\n scheduler.step()\n optimizer.zero_grad()\n\n return correct_predictions.double() / n_examples, np.mean(losses)", "_____no_output_____" ] ], [ [ "Training the model should look familiar, except for two things. The scheduler gets called every time a batch is fed to the model. We're avoiding exploding gradients by clipping the gradients of the model using [clip_grad_norm_](https://pytorch.org/docs/stable/nn.html#clip-grad-norm).\n\nLet's write another one that helps us evaluate the model on a given data loader:", "_____no_output_____" ] ], [ [ "def eval_model(model, data_loader, loss_fn, device, n_examples):\n model = model.eval()\n\n losses = []\n correct_predictions = 0\n\n with torch.no_grad():\n for d in data_loader:\n input_ids = d[\"input_ids\"].to(device)\n attention_mask = d[\"attention_mask\"].to(device)\n targets = d[\"targets\"].to(device)\n\n outputs = model(\n input_ids=input_ids,\n attention_mask=attention_mask\n )\n _, preds = torch.max(outputs, dim=1)\n\n loss = loss_fn(outputs, targets)\n\n correct_predictions += torch.sum(preds == targets)\n losses.append(loss.item())\n\n return correct_predictions.double() / n_examples, np.mean(losses)", "_____no_output_____" ] ], [ [ "Using those two, we can write our training loop. We'll also store the training history:", "_____no_output_____" ] ], [ [ "%%time\n\nhistory = defaultdict(list)\nbest_accuracy = 0\n\nfor epoch in range(EPOCHS):\n\n print(f'Epoch {epoch + 1}/{EPOCHS}')\n print('-' * 10)\n\n train_acc, train_loss = train_epoch(\n model,\n train_data_loader, \n loss_fn, \n optimizer, \n device, \n scheduler, \n len(df_train)\n )\n\n print(f'Train loss {train_loss} accuracy {train_acc}')\n\n val_acc, val_loss = eval_model(\n model,\n val_data_loader,\n loss_fn, \n device, \n len(df_val)\n )\n\n print(f'Val loss {val_loss} accuracy {val_acc}')\n print()\n\n history['train_acc'].append(train_acc)\n history['train_loss'].append(train_loss)\n history['val_acc'].append(val_acc)\n history['val_loss'].append(val_loss)\n\n if val_acc > best_accuracy:\n torch.save(model.state_dict(), 'best_model_state.bin')\n best_accuracy = val_acc", "Epoch 1/100\n----------\n" ] ], [ [ "Note that we're storing the state of the best model, indicated by the highest validation accuracy.", "_____no_output_____" ], [ "Whoo, this took some time! We can look at the training vs validation accuracy:", "_____no_output_____" ] ], [ [ "plt.plot(history['train_acc'], label='train accuracy')\nplt.plot(history['val_acc'], label='validation accuracy')\n\nplt.title('Training history')\nplt.ylabel('Accuracy')\nplt.xlabel('Epoch')\nplt.legend()\nplt.ylim([0, 1]);", "_____no_output_____" ] ], [ [ "The training accuracy starts to approach 100% after 10 epochs or so. You might try to fine-tune the parameters a bit more, but this will be good enough for us.\n\n", "_____no_output_____" ] ], [ [ "# !gdown --id 1V8itWtowCYnb2Bc9KlK9SxGff9WwmogA\n\n#model = SentimentClassifier(len(class_names))\n#model.load_state_dict(torch.load('best_model_state.bin'))\n#model = model.to(device)", "_____no_output_____" ] ], [ [ "## Evaluation\n\nSo how good is our model on predicting sentiment? Let's start by calculating the accuracy on the test data:", "_____no_output_____" ] ], [ [ "test_acc, _ = eval_model(\n model,\n test_data_loader,\n loss_fn,\n device,\n len(df_test)\n)\n\ntest_acc.item()", "_____no_output_____" ] ], [ [ "The accuracy is about 1% lower on the test set. Our model seems to generalize well.\n\nWe'll define a helper function to get the predictions from our model:", "_____no_output_____" ] ], [ [ "def get_predictions(model, data_loader):\n model = model.eval()\n \n review_texts = []\n predictions = []\n prediction_probs = []\n real_values = []\n\n with torch.no_grad():\n for d in data_loader:\n\n texts = d[\"review_text\"]\n input_ids = d[\"input_ids\"].to(device)\n attention_mask = d[\"attention_mask\"].to(device)\n targets = d[\"targets\"].to(device)\n\n outputs = model(\n input_ids=input_ids,\n attention_mask=attention_mask\n )\n _, preds = torch.max(outputs, dim=1)\n\n probs = F.softmax(outputs, dim=1)\n\n review_texts.extend(texts)\n predictions.extend(preds)\n prediction_probs.extend(probs)\n real_values.extend(targets)\n\n predictions = torch.stack(predictions).cpu()\n prediction_probs = torch.stack(prediction_probs).cpu()\n real_values = torch.stack(real_values).cpu()\n return review_texts, predictions, prediction_probs, real_values", "_____no_output_____" ] ], [ [ "This is similar to the evaluation function, except that we're storing the text of the reviews and the predicted probabilities (by applying the softmax on the model outputs):", "_____no_output_____" ] ], [ [ "y_review_texts, y_pred, y_pred_probs, y_test = get_predictions(\n model,\n test_data_loader\n)", "_____no_output_____" ] ], [ [ "Let's have a look at the classification report", "_____no_output_____" ] ], [ [ "print(classification_report(y_test, y_pred, target_names=class_names))", "_____no_output_____" ] ], [ [ "Looks like it is really hard to classify neutral (3 stars) reviews. And I can tell you from experience, looking at many reviews, those are hard to classify.\n\nWe'll continue with the confusion matrix:", "_____no_output_____" ] ], [ [ "def show_confusion_matrix(confusion_matrix):\n hmap = sns.heatmap(confusion_matrix, annot=True, fmt=\"d\", cmap=\"Blues\")\n hmap.yaxis.set_ticklabels(hmap.yaxis.get_ticklabels(), rotation=0, ha='right')\n hmap.xaxis.set_ticklabels(hmap.xaxis.get_ticklabels(), rotation=30, ha='right')\n plt.ylabel('True sentiment')\n plt.xlabel('Predicted sentiment');\n\ncm = confusion_matrix(y_test, y_pred)\ndf_cm = pd.DataFrame(cm, index=class_names, columns=class_names)\nshow_confusion_matrix(df_cm)", "_____no_output_____" ] ], [ [ "This confirms that our model is having difficulty classifying neutral reviews. It mistakes those for negative and positive at a roughly equal frequency.\n\nThat's a good overview of the performance of our model. But let's have a look at an example from our test data:", "_____no_output_____" ] ], [ [ "idx = 2\n\nreview_text = y_review_texts[idx]\ntrue_sentiment = y_test[idx]\n", "_____no_output_____" ], [ "print(\"\\n\".join(wrap(review_text)))\nprint()\nprint(f'True sentiment: {class_names[true_sentiment]}')", "_____no_output_____" ] ], [ [ "Now we can look at the confidence of each sentiment of our model:", "_____no_output_____" ] ], [ [ "\npred_df = pd.DataFrame({\n 'class_names': class_names,\n 'values': y_pred_probs[idx].tolist() #converting tensor to numbers\n})\n", "_____no_output_____" ], [ "sns.barplot(x='values', y='class_names', data=pred_df, orient='h')\nplt.ylabel('sentiment')\nplt.xlabel('probability')\nplt.xlim([0, 1]);", "_____no_output_____" ] ], [ [ "### Predicting on Raw Text\n\nLet's use our model to predict the sentiment of some raw text:", "_____no_output_____" ] ], [ [ "review_text = \"I love completing my todos! Best app ever!!!\"", "_____no_output_____" ] ], [ [ "We have to use the tokenizer to encode the text:", "_____no_output_____" ] ], [ [ "encoded_review = tokenizer.encode_plus(\n review_text,\n max_length=MAX_LEN,\n add_special_tokens=True,\n return_token_type_ids=False,\n pad_to_max_length=True,\n return_attention_mask=True,\n return_tensors='pt',\n)", "_____no_output_____" ] ], [ [ "Let's get the predictions from our model:", "_____no_output_____" ] ], [ [ "input_ids = encoded_review['input_ids'].to(device)\nattention_mask = encoded_review['attention_mask'].to(device)\n\noutput = model(input_ids, attention_mask)\n_, prediction = torch.max(output, dim=1)\n\nprint(f'Review text: {review_text}')\nprint(f'Sentiment : {class_names[prediction]}')", "_____no_output_____" ] ], [ [ "\nNice job! You learned how to use BERT for sentiment analysis. ", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown" ]

[ [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# This Python 3 environment comes with many helpful analytics libraries installed\n# It is defined by the kaggle/python Docker image: https://github.com/kaggle/docker-python\n# For example, here's several helpful packages to load\n\nimport numpy as np # linear algebra\nimport pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)\n\n# Input data files are available in the read-only \"../input/\" directory\n# For example, running this (by clicking run or pressing Shift+Enter) will list all files under the input directory\n\nimport os\nfor dirname, _, filenames in os.walk('/kaggle/input'):\n for filename in filenames:\n print(os.path.join(dirname, filename))\n\n# You can write up to 20GB to the current directory (/kaggle/working/) that gets preserved as output when you create a version using \"Save & Run All\" \n# You can also write temporary files to /kaggle/temp/, but they won't be saved outside of the current session", "/kaggle/input/asap-aes/valid_sample_submission_1_column.csv\n/kaggle/input/asap-aes/Training_Materials.zip\n/kaggle/input/asap-aes/training_set_rel3.xls\n/kaggle/input/asap-aes/valid_sample_submission_1_column_no_header.csv\n/kaggle/input/asap-aes/Essay_Set_Descriptions.zip\n/kaggle/input/asap-aes/training_set_rel3.xlsx\n/kaggle/input/asap-aes/valid_set.xls\n/kaggle/input/asap-aes/training_set_rel3.tsv\n/kaggle/input/asap-aes/valid_sample_submission_5_column.csv\n/kaggle/input/asap-aes/valid_set.xlsx\n/kaggle/input/asap-aes/valid_set.tsv\n/kaggle/input/asap-aes/test_set.tsv\n/kaggle/input/asap-aes/valid_sample_submission_2_column.csv\n" ], [ "import os\nimport pandas as pd\n\ndf = pd.read_csv(\"../input/asap-aes/training_set_rel3.tsv\", sep='\\t', encoding='ISO-8859-1')\ndf = df.dropna(axis=1)\ndf = df.drop(columns=['rater1_domain1', 'rater2_domain1'])\ndf = df.drop(columns=['essay_id', 'essay_set'])\ndf.head()", "_____no_output_____" ], [ "df['essay'][1]", "_____no_output_____" ], [ "import warnings\nwarnings.filterwarnings(\"ignore\") #Ignoring unnecessory warnings\n\nimport numpy as np #for large and multi-dimensional arrays\nimport pandas as pd #for data manipulation and analysis\nimport nltk #Natural language processing tool-kit\n\nfrom nltk.corpus import stopwords #Stopwords corpus\nfrom nltk.stem import PorterStemmer # Stemmer\n\nfrom sklearn.feature_extraction.text import CountVectorizer #For Bag of words\nfrom sklearn.feature_extraction.text import TfidfVectorizer #For TF-IDF\nfrom gensim.models import Word2Vec #For Word2Vec\n\nfrom tensorflow.keras.layers import Embedding\nfrom tensorflow.keras.preprocessing.sequence import pad_sequences\nfrom tensorflow.keras.models import Sequential\nfrom tensorflow.keras.preprocessing.text import one_hot\nfrom tensorflow.keras.layers import LSTM\nfrom tensorflow.keras.layers import Dropout\nfrom tensorflow.keras.layers import Dense", "_____no_output_____" ], [ "import numpy as np\nimport nltk\nimport re\nfrom nltk.corpus import stopwords\nfrom gensim.models import Word2Vec\n\ndef essay_to_wordlist(essay_v, remove_stopwords):\n \"\"\"Remove the tagged labels and word tokenize the sentence.\"\"\"\n essay_v = re.sub(\"[^a-zA-Z]\", \" \", essay_v)\n words = essay_v.lower().split()\n if remove_stopwords:\n stops = set(stopwords.words(\"english\"))\n words = [w for w in words if not w in stops]\n return (words)\n\ndef essay_to_sentences(essay_v, remove_stopwords):\n \"\"\"Sentence tokenize the essay and call essay_to_wordlist() for word tokenization.\"\"\"\n tokenizer = nltk.data.load('tokenizers/punkt/english.pickle')\n raw_sentences = tokenizer.tokenize(essay_v.strip())\n sentences = []\n for raw_sentence in raw_sentences:\n if len(raw_sentence) > 0:\n sentences.append(essay_to_wordlist(raw_sentence, remove_stopwords))\n return sentences\n\ndef makeFeatureVec(words, model, num_features):\n \"\"\"Make Feature Vector from the words list of an Essay.\"\"\"\n featureVec = np.zeros((num_features,),dtype=\"float32\")\n num_words = 0.\n index2word_set = set(model.wv.index2word)\n for word in words:\n if word in index2word_set:\n num_words += 1\n featureVec = np.add(featureVec,model[word]) \n featureVec = np.divide(featureVec,num_words)\n return featureVec\n\ndef getAvgFeatureVecs(essays, model, num_features):\n \"\"\"Main function to generate the word vectors for word2vec model.\"\"\"\n counter = 0\n essayFeatureVecs = np.zeros((len(essays),num_features),dtype=\"float32\")\n for essay in essays:\n essayFeatureVecs[counter] = makeFeatureVec(essay, model, num_features)\n counter = counter + 1\n return essayFeatureVecs", "_____no_output_____" ], [ "import nltk\n\nnltk.download('stopwords')", "[nltk_data] Downloading package stopwords to /usr/share/nltk_data...\n[nltk_data] Package stopwords is already up-to-date!\n" ], [ "from keras.layers import Embedding, LSTM, Dense, Dropout, Lambda, Flatten\nfrom keras.models import Sequential, load_model, model_from_config\nimport keras.backend as K\n\ndef get_model():\n \"\"\"Define the model.\"\"\"\n model = Sequential()\n model.add(LSTM(300, dropout=0.4, recurrent_dropout=0.4, input_shape=[1, 300], return_sequences=True))\n model.add(LSTM(64, recurrent_dropout=0.4))\n model.add(Dropout(0.5))\n model.add(Dense(1, activation='relu'))\n\n model.compile(loss='mean_squared_error', optimizer='rmsprop', metrics=['accuracy','mae'])\n model.summary()\n\n return model", "_____no_output_____" ], [ "X=df\ny = X['domain1_score']", "_____no_output_____" ], [ "from sklearn.model_selection import KFold\nfrom sklearn.linear_model import LinearRegression\nfrom sklearn.metrics import cohen_kappa_score\n\ncv = KFold(n_splits = 5, shuffle = True)\nresults = []\ny_pred_list = []\n\ncount = 1\nfor traincv, testcv in cv.split(X):\n print(\"\\n--------Fold {}--------\\n\".format(count))\n X_test, X_train, y_test, y_train = X.iloc[testcv], X.iloc[traincv], y.iloc[testcv], y.iloc[traincv]\n \n train_essays = X_train['essay']\n test_essays = X_test['essay']\n \n sentences = []\n \n for essay in train_essays:\n # Obtaining all sentences from the training essays.\n sentences += essay_to_sentences(essay, remove_stopwords = True)\n \n # Initializing variables for word2vec model.\n num_features = 300 \n min_word_count = 40\n num_workers = 4\n context = 10\n downsampling = 1e-3\n\n print(\"Training Word2Vec Model...\")\n model = Word2Vec(sentences, workers=num_workers, size=num_features, min_count = min_word_count, window = context, sample = downsampling)\n\n model.init_sims(replace=True)\n model.wv.save_word2vec_format('word2vecmodel.bin', binary=True)\n\n clean_train_essays = []\n \n # Generate training and testing data word vectors.\n for essay_v in train_essays:\n clean_train_essays.append(essay_to_wordlist(essay_v, remove_stopwords=True))\n trainDataVecs = getAvgFeatureVecs(clean_train_essays, model, num_features)\n \n clean_test_essays = []\n for essay_v in test_essays:\n clean_test_essays.append(essay_to_wordlist( essay_v, remove_stopwords=True ))\n testDataVecs = getAvgFeatureVecs( clean_test_essays, model, num_features )\n \n trainDataVecs = np.array(trainDataVecs)\n testDataVecs = np.array(testDataVecs)\n # Reshaping train and test vectors to 3 dimensions. (1 represnts one timestep)\n trainDataVecs = np.reshape(trainDataVecs, (trainDataVecs.shape[0], 1, trainDataVecs.shape[1]))\n testDataVecs = np.reshape(testDataVecs, (testDataVecs.shape[0], 1, testDataVecs.shape[1]))\n \n lstm_model = get_model()\n lstm_model.fit(trainDataVecs, y_train, batch_size=64, epochs=2)\n #lstm_model.load_weights('./model_weights/final_lstm.h5')\n y_pred = lstm_model.predict(testDataVecs)\n \n # Save any one of the 5 models.\n if count == 5:\n lstm_model.save('./model_weights/final_lstm.h5')\n \n # Round y_pred to the nearest integer.\n y_pred = np.around(y_pred)\n \n # Evaluate the model on the evaluation metric. \"Quadratic mean averaged Kappa\"\n result = cohen_kappa_score(y_test.values,y_pred,weights='quadratic')\n print(\"Kappa Score: {}\".format(result))\n results.append(result)\n\n count += 1\n", "\n--------Fold 1--------\n\nTraining Word2Vec Model...\nModel: \"sequential\"\n_________________________________________________________________\nLayer (type) Output Shape Param # \n=================================================================\nlstm (LSTM) (None, 1, 300) 721200 \n_________________________________________________________________\nlstm_1 (LSTM) (None, 64) 93440 \n_________________________________________________________________\ndropout (Dropout) (None, 64) 0 \n_________________________________________________________________\ndense (Dense) (None, 1) 65 \n=================================================================\nTotal params: 814,705\nTrainable params: 814,705\nNon-trainable params: 0\n_________________________________________________________________\nEpoch 1/2\n163/163 [==============================] - 2s 14ms/step - loss: 64.1702 - accuracy: 0.1120 - mae: 4.3562\nEpoch 2/2\n163/163 [==============================] - 2s 13ms/step - loss: 40.5542 - accuracy: 0.0803 - mae: 3.5957\nKappa Score: 0.7127520678030524\n\n--------Fold 2--------\n\nTraining Word2Vec Model...\nModel: \"sequential_1\"\n_________________________________________________________________\nLayer (type) Output Shape Param # \n=================================================================\nlstm_2 (LSTM) (None, 1, 300) 721200 \n_________________________________________________________________\nlstm_3 (LSTM) (None, 64) 93440 \n_________________________________________________________________\ndropout_1 (Dropout) (None, 64) 0 \n_________________________________________________________________\ndense_1 (Dense) (None, 1) 65 \n=================================================================\nTotal params: 814,705\nTrainable params: 814,705\nNon-trainable params: 0\n_________________________________________________________________\nEpoch 1/2\n163/163 [==============================] - 2s 12ms/step - loss: 63.9632 - accuracy: 0.1127 - mae: 4.3208\nEpoch 2/2\n163/163 [==============================] - 2s 12ms/step - loss: 39.9875 - accuracy: 0.0761 - mae: 3.6013\nKappa Score: 0.7534653510180308\n\n--------Fold 3--------\n\nTraining Word2Vec Model...\nModel: \"sequential_2\"\n_________________________________________________________________\nLayer (type) Output Shape Param # \n=================================================================\nlstm_4 (LSTM) (None, 1, 300) 721200 \n_________________________________________________________________\nlstm_5 (LSTM) (None, 64) 93440 \n_________________________________________________________________\ndropout_2 (Dropout) (None, 64) 0 \n_________________________________________________________________\ndense_2 (Dense) (None, 1) 65 \n=================================================================\nTotal params: 814,705\nTrainable params: 814,705\nNon-trainable params: 0\n_________________________________________________________________\nEpoch 1/2\n163/163 [==============================] - 2s 12ms/step - loss: 62.7267 - accuracy: 0.1150 - mae: 4.3239\nEpoch 2/2\n163/163 [==============================] - 2s 12ms/step - loss: 38.9225 - accuracy: 0.0753 - mae: 3.5816\nKappa Score: 0.760911157888906\n\n--------Fold 4--------\n\nTraining Word2Vec Model...\nModel: \"sequential_3\"\n_________________________________________________________________\nLayer (type) Output Shape Param # \n=================================================================\nlstm_6 (LSTM) (None, 1, 300) 721200 \n_________________________________________________________________\nlstm_7 (LSTM) (None, 64) 93440 \n_________________________________________________________________\ndropout_3 (Dropout) (None, 64) 0 \n_________________________________________________________________\ndense_3 (Dense) (None, 1) 65 \n=================================================================\nTotal params: 814,705\nTrainable params: 814,705\nNon-trainable params: 0\n_________________________________________________________________\nEpoch 1/2\n163/163 [==============================] - 2s 15ms/step - loss: 63.2130 - accuracy: 0.1110 - mae: 4.3320\nEpoch 2/2\n163/163 [==============================] - 2s 12ms/step - loss: 40.5429 - accuracy: 0.0809 - mae: 3.6119\nKappa Score: 0.7429735456201619\n\n--------Fold 5--------\n\nTraining Word2Vec Model...\nModel: \"sequential_4\"\n_________________________________________________________________\nLayer (type) Output Shape Param # \n=================================================================\nlstm_8 (LSTM) (None, 1, 300) 721200 \n_________________________________________________________________\nlstm_9 (LSTM) (None, 64) 93440 \n_________________________________________________________________\ndropout_4 (Dropout) (None, 64) 0 \n_________________________________________________________________\ndense_4 (Dense) (None, 1) 65 \n=================================================================\nTotal params: 814,705\nTrainable params: 814,705\nNon-trainable params: 0\n_________________________________________________________________\nEpoch 1/2\n163/163 [==============================] - 2s 12ms/step - loss: 63.6962 - accuracy: 0.1113 - mae: 4.3485\nEpoch 2/2\n163/163 [==============================] - 2s 12ms/step - loss: 40.5028 - accuracy: 0.0794 - mae: 3.6451\n" ], [ "df.info()", "<class 'pandas.core.frame.DataFrame'>\nRangeIndex: 12976 entries, 0 to 12975\nData columns (total 2 columns):\n # Column Non-Null Count Dtype \n--- ------ -------------- ----- \n 0 essay 12976 non-null object\n 1 domain1_score 12976 non-null int64 \ndtypes: int64(1), object(1)\nmemory usage: 202.9+ KB\n" ], [ "print(\"Average Kappa score after a 5-fold cross validation: \",np.around(np.array(results).mean(),decimals=4))", "Average Kappa score after a 5-fold cross validation: 0.9098\n" ], [ "demo = [\"Dear@CAPS1 @CAPS2, I believe that using computers will benefit us in many ways like talking and becoming friends will others through websites like facebook and mysace. Using computers can help us find coordibates, locations, and able ourselfs to millions of information. Also computers will benefit us by helping with jobs as in planning a house plan and typing a @NUM1 page report for one of our jobs in less than writing it. Now lets go into the wonder world of technology. Using a computer will help us in life by talking or making friends on line. Many people have myspace, facebooks, aim, these all benefit us by having conversations with one another. Many people believe computers are bad but how can you make friends if you can never talk to them? I am very fortunate for having a computer that can help with not only school work but my social life and how I make friends. Computers help us with finding our locations, coordibates and millions of information online. If we didn't go on the internet a lot we wouldn't know how to go onto websites that @MONTH1 help us with locations and coordinates like @LOCATION1. Would you rather use a computer or be in @LOCATION3. When your supposed to be vacationing in @LOCATION2. Million of information is found on the internet. You can as almost every question and a computer will have it. Would you rather easily draw up a house plan on the computers or take @NUM1 hours doing one by hand with ugly erazer marks all over it, you are garrenteed that to find a job with a drawing like that. Also when appling for a job many workers must write very long papers like a @NUM3 word essay on why this job fits you the most, and many people I know don't like writing @NUM3 words non-stopp for hours when it could take them I hav an a computer. That is why computers we needed a lot now adays. I hope this essay has impacted your descion on computers because they are great machines to work with. The other day I showed my mom how to use a computer and she said it was the greatest invention sense sliced bread! Now go out and buy a computer to help you chat online with friends, find locations and millions of information on one click of the button and help your self with getting a job with neat, prepared, printed work that your boss will love.\"]\ndemo_df = pd.DataFrame(demo,columns=['essay'])\ndemo_df.head()", "_____no_output_____" ], [ "type(demo_df['essay'])", "_____no_output_____" ], [ "content = \"Dear@CAPS1 @CAPS2, I believe that using computers will benefit us in many ways like talking and becoming friends will others through websites like facebook and mysace. Using computers can help us find coordibates, locations, and able ourselfs to millions of information. Also computers will benefit us by helping with jobs as in planning a house plan and typing a @NUM1 page report for one of our jobs in less than writing it. Now lets go into the wonder world of technology. Using a computer will help us in life by talking or making friends on line. Many people have myspace, facebooks, aim, these all benefit us by having conversations with one another. Many people believe computers are bad but how can you make friends if you can never talk to them? I am very fortunate for having a computer that can help with not only school work but my social life and how I make friends. Computers help us with finding our locations, coordibates and millions of information online. If we didn't go on the internet a lot we wouldn't know how to go onto websites that @MONTH1 help us with locations and coordinates like @LOCATION1. Would you rather use a computer or be in @LOCATION3. When your supposed to be vacationing in @LOCATION2. Million of information is found on the internet. You can as almost every question and a computer will have it. Would you rather easily draw up a house plan on the computers or take @NUM1 hours doing one by hand with ugly erazer marks all over it, you are garrenteed that to find a job with a drawing like that. Also when appling for a job many workers must write very long papers like a @NUM3 word essay on why this job fits you the most, and many people I know don't like writing @NUM3 words non-stopp for hours when it could take them I hav an a computer. That is why computers we needed a lot now adays. I hope this essay has impacted your descion on computers because they are great machines to work with. The other day I showed my mom how to use a computer and she said it was the greatest invention sense sliced bread! Now go out and buy a computer to help you chat online with friends, find locations and millions of information on one click of the button and help your self with getting a job with neat, prepared, printed work that your boss will love.\"", "_____no_output_____" ], [ "from gensim.models import Word2Vec\nfrom gensim.models import KeyedVectors\nnum_features = 300\n \nmodel = KeyedVectors.load_word2vec_format( \"./word2vecmodel.bin\", binary=True)\nclean_test_essays = []\nclean_test_essays.append(essay_to_wordlist( content, remove_stopwords=True ))\ntestDataVecs = getAvgFeatureVecs( clean_test_essays, model, num_features )\ntestDataVecs = np.array(testDataVecs)\ntestDataVecs = np.reshape(testDataVecs, (testDataVecs.shape[0], 1, testDataVecs.shape[1]))\n\n# lstm_model = get_model()\nlstm_model.load_weights(\"./final_lstm.h5\")\npreds = lstm_model.predict(testDataVecs)", "_____no_output_____" ], [ "int(np.around(preds))", "_____no_output_____" ], [ "# val_essays = demo_df['essay']\n# sentences = []\n\n# for essay in val_essays:\n# sentences += essay_to_sentences(essay, remove_stopwords = True)\n \n\n# num_features = 300 \n# min_word_count = 40\n# num_workers = 4\n# context = 10\n# downsampling = 1e-3\n\n\n# model = Word2Vec(sentences, workers=num_workers, size=num_features, min_count = min_word_count, window = context, sample = downsampling)\n\n# model.init_sims(replace=True)\n\n# clean_train_essays = []\n\n# # Generate training and testing data word vectors.\n# for essay_v in val_essays:\n# clean_train_essays.append(essay_to_wordlist(essay_v, remove_stopwords=True))\n# trainDataVecs = getAvgFeatureVecs(clean_train_essays, model, num_features)\n\n# trainDataVecs = np.array(trainDataVecs)\n# # Reshaping train and test vectors to 3 dimensions. (1 represnts one timestep)\n# trainDataVecs = np.reshape(trainDataVecs, (trainDataVecs.shape[0], 1, trainDataVecs.shape[1]))\n\n\n# y_pred = lstm_model.predict(trainDataVecs)\n\n\n\n# y_pred = np.around(y_pred)\n\n\n# # result = cohen_kappa_score(y_test.values,y_pred,weights='quadratic')\n# # print(\"Kappa Score: {}\".format(result))\n# print(y_pred)", "_____no_output_____" ], [ "y_test.T.shape", "_____no_output_____" ], [ "y_pred1 = y_pred.T.reshape(2595,)", "_____no_output_____" ], [ "y_pred1", "_____no_output_____" ], [ "y_test1 = y_test.T.reshape", "_____no_output_____" ], [ "from sklearn.metrics import confusion_matrix\ncm = confusion_matrix(y_test, y_pred1)", "_____no_output_____" ], [ "# Save a palette to a variable:\npalette = sns.color_palette(\"bright\")\n \n# Use palplot and pass in the variable:\nsns.palplot(palette)", "_____no_output_____" ], [ "import seaborn as sns\nimport pandas as pd\nimport matplotlib.pyplot as plt\narray =cm[0:10,0:10]\ndf_cm = pd.DataFrame(array)\nplt.figure(figsize = (20,20))\nsns.heatmap(df_cm,cmap = 'Blues',square=True, annot=True)", "_____no_output_____" ], [ "cm[0:10,0:10]", "_____no_output_____" ], [ "arr = np.array([[1, 2, 3, 4, 5], [6, 7, 8, 9, 10]])\nprint(arr.shape)\nprint(arr[1:4, 4:1])\n# print(arr)", "(2, 5)\n[]\n" ], [ "cm.shape", "_____no_output_____" ], [ "corr = cm\n\nmask = np.triu(np.ones_like(corr, dtype=np.bool))\n\n# Set up the matplotlib figure\nf, ax = plt.subplots(figsize=(11, 9))\n\n# Generate a custom diverging colormap\ncmap = sns.diverging_palette(220, 10, as_cmap=True)\n\n# Draw the heatmap with the mask and correct aspect ratio\nsns.heatmap(corr, mask=mask, cmap=cmap, vmax=.3, center=0,\n square=True, linewidths=.5, cbar_kws={\"shrink\": .5})\nplt.title('Fig:1',size=15)", "_____no_output_____" ], [ "from sklearn.metrics import f1_score\nf1_score(y_test, y_pred1, average='macro')", "_____no_output_____" ], [ "from sklearn.metrics import classification_report\ncr = classification_report(y_test, y_pred1)", "_____no_output_____" ], [ "cr", "_____no_output_____" ], [ "import seaborn as sns\nimport numpy as np\nfrom sklearn.metrics import precision_recall_fscore_support\nimport matplotlib.pyplot as plt\n\ndef plot_classification_report(y_tru, y_prd, figsize=(10, 10), ax=None):\n\n plt.figure(figsize=figsize)\n\n xticks = ['precision', 'recall', 'f1-score', 'support']\n yticks = list(np.unique(y_tru))\n yticks += ['avg']\n\n rep = np.array(precision_recall_fscore_support(y_tru, y_prd)).T\n avg = np.mean(rep, axis=0)\n avg[-1] = np.sum(rep[:, -1])\n rep = np.insert(rep, rep.shape[0], avg, axis=0)\n\n sns.heatmap(rep,\n annot=True, \n cbar=False, \n xticklabels=xticks, \n yticklabels=yticks,\n ax=ax)\n\nplot_classification_report(y_test, y_pred1)", "_____no_output_____" ], [ "import matplotlib.pyplot as plt\nimport numpy as np\nimport itertools\n\n\ndef plot_classification_report(classificationReport,\n title='Classification report',\n cmap='RdBu'):\n\n classificationReport = classificationReport.replace('\\n\\n', '\\n')\n classificationReport = classificationReport.replace(' / ', '/')\n lines = classificationReport.split('\\n')\n\n classes, plotMat, support, class_names = [], [], [], []\n for line in lines[1:]: # if you don't want avg/total result, then change [1:] into [1:-1]\n t = line.strip().split()\n if len(t) < 2:\n continue\n classes.append(t[0])\n v = [float(x) for x in t[1: len(t) - 1]]\n support.append(int(t[-1]))\n class_names.append(t[0])\n plotMat.append(v)\n\n plotMat = np.array(plotMat)\n xticklabels = ['Precision', 'Recall', 'F1-score']\n yticklabels = ['{0} ({1})'.format(class_names[idx], sup)\n for idx, sup in enumerate(support)]\n\n plt.imshow(plotMat, interpolation='nearest', cmap=cmap, aspect='auto')\n plt.title(title)\n plt.colorbar()\n plt.xticks(np.arange(3), xticklabels, rotation=45)\n plt.yticks(np.arange(len(classes)), yticklabels)\n\n upper_thresh = plotMat.min() + (plotMat.max() - plotMat.min()) / 10 * 8\n lower_thresh = plotMat.min() + (plotMat.max() - plotMat.min()) / 10 * 2\n for i, j in itertools.product(range(plotMat.shape[0]), range(plotMat.shape[1])):\n plt.text(j, i, format(plotMat[i, j], '.2f'),\n horizontalalignment=\"center\",\n color=\"white\" if (plotMat[i, j] > upper_thresh or plotMat[i, j] < lower_thresh) else \"black\")\n\n plt.ylabel('Metrics')\n plt.xlabel('Classes')\n plt.tight_layout()\n\n\ndef main():\n\n sampleClassificationReport = cr\n plot_classification_report(sampleClassificationReport)\n plt.show()\n plt.close()\n\n\nif __name__ == '__main__':\n main()", "_____no_output_____" ], [ "from sklearn.linear_model import LogisticRegression\nfrom sklearn.svm import SVC\nfrom sklearn.neighbors import KNeighborsClassifier\nfrom sklearn.tree import DecisionTreeClassifier\nfrom sklearn.ensemble import RandomForestClassifier\nfrom sklearn.naive_bayes import MultinomialNB\nfrom sklearn.metrics import classification_report,confusion_matrix,accuracy_score,plot_confusion_matrix", "_____no_output_____" ], [ "classifiers = {\n \"LogisiticRegression\": LogisticRegression(),\n \"KNearest\": KNeighborsClassifier(n_neighbors=1),\n \"Support Vector Classifier\": SVC(),\n \"DecisionTreeClassifier\": DecisionTreeClassifier(),\n \"MultinimialNB\": MultinomialNB()\n}\n", "_____no_output_____" ], [ "trainDataVecs.shape[2]", "_____no_output_____" ], [ "trainDataVecs1 = np.reshape(trainDataVecs,trainDataVecs.shape[0],trainDataVecs.shape[2])\ntrainDataVecs1.shape", "_____no_output_____" ], [ "from sklearn.model_selection import cross_val_score\n\nclassifier = KNeighborsClassifier()\n\nclassifier.fit(trainDataVecs, y_train)\ntraining_score = cross_val_score(classifier, train_vectors, df_train[\"domain1_score\"], cv=5)\nprint(\"Classifiers: \", classifier.__class__.__name__, \"Has a training score of\", round(training_score.mean(), 2) * 100, \"% accuracy score\")", "_____no_output_____" ], [ "np.unique(df['domain1_score'],return_counts=True)", "_____no_output_____" ], [ "df", "_____no_output_____" ], [ "from sklearn.utils import class_weight\nclass_weights = class_weight.compute_class_weight('balanced',\n np.unique(df['domain1_score']),\n train)", "_____no_output_____" ] ] ]

[ "code" ]

[ [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# 1. Introduction to Natural Language Processing\nNatural Language Processing is certainly one of the most fascinating and exciting areas to be involved with at this point in time. It is a wonderful intersection of computer science, artificial intelligence, machine learning and linguistics. With the (somewhat) recent rise of Deep Learning, Natural Language Processing currently has a great deal of buzz surrounding it, and for good reason. The goal of this post is to do three things:\n\n1. Inspire the reader with the beauty of the problem of NLP\n2. Explain how machine learning techniques (i.e. something as simple as Logistic Regression) can be applied to text data.\n3. Prepare the reader for the next sections surrounding Deep Learning as it is applied to NLP.\n\nBefore we dive in, I would like to share the poem _Jabberwocky_ by Lewis Carrol, and an accompanying excerpt from the book \"_Godel, Escher, Bach_\", by Douglas Hofstadter.\n\n<img src=\"https://drive.google.com/uc?id=1ROLVf2p6xYyTqQ3fmeky0eSD6ZCdfJ3M\" width=\"300\">\n\nAnd now, the corresponding excerpt, _**Translations of Jabberwocky**_. \n\n> ### Translations of Jabberwocky<br>\nDouglas R. Hofstadter\nImagine native speakers of English, French, and German, all of whom have excellent command of their respective native languages, and all of whom enjoy wordplay in their own language. Would their symbol networks be similar on a local level, or on a global level? Or is it meaningful to ask such a question? The question becomes concrete when you look at the preceding translations of Lewis Carroll's famous \"Jabberwocky\".\n<br>\n<br>\n[The \"preceding translations\" were \"Jabberwocky\" (English, original), by Lewis Carroll, \"Le Jaseroque\", (French), by Frank L. Warrin, and \"Der Jammerwoch\" (German), by Robert Scott. --kl]\n<br>\n<br>\nI chose this example because it demonstrates, perhaps better than an example in ordinary prose, the problem of trying to find \"the same node\" in two different networks which are, on some level of analysis, extremely nonisomorphic. In ordinary language, the task of translation is more straightforward, since to each word or phrase in the original language, there can usually be found a corresponding word or phrase in the new language. By contrast, in a poem of this type, many \"words\" do not carry ordinary meaning, but act purely as exciters of nearby symbols. However, what is nearby in one language may be remote in another.\n<br>\n<br>\nThus, in the brain of a native speaker of English, \"slithy\" probably activates such symbols as \"slimy\", \"slither\", \"slippery\", \"lithe\", and \"sly\", to varying extents. Does \"lubricilleux\" do the corresponding thing in the brain of a Frenchman? What indeed would be \"the corresponding thing\"? Would it be to activate symbols which are the ordinary translations of those words? What if there is no word, real or fabricated, which will accomplish that? Or what if a word does exist, but it is very intellectual-sounding and Latinate (\"lubricilleux\"), rather than earthy and Anglo-Saxon (\"slithy\")? Perhaps \"huilasse\" would be better than \"lubricilleux\"? Or does the Latin origin of the word \"lubricilleux\" not make itself felt to a speaker of French in the way that it would if it were an English word (\"lubricilious\", perhaps)?\n<br>\n<br>\nAn interesting feature of the translation into French is the transposition into the present tense. To keep it in the past would make some unnatural turns of phrase necessary, and the present tense has a much fresher flavour in French than in the past. The translator sensed that this would be \"more appropriate\"--in some ill-defined yet compelling sense--and made the switch. Who can say whether remaining faithful to the English tense would have been better?\n<br>\n<br>\nIn the German version, the droll phrase \"er an-zu-denken-fing\" occurs; it does not correspond to any English original. It is a playful reversal of words, whose flavour vaguely resembles that of the English phrase \"he out-to-ponder set\", if I may hazard a reverse translation. Most likely this funny turnabout of words was inspired by the similar playful reversal in the English of one line earlier: \"So rested he by the Tumtum tree\". It corresponds, yet doesn't correspond.\n<br>\n<br>\nIncidentally, why did the Tumtum tree get changed into an \"arbre Té-té\" in French? Figure it out for yourself.\n<br>\n<br>\nThe word \"manxome\" in the original, whose \"x\" imbues it with many rich overtones, is weakly rendered in German by \"manchsam\", which back-translates into English as \"maniful\". The French \"manscant\" also lacks the manifold overtones of \"manxome\". There is no end to the interest of this kind of translation task.\n<br>\n<br>\nWhen confronted with such an example, one realizes that it is utterly impossible to make an exact translation. Yet even in this pathologically difficult case of translation, there seems to be some rough equivalence obtainable. Why is this so, if there really is no isomorphism between the brains of people who will read the different versions? The answer is that there is a kind of rough isomorphism, partly global, partly local, between the brains of all the readers of these three poems.\n\n\nNow, the purpose of sharing the above is because if you are reading these posts (and are anything like me), you may very well spend a large chunk of your time studying mathematics, computer science, machine learning, writing code, and so on. But, if you are new to NLP the appreciation for the beauty and deeper meaning surrounding language may not be on the forefront of your mind-that is understandable! But hopefully the passage and commentary above ignited some interest in the wonderfully complex and worthwhile problem of Natural Language Processing and Understanding.\n\n## 2. Spam Detection\nNow, especially at first, I don't want to dive into phonemes, morphemes, syntactical structure, and the like. We will leave those linguistic concepts for later on. The goal here is to quickly allow someone with an understanding of basic machine learning algorithms and techniques to implement them in the domain of NLP. \n\nWe will see that, at least at first, a lot of NLP deals with preprocessing data, which allows us to use algorithms that we already know. The question that most definitely arises is: How do we take a bunch of documents which are basically a bunch of text, and feed them into other machine learning algorithms where the input is usually a vector of numbers? \n\nWell, before we even get to that, let's take a preprocessed data set from the [uci archive](https://archive.ics.uci.edu/ml/datasets/Spambase) and perform a simple classification on it. The data has been processed in such a way that we can consider columns 1-48 to the be the input, and column 49 to the be label (1 = spam, 0 = not spam). \n\nThe input columns are considered the input, and they are a **word frequency measure**. This measure can be calculated via:\n\n$$\\text{Word Frequency Measure} = \\frac{\\text{# of times word appears in a document}}{\\text{Number of words in document}} * 100$$\n\nThis will result in a **Document Term matrix**, which is a matrix where _terms_ (words that appeared in the document) go along the columns, and _documents_ (emails in this case) go along the rows:\n\n| |word 1|word 2|word 3|word 4|word 5|word 6|word 7|word 8|\n|-------|------|------|------|------|------|------|------|------|\n|Email 1|||||||||\n|Email 2|||||||||\n|Email 3|||||||||\n|Email 4|||||||||\n|Email 5|||||||||\n\n### 2.1 Implementation in Code\nWe will now use `Scikit Learn` to show that we can use _any_ model on NLP data, as long as it has been preprocessed correctly. First, let's use scikit learns `NaiveBayes` classifier:", "_____no_output_____" ] ], [ [ "from sklearn.naive_bayes import MultinomialNB\nimport pandas as pd\nimport numpy as np", "_____no_output_____" ], [ "data = pd.read_csv('../../data/nlp/spambase.data')\ndata.head()", "_____no_output_____" ], [ "data = data.values\nnp.random.shuffle(data) # randomly split data into train and test sets\n\nX = data[:, :48]\nY = data[:, -1]\n\nXtrain = X[:-100,]\nYtrain = Y[:-100,]\nXtest = X[-100:,]\nYtest = Y[-100:,]\n\nmodel = MultinomialNB()\nmodel.fit(Xtrain, Ytrain)\nprint (\"Classifcation Rate for NB: \", model.score(Xtest, Ytest))", "Classifcation Rate for NB: 0.87\n" ] ], [ [ "Excellent, a classification rate of 92%! Let's now look utilize `AdaBoost`:", "_____no_output_____" ] ], [ [ "from sklearn.ensemble import AdaBoostClassifier\n\nmodel = AdaBoostClassifier()\nmodel.fit(Xtrain, Ytrain)\nprint (\"Classifcation Rate for Adaboost: \", model.score(Xtest, Ytest))", "Classifcation Rate for Adaboost: 0.94\n" ] ], [ [ "Great, a nice improvement, but more importantly, we have shown that we can take text data and that via correct preprocessing we are able to utilize it with standard machine learning API's. The next step is to dig into _how_ basic preprocessing is performed.", "_____no_output_____" ], [ "---\n\n# 3. Sentiment Analysis\nTo go through the basic preprocessing steps that are frequently used when performing machine learning on text data (often referred to an NLP pipeline) we are going to want to work on the problem of **sentiment analysis**. Sentiment is a measure of how positive or negative something is, and we are going to build a very simple sentiment analyzer to predict the sentiment of Amazon reviews. These are reviews, so they come with 5 star ratings, and we are going to look at the electronics category in particular. These are XML files, so we will need an XML parser. \n\n### 3.1 NLP Terminology \nBefore we begin, I would just like to quickly go over some basic NLP terminology that will come up frequently throughout this post.\n* **Corpus**: Collection of text\n* **Tokens**: Words and punctuation that make up the corpus. \n* **Type**: a distinct token. Ex. \"Run, Lola Run\" has four tokens (comma counts as one) and 3 types.\n* **Vocabulary**: The set of all types. \n* The google corpus (collection of text) has 1 trillion tokens, and only 13 million types. English only has 1 million dictionary words, but the google corpus includes types such as \"www.facebook.com\". \n\n### 3.2 Problem Overview\nNow, we are just going to be looking at the electronics category. We could use the 5 star targets to do regression, but instead we will just do classification since they are already marked \"positive\" and \"negative\". As I mentioned, we are going to be working with XML data, so we will need an XML parser, for which we will use `BeautifulSoup`. We will only look at the `review_text` attribute. To create our feature vector, we will count up the number of occurences of each word, and divided it by the total number of words. However, for that to work we will need two passes through the data:\n\n1. One to collect the total number of distinct words, so that we know the size of our feature vector, in other words the vocabulary size, and possibly remove stop words like \"this\", \"is\", \"I\", \"to\", etc, to decrease the vocabulary size. The goal here is to know the index of each token\n2. On the second pass, we will be able to assign values to each data vector whose index corresponds to which words, and one to create data vectors \n\nOnce we have that, it is simply a matter of creating a classifier like the one we did for our spam detector! Here, we will use logistic regression, so we can intepret the weights! For example, if you see a word like horrible and it has a weight of minus 1, it is associated with negative reviews. With that started, let's begin!", "_____no_output_____" ], [ "## 3.3 Sentiment Analysis in Code", "_____no_output_____" ] ], [ [ "import nltk\nimport numpy as np\n\nfrom nltk.stem import WordNetLemmatizer\nfrom sklearn.linear_model import LogisticRegression\nfrom bs4 import BeautifulSoup\n\nwordnet_lemmatizer = WordNetLemmatizer() # this turns words into their base form \n\nstopwords = set(w.rstrip() for w in open('../../data/nlp/stopwords.txt')) # grab stop words \n\n# get pos reviews\n# only want rev text\npositive_reviews = BeautifulSoup(open('../../data/nlp/electronics/positive.review').read(), \"lxml\") \npositive_reviews = positive_reviews.findAll('review_text') \n\nnegative_reviews = BeautifulSoup(open('../../data/nlp/electronics/negative.review').read(), \"lxml\")\nnegative_reviews = negative_reviews.findAll('review_text')", "_____no_output_____" ] ], [ [ "### 3.3.1 Class Imbalance\nThere are more positive than negative reviews, so we are going to shuffle the positive reviews and then cut off any extra that we may have so that they are both the same size.", "_____no_output_____" ] ], [ [ "np.random.shuffle(positive_reviews)\npositive_reviews = positive_reviews[:len(negative_reviews)]", "_____no_output_____" ] ], [ [ "### 3.3.2 Tokenizer function\nLets now create a tokenizer function that can be used on our specific reviews.", "_____no_output_____" ] ], [ [ "def my_tokenizer(s):\n s = s.lower()\n tokens = nltk.tokenize.word_tokenize(s) # essentially string.split()\n tokens = [t for t in tokens if len(t) > 2] # get rid of short words\n tokens = [wordnet_lemmatizer.lemmatize(t) for t in tokens] # get words to base form\n tokens = [t for t in tokens if t not in stopwords]\n return tokens", "_____no_output_____" ] ], [ [ "### 3.3.3 Index each word\nWe now need to create an index for each of the words, so that each word has an index in the final data vector. However, to able able to do that we need to know the size of the final data vector, and to be able to know that we need to know how big the vocabulary is. Remember, the **vocabulary** is just the set of all types!\n\nWe are essentially going to look at every individual review, tokenize them, and then add those tokens 1 by 1 to the map if they do not exist yet.", "_____no_output_____" ] ], [ [ "word_index_map = {} # our vocabulary - dictionary that will map words to dictionaries\ncurrent_index = 0 # counter increases whenever we see a new word\n\npositive_tokenized = []\nnegative_tokenized = []\n\n# --------- loop through positive reviews ---------\nfor review in positive_reviews: \n tokens = my_tokenizer(review.text) # converts single review into array of tokens (split function)\n positive_tokenized.append(tokens)\n for token in tokens: # loops through array of tokens for specific review\n if token not in word_index_map: # if the token is not in the map, add it\n word_index_map[token] = current_index \n current_index += 1 # increment current index\n \n# --------- loop through negative reviews ---------\nfor review in negative_reviews: \n tokens = my_tokenizer(review.text) \n negative_tokenized.append(tokens)\n for token in tokens: \n if token not in word_index_map: \n word_index_map[token] = current_index \n current_index += 1 ", "_____no_output_____" ] ], [ [ "And we can actually take a look at the contents of `word_index_map` by making use of the `random` module (part of the Python Standard Library):", "_____no_output_____" ] ], [ [ "import random\nprint(dict(random.sample(word_index_map.items(), 20)))", "{'tech-savvy': 5921, 'downloads': 2029, 're-acquire': 9930, 'megapixels': 7066, 'dual-amping': 6499, 'unsupported': 10981, 'configuration': 1183, '6000': 3246, 'obviously..': 9627, 'didn': 10133, 'eligible': 2440, 'lawn': 748, '50-pack': 1002, 'yearly..': 10956, '192.168.1.245': 2607, 'glad': 1844, 'occasionally': 1631, 'floppy': 7170, 'criminal': 4786, 'emptying': 4382}\n" ], [ "print('Vocabulary Size', len(word_index_map))", "Vocabulary Size 11088\n" ] ], [ [ "### 3.3.4 Convert tokens into vector\nNow that we have our tokens and vocabulary, we need to convert our tokens into a vector. Because we are going to shuffle our train and test sets again, we are going to want to put labels and vector into same array for now since it makes it easier to shuffle. \n\nNote, this function operates on **one** review. So the +1 is creating our label, and this function is basically designed to take our input vector from an english form to a numeric vector form.", "_____no_output_____" ] ], [ [ "def tokens_to_vector(tokens, label):\n xy_data = np.zeros(len(word_index_map) + 1) # equal to the vocab size + 1 for the label \n for t in tokens: # loop through every token\n i = word_index_map[t] # get index from word index map\n xy_data[i] += 1 # increment data at that index \n xy_data = xy_data / xy_data.sum() # divide entire array by total, so they add to 1\n xy_data[-1] = label # set last element to label\n return xy_data", "_____no_output_____" ] ], [ [ "Time to actually assign these tokens to vectors.", "_____no_output_____" ] ], [ [ "N = len(positive_tokenized) + len(negative_tokenized) # total number of examples \ndata = np.zeros((N, len(word_index_map) + 1)) # N examples x vocab size + 1 for label\ni = 0 # counter to keep track of sample\n\nfor tokens in positive_tokenized: # loop through postive tokenized reviews\n xy = tokens_to_vector(tokens, 1) # passing in 1 because these are pos reviews\n data[i,:] = xy # set data row to that of the input vector\n i += 1 # increment 1\n \nfor tokens in negative_tokenized: \n xy = tokens_to_vector(tokens, 0) \n data[i,:] = xy \n i += 1 ", "_____no_output_____" ], [ "print(data.shape)", "(2000, 11089)\n" ] ], [ [ "Our data is now 1000 rows of positively labeled reviews, followed by 1000 rows of negatively labeled reviews. We have `11089` columns, which is one more than our vocabulary size because we have a column for the label (positive or negative). Lets shuffle before getting our train and test set.", "_____no_output_____" ] ], [ [ "np.random.shuffle(data)\nX = data[:, :-1]\nY = data[:, -1]\n\nXtrain = X[:-100,]\nYtrain = Y[:-100,]\nXtest = X[-100:,]\nYtest = Y[-100:,]\n\n\nmodel = LogisticRegression()\nmodel.fit(Xtrain, Ytrain)\nprint(\"Classification Rate: \", model.score(Xtest, Ytest))", "Classification Rate: 0.7\n" ] ], [ [ "### 3.3.5 Classification Rate\nWe end up with a classification rate of 0.71, which is not ideal, but it is better than random guessing. \n\n### 3.3.6 Sentiment Analysis\nSomething interesting that we can do is look at the weights of each word, to see if that word has positive or negative sentiment. ", "_____no_output_____" ] ], [ [ "threshold = 0.7 \nlarge_magnitude_weights = []\nfor word, index in word_index_map.items():\n weight = model.coef_[0][index]\n if weight > threshold or weight < -threshold:\n large_magnitude_weights.append((word, weight))\n\ndef sort_by_magnitude(sentiment_dict):\n return sentiment_dict[1]\n \nlarge_magnitude_weights.sort(reverse=True, key=sort_by_magnitude)\nprint(large_magnitude_weights)", "[('price', 2.808163204024058), ('easy', 1.7646511704661152), ('quality', 1.3716522244882545), ('excellent', 1.319811182219224), ('love', 1.237745876552362), ('you', 1.155006377913112), ('perfect', 1.0324004425098248), ('sound', 0.9780126530219685), ('highly', 0.9778749978617105), ('memory', 0.9398953342479317), ('little', 0.9262682823592787), ('fast', 0.905207610856845), ('speaker', 0.8965845758701319), ('ha', 0.8111001120921802), ('pretty', 0.7764302324793534), ('cable', 0.7712191036378001), (\"'ve\", 0.7170298751638035), ('week', -0.7194449455694366), ('returned', -0.7482471935264389), ('bad', -0.7542948554985326), ('poor', -0.7555447694156194), ('tried', -0.7892866982929136), ('buy', -0.8504195601103998), ('month', -0.8771148641617261), ('support', -0.9163137326943319), ('waste', -0.946863186564699), ('item', -0.9518247418299971), ('money', -1.1086664158434432), ('return', -1.1512973579906935), ('then', -1.2084513223482118), ('doe', -1.2197007105871698), ('wa', -1.6630639259918825), (\"n't\", -2.0687949024413546)]\n" ] ], [ [ "Clearly the above list is not perfect, _but_ it should give some insight on what is possible for us already. The logistic regression model was able to pick out `easy`, `quality`, and `excellent` as words that correlate to a positive response, and it was able to find `poor`, `returned`, and `waste` as words the correlate to a negative response. ", "_____no_output_____" ], [ "---\n\n# 4. NLTK Exploration \nBefore we move on any further, I wanted to take a minute to go over a few of the most useful tools for the `nltk` (Natural Language Toolkit) library. This library will encapsulate many NLP tasks for us.\n\n### 4.1 Parts of Speech (POS) Tagging\nParts of speech tagging is meant to do just what it sound like: tag each word with a given part of speech within a document. For example, in the following sentence:\n\n> \"Bob is great.\"\n\n`Bob` is a noun, `is` is a verb, and `great` is an adjective. We can utilize `nltk`'s POS tagger on that sentence and see the same result:", "_____no_output_____" ] ], [ [ "import nltk\nnltk.pos_tag(\"Bob is great\".split())", "_____no_output_____" ], [ "nltk.pos_tag(\"Machine learning is great\".split())", "_____no_output_____" ] ], [ [ "The second entry in the above tuples `NN`, `VBZ`, etc, represents the determined tag of the word. For a description of each tag, check out [this link](https://www.ling.upenn.edu/courses/Fall_2003/ling001/penn_treebank_pos.html). ", "_____no_output_____" ], [ "### 4.2 Stemming and Lemmatization\nBoth the process of **stemming** and **lemmatization** are used in reducing words to a \"base\" form. This is very useful because a vocabulary can get very large, while certain words tend to have the same meaning. For example _dog_ and _dogs_, and _jump_ and _jumping_ both have similar meanings. The main difference between stemming and lemmatization is that stemming is a bit more basic. ", "_____no_output_____" ] ], [ [ "porter_stemmer = nltk.stem.porter.PorterStemmer()\nprint(porter_stemmer.stem('dogs'))", "dog\n" ], [ "print(porter_stemmer.stem('wolves'))", "wolv\n" ], [ "lemmatizer = nltk.stem.WordNetLemmatizer()\nprint(lemmatizer.lemmatize('dogs'))", "dog\n" ], [ "print(lemmatizer.lemmatize('wolves'))", "wolf\n" ] ], [ [ "Both the stemmer and lemmatizer managed to get `dogs` correct, but only the lemmatizer managed to correctly convert `wolves` to base form. ", "_____no_output_____" ], [ "### 4.3 Named Entity Recognition \nFinally there is **Named Entity** recognition. Entities refer to nouns such as:\n* \"Albert Einstein\" - a person\n* \"Apple\" - an organization", "_____no_output_____" ] ], [ [ "s = \"Albert Einstein was born on March 14, 1879\"\ntags = nltk.pos_tag(s.split())\nprint(tags)", "[('Albert', 'NNP'), ('Einstein', 'NNP'), ('was', 'VBD'), ('born', 'VBN'), ('on', 'IN'), ('March', 'NNP'), ('14,', 'CD'), ('1879', 'CD')]\n" ], [ "nltk.ne_chunk(tags)", "_____no_output_____" ], [ "s = \"Steve Jobs was the CEO of Apple Corp.\"\ntags = nltk.pos_tag(s.split())\nprint(tags)", "[('Steve', 'NNP'), ('Jobs', 'NNP'), ('was', 'VBD'), ('the', 'DT'), ('CEO', 'NNP'), ('of', 'IN'), ('Apple', 'NNP'), ('Corp.', 'NNP')]\n" ], [ "nltk.ne_chunk(tags)", "_____no_output_____" ] ], [ [ "---\n\n# 5. Latent Semantic Analysis\nWe will now take a moment to extend our semantic analysis example from before, instead now performing **Latent Semantic Analysis**. Latent semantic analysis is utilized to deal with the reality that we will often have _multiple_ words with the _same_ meaning, or on the other hand, _one_ word with _multiple_ meanings. These are referred to as _synonomy_ and _polysemy_ respectively. \n\nIn the case of synonyms here are a few basic examples:\n* \"Buy\" and \"Purchase\"\n* \"Big\" and \"Large\"\n* \"Quick\" and \"Speedy\"\n\nAnd in the case of polysemes:\n* \"Man\" (man as in human, and man as in a male opposed to a female)\n* \"Milk\" (can be a noun or a verb)\n\nIn order to solve this problem, we will need to introduce _Latent Variables_.\n\n## 5.1 Latent Variables\nThe easiest way to get your head around latent variables at first is via an example. Consider the words \"computer\", \"laptop\", and \"PC\"; these words are most likely seen together very often, meaning they are highly correlated. We can thinking a _latent_ or _hidden_ variable that is below representing them all, and we can call that $z$. We can mathematically define $z$ as:\n\n$$z = 0.7*computer \\; + 0.5*PC \\; + 0.6*laptop$$\n\nSo, we now have an idea of what a latent variable is, but what is the job of Latent Semantic Analysis? The entire goal of LSA is:\n\n1. To find the latent/hidden variables.\n2. Then, transform original data into these new variables. \n\nIdeally, after the above has been performed, the dimensionality of the new data will be much smaller than that of the original data set. It is important to note that LSA definitely helps solve the synonomy problem, by combining correlated variables. However, there are conflicting view points about whether or not it helps with polysemy. \n\n## 5.2 The Math Behind LSA\nAs we just discussed, the main goal when applying LSA is to deal with synonyms. For example, \"small\" and \"little\" would each make up their own unique variable, but in reality we know that they mean the same thing, so that is redundant. We could combine them into a single variable, reducing the dimensionality of our data set by one. So, to be clear the goal of LSA is:\n\n> **Goal of LSA**: Reduce redundancy.\n\n### 5.2.1 Redundancy in Numbers\nNow, machine learning at its core is always dealing with numbers, so what exactly do I mean by redundancy from a numerical standpoint? Take a look at the the plot below:\n\n<img src=\"https://drive.google.com/uc?id=1GK0COCtvumKXTI0nB0qHB_e696p2IE_J\" width=\"300\">\n\nWe can see clearly that there is a linear relationship between the dependent and independent variable. In other words, there is a linear relationship between lean body mass and muscle strength. So, we could say that one of these variables is redundant; if we know someones lean body mass, we can accurately predict their muscle strength, and vice versa. If we want a compact representation of attributes related to someones athletic performance, then we may only need to know one of these variables, since the other can be predicted from it. This advantage becomes more apparent as our dimensionality grows; if we could go from 1 million variables down to variables, that is a 200,000x's savings of space! Saving space is good, and hence reducing redundancy is good! \n\nNow the math behind LSA is rather complex and involves a good deal of linear algebra, and to be honest it would slightly bloat this notebook if I placed it here. Because of this, I have decided to move it to my mathematics section under linear algebra. With that said, LSA is essentially just the application of **Singular Value Decomposition** (SVD) to a term document matrix. I highly encourage you to go over my notebook explaining SVD and PCA before continuing, to have a better understanding of how the underlying mechanics work in the code we are about to implement. \n\nNow, we can begin by gaining a brief bit of intuition behind what LSA may look like in code. As usual, we are going to begin with an input matrix `X` of shape $NxD$, where $N$ is the number of samples and $D$ is the number of features. This will be passed into scikit learns svd model, `TruncatedSVD`, call the `fit`, `transform` function, and finally receive an output matrix `Z` of shape $Nx2$, or $Nxd$, where $d << D$. \n\n```\nmodel = TruncatedSVD()\nmodel.fit(X)\nZ = model.transform(X)\n# equivalent: Z = model.fit_transform(X)\n```\n\n## 5.3 LSA in Code", "_____no_output_____" ] ], [ [ "import nltk \nimport numpy as np\nimport matplotlib.pyplot as plt\nfrom nltk.stem import WordNetLemmatizer\nfrom sklearn.decomposition import TruncatedSVD", "_____no_output_____" ] ], [ [ "### Process:\n* we start by pulling in all of the titles, and all of the stop words. Our titles will look like:\n```\n['Philosophy of Sex and Love A Reader',\n 'Readings in Judaism, Christianity, and Islam',\n 'Microprocessors Principles and Applications',\n 'Bernhard Edouard Fernow: Story of North American Forestry',\n 'Encyclopedia of Buddhism',...]\n```\n* we then define our tokenizer which will convert our list of strings into specific tokens, which will look like:\n```\n[['philosophy', 'sex', 'love', 'reader'],\n ['reading', 'judaism', 'christianity', 'islam'],\n ['microprocessor', 'principle'],\n ['bernhard', 'edouard', 'fernow', 'story', 'north', 'american', 'forestry'],\n ['encyclopedia', 'buddhism'],\n```\n* we then create our input matrix. This is going to be D x N, where D is the length of the total number of terms we are using (input features, 2070) and where N is the length of all tokens (2373, the total number of titles)\n* This is essentially the transpose of how our input matrix is generally setup. Usually we have our examples along the rows, and our input features along the columns, however, in NLP it is sometimes the opposite\n* we then loop through all tokens, and create a vector for each one (essentially, if a word occurs, its value in the vector is incremented by 1)\n* the final input matrix is fed into the SVD, where the X matrix is transformed into a Z matrix of only 2 dimensions", "_____no_output_____" ] ], [ [ "wordnet_lemmatizer = WordNetLemmatizer()\n\ntitles = [line.rstrip() for line in open('../../data/nlp/all_book_titles.txt')] # Load all book titles in to an array\n\nstopwords = set(w.rstrip() for w in open('../../data/nlp/stopwords.txt')) # loading stop words (irrelevant)\nstopwords = stopwords.union({\n 'introduction', 'edition', 'series', 'application',\n 'approach', 'card', 'access', 'package', 'plus', 'etext',\n 'brief', 'vol', 'fundamental', 'guide', 'essential', 'printed',\n 'third', 'second', 'fourth', }) # adding additional stop words \n \n\ndef my_tokenizer(s):\n s = s.lower()\n tokens = nltk.tokenize.word_tokenize(s) # essentially string.split()\n tokens = [t for t in tokens if len(t) > 2] # get rid of short words\n tokens = [wordnet_lemmatizer.lemmatize(t) for t in tokens] # get words to base form\n tokens = [t for t in tokens if t not in stopwords] # remove stop words\n tokens = [t for t in tokens if not any(c.isdigit() for c in t)] # get rid of any token that includes a number\n return tokens\n\n# Lets now figure out the index of each word, by going through the entire vocabularly \n# create a word-to-index map so that we can create our word-frequency vectors later\n# let's also save the tokenized versions so we don't have to tokenize again later\nword_index_map = {}\ncurrent_index = 0\nall_tokens = []\nall_titles = []\nindex_word_map = []\nerror_count = 0\n\nfor title in titles:\n try:\n title = title.encode('ascii', 'ignore').decode('utf-8') # this will throw exception if bad characters\n all_titles.append(title)\n tokens = my_tokenizer(title)\n all_tokens.append(tokens)\n for token in tokens:\n if token not in word_index_map:\n word_index_map[token] = current_index\n current_index += 1\n index_word_map.append(token)\n except Exception as e:\n print(e)\n print(title)\n error_count += 1 \n \n# now let's create our input matrices - just indicator variables for this example - works better than proportions\ndef tokens_to_vector(tokens):\n x = np.zeros(len(word_index_map))\n for t in tokens:\n i = word_index_map[t]\n x[i] = 1\n return x\n\n\nN = len(all_tokens) # nested list, has 2373 total entries, each which has several words\nD = len(word_index_map) # total number of words that we are working with (2070)\nX = np.zeros((D, N)) # terms will go along rows, documents along columns\ni = 0\nfor tokens in all_tokens:\n X[:,i] = tokens_to_vector(tokens)\n i += 1\n\ndef main():\n fig, ax = plt.subplots(figsize=(12,8))\n svd = TruncatedSVD()\n Z = svd.fit_transform(X)\n plt.scatter(Z[:,0], Z[:,1])\n for i in range(D):\n plt.annotate(s=index_word_map[i], xy=(Z[i,0], Z[i,1]))\n plt.show()\n\nif __name__ == '__main__':\n main()\n", "_____no_output_____" ] ], [ [ "## Keep in mind...\nWhat is important to remember here is that the main point of this process is to take a bunch of words (2070) be able to plot their relationship to eachother in a 2-d surface. To do this, their use in 2373 book titles are considered, and SVD is performed. ", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown" ]

[ [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code", "code" ], [ "markdown", "markdown" ], [ "code", "code", "code", "code" ], [ "markdown", "markdown" ], [ "code", "code", "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Plagiarism Detection Model\n\nNow that you've created training and test data, you are ready to define and train a model. Your goal in this notebook, will be to train a binary classification model that learns to label an answer file as either plagiarized or not, based on the features you provide the model.\n\nThis task will be broken down into a few discrete steps:\n\n* Upload your data to S3.\n* Define a binary classification model and a training script.\n* Train your model and deploy it.\n* Evaluate your deployed classifier and answer some questions about your approach.\n\nTo complete this notebook, you'll have to complete all given exercises and answer all the questions in this notebook.\n> All your tasks will be clearly labeled **EXERCISE** and questions as **QUESTION**.\n\nIt will be up to you to explore different classification models and decide on a model that gives you the best performance for this dataset.\n\n---", "_____no_output_____" ], [ "## Load Data to S3\n\nIn the last notebook, you should have created two files: a `training.csv` and `test.csv` file with the features and class labels for the given corpus of plagiarized/non-plagiarized text data. \n\n>The below cells load in some AWS SageMaker libraries and creates a default bucket. After creating this bucket, you can upload your locally stored data to S3.\n\nSave your train and test `.csv` feature files, locally. To do this you can run the second notebook \"2_Plagiarism_Feature_Engineering\" in SageMaker or you can manually upload your files to this notebook using the upload icon in Jupyter Lab. Then you can upload local files to S3 by using `sagemaker_session.upload_data` and pointing directly to where the training data is saved.", "_____no_output_____" ] ], [ [ "import pandas as pd\nimport boto3\nimport sagemaker", "_____no_output_____" ], [ "\"\"\"\nDON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE\n\"\"\"\n# session and role\nsagemaker_session = sagemaker.Session()\nrole = sagemaker.get_execution_role()\n\n# create an S3 bucket\nbucket = sagemaker_session.default_bucket()", "_____no_output_____" ] ], [ [ "## EXERCISE: Upload your training data to S3\n\nSpecify the `data_dir` where you've saved your `train.csv` file. Decide on a descriptive `prefix` that defines where your data will be uploaded in the default S3 bucket. Finally, create a pointer to your training data by calling `sagemaker_session.upload_data` and passing in the required parameters. It may help to look at the [Session documentation](https://sagemaker.readthedocs.io/en/stable/session.html#sagemaker.session.Session.upload_data) or previous SageMaker code examples.\n\nYou are expected to upload your entire directory. Later, the training script will only access the `train.csv` file.", "_____no_output_____" ] ], [ [ "# should be the name of directory you created to save your features data\ndata_dir = 'plagiarism_data'\n\n# set prefix, a descriptive name for a directory \nprefix = 'sagemaker/plagiarism-detection'\n\n# upload all data to S3\ninput_data = sagemaker_session.upload_data(path=data_dir, bucket=bucket, key_prefix=prefix)", "_____no_output_____" ] ], [ [ "### Test cell\n\nTest that your data has been successfully uploaded. The below cell prints out the items in your S3 bucket and will throw an error if it is empty. You should see the contents of your `data_dir` and perhaps some checkpoints. If you see any other files listed, then you may have some old model files that you can delete via the S3 console (though, additional files shouldn't affect the performance of model developed in this notebook).", "_____no_output_____" ] ], [ [ "\"\"\"\nDON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE\n\"\"\"\n# confirm that data is in S3 bucket\nempty_check = []\nfor obj in boto3.resource('s3').Bucket(bucket).objects.all():\n empty_check.append(obj.key)\n print(obj.key)\n\nassert len(empty_check) !=0, 'S3 bucket is empty.'\nprint('Test passed!')", "Lambda/\nLambda/lambda_function.zip\nLambda/package.zip\nLambda/plagiarism_detection_func-8a2856a0-4389-44ac-a08f-24e25d799fca.zip\nLambda/sample-site-packages-2016-02-20.zip\nPanda_Layer.zip\nboston-update-endpoints/train.csv\nboston-update-endpoints/validation.csv\nboston-xgboost-HL/output/xgboost-2020-10-05-09-40-13-851/output/model.tar.gz\nboston-xgboost-HL/test.csv\nboston-xgboost-HL/train.csv\nboston-xgboost-HL/validation.csv\nboston-xgboost-LL/batch-bransform/test.csv.out\nboston-xgboost-LL/output/boston-xgboost-2020-10-05-08-45-40/output/model.tar.gz\nboston-xgboost-LL/test.csv\nboston-xgboost-LL/train.csv\nboston-xgboost-LL/validation.csv\ncounties/kmeans-2020-10-12-15-19-49-497/output/model.tar.gz\ncounties/kmeans-2020-10-12-16-12-40-795/output/model.tar.gz\ncounties/pca-2020-10-12-11-38-56-130/output/model.tar.gz\ncounties/pca-2020-10-12-12-36-11-407/output/model.tar.gz\nfraund_detection/linear-learner-2020-10-13-16-25-53-808/output/model.tar.gz\nfraund_detection/linear-learner-2020-10-13-18-26-50-429/output/model.tar.gz\nfraund_detection/linear-learner-2020-10-13-18-57-22-416/output/model.tar.gz\nfraund_detection/linear-learner-2020-10-13-19-27-34-058/output/model.tar.gz\nfraund_detection/linear-learner-2020-10-13-19-52-03-257/output/model.tar.gz\nfraund_detection/linear-learner-2020-10-13-20-12-26-687/output/model.tar.gz\nfraund_detection/linear-learner-2020-10-13-20-33-19-916/output/model.tar.gz\nfraund_detection/linear-learner-2020-10-13-20-40-04-004/output/model.tar.gz\nfraund_detection/linear-learner-2020-10-13-21-00-43-603/output/model.tar.gz\nfraund_detection/linear-learner-2020-10-14-08-57-57-060/output/model.tar.gz\nlambda.zip\nlambda_panda_layer-835b6e94-93de-4e5d-abda-4586b6b7f69d.zip\nmoon-data/sagemaker-pytorch-2020-10-15-11-44-11-502/debug-output/training_job_end.ts\nmoon-data/sagemaker-pytorch-2020-10-15-11-44-11-502/output/model.tar.gz\nmoon-data/sagemaker-pytorch-2020-10-15-12-19-41-161/debug-output/training_job_end.ts\nmoon-data/sagemaker-pytorch-2020-10-15-12-19-41-161/output/model.tar.gz\nmoon-data/sagemaker-pytorch-2020-10-15-12-37-52-303/output/model.tar.gz\nmoon-data/sagemaker-pytorch-2020-10-15-12-55-13-731/debug-output/training_job_end.ts\nmoon-data/sagemaker-pytorch-2020-10-15-12-55-13-731/output/model.tar.gz\npython.zip\npython3.zip\nsagemaker-pytorch-2020-10-23-18-08-26-412/source/sourcedir.tar.gz\nsagemaker-pytorch-2020-10-23-18-14-07-463/source/sourcedir.tar.gz\nsagemaker-pytorch-2020-10-23-18-19-37-689/source/sourcedir.tar.gz\nsagemaker-record-sets/KMeans-2020-10-12-15-19-36-131/.amazon.manifest\nsagemaker-record-sets/KMeans-2020-10-12-15-19-36-131/matrix_0.pbr\nsagemaker-record-sets/LinearLearner-2020-10-13-16-12-30-860/.amazon.manifest\nsagemaker-record-sets/LinearLearner-2020-10-13-16-12-30-860/matrix_0.pbr\nsagemaker-record-sets/LinearLearner-2020-10-13-16-25-36-894/.amazon.manifest\nsagemaker-record-sets/LinearLearner-2020-10-13-16-25-36-894/matrix_0.pbr\nsagemaker-record-sets/LinearLearner-2020-10-13-18-26-27-878/.amazon.manifest\nsagemaker-record-sets/LinearLearner-2020-10-13-18-26-27-878/matrix_0.pbr\nsagemaker-record-sets/PCA-2020-10-12-11-29-48-805/.amazon.manifest\nsagemaker-record-sets/PCA-2020-10-12-11-29-48-805/matrix_0.pbr\nsagemaker-record-sets/PCA-2020-10-12-12-36-08-957/.amazon.manifest\nsagemaker-record-sets/PCA-2020-10-12-12-36-08-957/matrix_0.pbr\nsagemaker-scikit-learn-2020-10-20-13-20-22-809/debug-output/training_job_end.ts\nsagemaker-scikit-learn-2020-10-20-13-20-22-809/output/model.tar.gz\nsagemaker-scikit-learn-2020-10-20-13-20-22-809/source/sourcedir.tar.gz\nsagemaker-scikit-learn-2020-10-23-17-25-30-271/source/sourcedir.tar.gz\nsagemaker-scikit-learn-2020-10-23-17-32-58-378/source/sourcedir.tar.gz\nsagemaker-scikit-learn-2020-10-23-17-40-45-279/source/sourcedir.tar.gz\nsagemaker-scikit-learn-2020-10-23-17-52-31-242/source/sourcedir.tar.gz\nsagemaker-scikit-learn-2020-10-23-18-02-22-277/debug-output/training_job_end.ts\nsagemaker-scikit-learn-2020-10-23-18-02-22-277/output/model.tar.gz\nsagemaker-scikit-learn-2020-10-23-18-02-22-277/source/sourcedir.tar.gz\nsagemaker/energy_consumption/forecasting-deepar-2020-10-15-20-57-08-845/output/model.tar.gz\nsagemaker/energy_consumption/test.json\nsagemaker/energy_consumption/train.json\nsagemaker/moon-data/train.csv\nsagemaker/plagiarism-detection/pytorch/output/sagemaker-pytorch-2020-10-19-20-24-35-558/debug-output/training_job_end.ts\nsagemaker/plagiarism-detection/pytorch/output/sagemaker-pytorch-2020-10-19-20-24-35-558/output/model.tar.gz\nsagemaker/plagiarism-detection/pytorch/output/sagemaker-pytorch-2020-10-19-20-47-37-049/debug-output/training_job_end.ts\nsagemaker/plagiarism-detection/pytorch/output/sagemaker-pytorch-2020-10-19-20-47-37-049/output/model.tar.gz\nsagemaker/plagiarism-detection/pytorch/output/sagemaker-pytorch-2020-10-19-21-08-53-391/debug-output/training_job_end.ts\nsagemaker/plagiarism-detection/pytorch/output/sagemaker-pytorch-2020-10-19-21-08-53-391/output/model.tar.gz\nsagemaker/plagiarism-detection/pytorch/output/sagemaker-pytorch-2020-10-23-18-19-37-689/debug-output/training_job_end.ts\nsagemaker/plagiarism-detection/pytorch/output/sagemaker-pytorch-2020-10-23-18-19-37-689/output/model.tar.gz\nsagemaker/plagiarism-detection/test.csv\nsagemaker/plagiarism-detection/train.csv\nsagemaker/sentiment_rnn/train.csv\nsagemaker/sentiment_rnn/word_dict.pkl\nsample-site-packages-2016-02-20.zip\nsentiment-web-app/output/xgboost-2020-10-07-14-49-14-975/output/model.tar.gz\nsentiment-web-app/test.csv\nsentiment-web-app/train.csv\nsentiment-web-app/validation.csv\nsklearn-build-lambda-master.zip\nsklearn.zip\nsklearn1.zip\nTest passed!\n" ] ], [ [ "---\n\n# Modeling\n\nNow that you've uploaded your training data, it's time to define and train a model!\n\nThe type of model you create is up to you. For a binary classification task, you can choose to go one of three routes:\n* Use a built-in classification algorithm, like LinearLearner.\n* Define a custom Scikit-learn classifier, a comparison of models can be found [here](https://scikit-learn.org/stable/auto_examples/classification/plot_classifier_comparison.html).\n* Define a custom PyTorch neural network classifier. \n\nIt will be up to you to test out a variety of models and choose the best one. Your project will be graded on the accuracy of your final model. \n \n---\n\n## EXERCISE: Complete a training script \n\nTo implement a custom classifier, you'll need to complete a `train.py` script. You've been given the folders `source_sklearn` and `source_pytorch` which hold starting code for a custom Scikit-learn model and a PyTorch model, respectively. Each directory has a `train.py` training script. To complete this project **you only need to complete one of these scripts**; the script that is responsible for training your final model.\n\nA typical training script:\n* Loads training data from a specified directory\n* Parses any training & model hyperparameters (ex. nodes in a neural network, training epochs, etc.)\n* Instantiates a model of your design, with any specified hyperparams\n* Trains that model \n* Finally, saves the model so that it can be hosted/deployed, later\n\n### Defining and training a model\nMuch of the training script code is provided for you. Almost all of your work will be done in the `if __name__ == '__main__':` section. To complete a `train.py` file, you will:\n1. Import any extra libraries you need\n2. Define any additional model training hyperparameters using `parser.add_argument`\n2. Define a model in the `if __name__ == '__main__':` section\n3. Train the model in that same section\n\nBelow, you can use `!pygmentize` to display an existing `train.py` file. Read through the code; all of your tasks are marked with `TODO` comments. \n\n**Note: If you choose to create a custom PyTorch model, you will be responsible for defining the model in the `model.py` file,** and a `predict.py` file is provided. If you choose to use Scikit-learn, you only need a `train.py` file; you may import a classifier from the `sklearn` library.", "_____no_output_____" ] ], [ [ "# directory can be changed to: source_sklearn or source_pytorch\n!pygmentize source_sklearn/train.py", "\u001b[34mfrom\u001b[39;49;00m \u001b[04m\u001b[36m__future__\u001b[39;49;00m \u001b[34mimport\u001b[39;49;00m print_function\r\n\r\n\u001b[34mimport\u001b[39;49;00m \u001b[04m\u001b[36margparse\u001b[39;49;00m\r\n\u001b[34mimport\u001b[39;49;00m \u001b[04m\u001b[36mos\u001b[39;49;00m\r\n\u001b[34mimport\u001b[39;49;00m \u001b[04m\u001b[36mpandas\u001b[39;49;00m \u001b[34mas\u001b[39;49;00m \u001b[04m\u001b[36mpd\u001b[39;49;00m\r\n\r\n\u001b[34mfrom\u001b[39;49;00m \u001b[04m\u001b[36msklearn\u001b[39;49;00m\u001b[04m\u001b[36m.\u001b[39;49;00m\u001b[04m\u001b[36mexternals\u001b[39;49;00m \u001b[34mimport\u001b[39;49;00m joblib\r\n\u001b[34mfrom\u001b[39;49;00m \u001b[04m\u001b[36mskorch\u001b[39;49;00m \u001b[34mimport\u001b[39;49;00m NeuralNetRegressor\r\n\u001b[34mfrom\u001b[39;49;00m \u001b[04m\u001b[36mmodel\u001b[39;49;00m \u001b[34mimport\u001b[39;49;00m BinaryClassifier\r\n\u001b[34mfrom\u001b[39;49;00m \u001b[04m\u001b[36msklearn\u001b[39;49;00m\u001b[04m\u001b[36m.\u001b[39;49;00m\u001b[04m\u001b[36mmodel_selection\u001b[39;49;00m \u001b[34mimport\u001b[39;49;00m GridSearchCV\r\n\r\n\u001b[37m#from sklearn.svm import LinearSVC\u001b[39;49;00m\r\n\u001b[37m## TODO: Import any additional libraries you need to define a model\u001b[39;49;00m\r\n\r\n\u001b[37m#Begin Yanfei's first try#\u001b[39;49;00m\r\n\u001b[37m#from sklearn.svm import LinearSVC\u001b[39;49;00m\r\n\u001b[37m#End Yanfei's first try#\u001b[39;49;00m\r\n\r\n\u001b[37m#Begin Yanfei's second try#\u001b[39;49;00m\r\n\u001b[34mfrom\u001b[39;49;00m \u001b[04m\u001b[36msklearn\u001b[39;49;00m\u001b[04m\u001b[36m.\u001b[39;49;00m\u001b[04m\u001b[36mlinear_model\u001b[39;49;00m \u001b[34mimport\u001b[39;49;00m LogisticRegression\r\n\u001b[37m#Begin Yanfei's second try#\u001b[39;49;00m\r\n\u001b[37m# Provided model load function\u001b[39;49;00m\r\n\u001b[34mdef\u001b[39;49;00m \u001b[32mmodel_fn\u001b[39;49;00m(model_dir):\r\n \u001b[33m\"\"\"Load model from the model_dir. This is the same model that is saved\u001b[39;49;00m\r\n\u001b[33m in the main if statement.\u001b[39;49;00m\r\n\u001b[33m \"\"\"\u001b[39;49;00m\r\n \u001b[36mprint\u001b[39;49;00m(\u001b[33m\"\u001b[39;49;00m\u001b[33mLoading model.\u001b[39;49;00m\u001b[33m\"\u001b[39;49;00m)\r\n \r\n \u001b[37m# load using joblib\u001b[39;49;00m\r\n model = joblib.load(os.path.join(model_dir, \u001b[33m\"\u001b[39;49;00m\u001b[33mmodel.joblib\u001b[39;49;00m\u001b[33m\"\u001b[39;49;00m))\r\n \u001b[36mprint\u001b[39;49;00m(\u001b[33m\"\u001b[39;49;00m\u001b[33mDone loading model.\u001b[39;49;00m\u001b[33m\"\u001b[39;49;00m)\r\n \r\n \u001b[34mreturn\u001b[39;49;00m model\r\n\r\n\r\n\u001b[37m## TODO: Complete the main code\u001b[39;49;00m\r\n\u001b[34mif\u001b[39;49;00m \u001b[31m__name__\u001b[39;49;00m == \u001b[33m'\u001b[39;49;00m\u001b[33m__main__\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m:\r\n \r\n \u001b[37m# All of the model parameters and training parameters are sent as arguments\u001b[39;49;00m\r\n \u001b[37m# when this script is executed, during a training job\u001b[39;49;00m\r\n \r\n \u001b[37m# Here we set up an argument parser to easily access the parameters\u001b[39;49;00m\r\n parser = argparse.ArgumentParser()\r\n\r\n \u001b[37m# SageMaker parameters, like the directories for training data and saving models; set automatically\u001b[39;49;00m\r\n \u001b[37m# Do not need to change\u001b[39;49;00m\r\n parser.add_argument(\u001b[33m'\u001b[39;49;00m\u001b[33m--output-data-dir\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m, \u001b[36mtype\u001b[39;49;00m=\u001b[36mstr\u001b[39;49;00m, default=os.environ[\u001b[33m'\u001b[39;49;00m\u001b[33mSM_OUTPUT_DATA_DIR\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m])\r\n parser.add_argument(\u001b[33m'\u001b[39;49;00m\u001b[33m--model-dir\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m, \u001b[36mtype\u001b[39;49;00m=\u001b[36mstr\u001b[39;49;00m, default=os.environ[\u001b[33m'\u001b[39;49;00m\u001b[33mSM_MODEL_DIR\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m])\r\n parser.add_argument(\u001b[33m'\u001b[39;49;00m\u001b[33m--data-dir\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m, \u001b[36mtype\u001b[39;49;00m=\u001b[36mstr\u001b[39;49;00m, default=os.environ[\u001b[33m'\u001b[39;49;00m\u001b[33mSM_CHANNEL_TRAIN\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m])\r\n \r\n \u001b[37m## TODO: Add any additional arguments that you will need to pass into your model\u001b[39;49;00m\r\n parser.add_argument(\u001b[33m'\u001b[39;49;00m\u001b[33m--random_state\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m, \u001b[36mtype\u001b[39;49;00m=\u001b[36mint\u001b[39;49;00m, default=\u001b[34m0\u001b[39;49;00m, metavar=\u001b[33m'\u001b[39;49;00m\u001b[33mN\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m,\r\n help=\u001b[33m'\u001b[39;49;00m\u001b[33mint, RandomState instance, default=0\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m)\r\n parser.add_argument(\u001b[33m'\u001b[39;49;00m\u001b[33m--solver\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m, \u001b[36mtype\u001b[39;49;00m=\u001b[36mstr\u001b[39;49;00m, default=\u001b[33m'\u001b[39;49;00m\u001b[33mlbfgs\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m, metavar=\u001b[33m'\u001b[39;49;00m\u001b[33mS\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m,\r\n help=\u001b[33m'\u001b[39;49;00m\u001b[33mPossible values: \u001b[39;49;00m\u001b[33m{\u001b[39;49;00m\u001b[33mnewton-cg, lbfgs, liblinear, sag, saga}, default is lbfgs\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m)\r\n parser.add_argument(\u001b[33m'\u001b[39;49;00m\u001b[33m--multi_class\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m, \u001b[36mtype\u001b[39;49;00m=\u001b[36mstr\u001b[39;49;00m, default=\u001b[33m'\u001b[39;49;00m\u001b[33movr\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m, metavar=\u001b[33m'\u001b[39;49;00m\u001b[33mS\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m,\r\n help=\u001b[33m'\u001b[39;49;00m\u001b[33mPossible values: \u001b[39;49;00m\u001b[33m{\u001b[39;49;00m\u001b[33mauto, ovr, multinomial}, default is ovr\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m)\r\n \u001b[37m# args holds all passed-in arguments\u001b[39;49;00m\r\n args = parser.parse_args()\r\n\r\n \u001b[37m# Read in csv training file\u001b[39;49;00m\r\n training_dir = args.data_dir\r\n train_data = pd.read_csv(os.path.join(training_dir, \u001b[33m\"\u001b[39;49;00m\u001b[33mtrain.csv\u001b[39;49;00m\u001b[33m\"\u001b[39;49;00m), header=\u001b[34mNone\u001b[39;49;00m, names=\u001b[34mNone\u001b[39;49;00m)\r\n\r\n \u001b[37m# Labels are in the first column\u001b[39;49;00m\r\n train_y = train_data.iloc[:,\u001b[34m0\u001b[39;49;00m]\r\n train_x = train_data.iloc[:,\u001b[34m1\u001b[39;49;00m:]\r\n \r\n \r\n \u001b[37m## --- Your code here --- ##\u001b[39;49;00m\r\n \r\n\r\n \u001b[37m## TODO: Define a model \u001b[39;49;00m\r\n \r\n \u001b[37m#Begin Yanfei's first try#\u001b[39;49;00m\r\n \u001b[37m#model=LinearSVC()\u001b[39;49;00m\r\n \u001b[37m#End Yanfei's first try#\u001b[39;49;00m\r\n net = NeuralNetRegressor(BinaryClassifier(args.input_features,args.hidden_dim,args.output_dim)\r\n , max_epochs=\u001b[34m100\u001b[39;49;00m\r\n , lr=\u001b[34m0.001\u001b[39;49;00m\r\n , verbose=\u001b[34m1\u001b[39;49;00m)\r\n \r\n \u001b[37m#Begin Yanfei's second try#\u001b[39;49;00m\r\n model = LogisticRegression(random_state=args.random_state, solver=args.solver, multi_class=args.multi_class)\r\n \u001b[37m#End Yanfei's first try#\u001b[39;49;00m\r\n params = {\r\n \u001b[33m'\u001b[39;49;00m\u001b[33mlr\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m: [\u001b[34m0.001\u001b[39;49;00m,\u001b[34m0.005\u001b[39;49;00m, \u001b[34m0.01\u001b[39;49;00m, \u001b[34m0.05\u001b[39;49;00m, \u001b[34m0.1\u001b[39;49;00m, \u001b[34m0.2\u001b[39;49;00m, \u001b[34m0.3\u001b[39;49;00m],\r\n \u001b[33m'\u001b[39;49;00m\u001b[33mmax_epochs\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m: \u001b[36mlist\u001b[39;49;00m(\u001b[36mrange\u001b[39;49;00m(\u001b[34m500\u001b[39;49;00m,\u001b[34m5500\u001b[39;49;00m, \u001b[34m500\u001b[39;49;00m))\r\n }\r\n\r\n \u001b[37m## TODO: Train the model\u001b[39;49;00m\r\n \u001b[37m# model.fit(train_x,train_y)\u001b[39;49;00m\r\n \u001b[37m#model = GridSearchCV(net, params, refit=False, scoring='r2', verbose=1, cv=10)\u001b[39;49;00m\r\n\r\n model.fit(train_x, train_y)\r\n \r\n \u001b[37m## --- End of your code --- ##\u001b[39;49;00m\r\n \r\n\r\n \u001b[37m# Save the trained model\u001b[39;49;00m\r\n joblib.dump(model, os.path.join(args.model_dir, \u001b[33m\"\u001b[39;49;00m\u001b[33mmodel.joblib\u001b[39;49;00m\u001b[33m\"\u001b[39;49;00m))\r\n" ] ], [ [ "### Provided code\n\nIf you read the code above, you can see that the starter code includes a few things:\n* Model loading (`model_fn`) and saving code\n* Getting SageMaker's default hyperparameters\n* Loading the training data by name, `train.csv` and extracting the features and labels, `train_x`, and `train_y`\n\nIf you'd like to read more about model saving with [joblib for sklearn](https://scikit-learn.org/stable/modules/model_persistence.html) or with [torch.save](https://pytorch.org/tutorials/beginner/saving_loading_models.html), click on the provided links.", "_____no_output_____" ], [ "---\n# Create an Estimator\n\nWhen a custom model is constructed in SageMaker, an entry point must be specified. This is the Python file which will be executed when the model is trained; the `train.py` function you specified above. To run a custom training script in SageMaker, construct an estimator, and fill in the appropriate constructor arguments:\n\n* **entry_point**: The path to the Python script SageMaker runs for training and prediction.\n* **source_dir**: The path to the training script directory `source_sklearn` OR `source_pytorch`.\n* **entry_point**: The path to the Python script SageMaker runs for training and prediction.\n* **source_dir**: The path to the training script directory `train_sklearn` OR `train_pytorch`.\n* **entry_point**: The path to the Python script SageMaker runs for training.\n* **source_dir**: The path to the training script directory `train_sklearn` OR `train_pytorch`.\n* **role**: Role ARN, which was specified, above.\n* **train_instance_count**: The number of training instances (should be left at 1).\n* **train_instance_type**: The type of SageMaker instance for training. Note: Because Scikit-learn does not natively support GPU training, Sagemaker Scikit-learn does not currently support training on GPU instance types.\n* **sagemaker_session**: The session used to train on Sagemaker.\n* **hyperparameters** (optional): A dictionary `{'name':value, ..}` passed to the train function as hyperparameters.\n\nNote: For a PyTorch model, there is another optional argument **framework_version**, which you can set to the latest version of PyTorch, `1.0`.\n\n## EXERCISE: Define a Scikit-learn or PyTorch estimator\n\nTo import your desired estimator, use one of the following lines:\n```\nfrom sagemaker.sklearn.estimator import SKLearn\n```\n```\nfrom sagemaker.pytorch import PyTorch\n```", "_____no_output_____" ] ], [ [ "# your import and estimator code, here\nfrom sagemaker.sklearn.estimator import SKLearn\n\n# specify an output path\nprefix = 'sagemaker/plagiarism-detection/output'\n\n# define location to store model artifacts\n\noutput_path='s3://{}/{}/'.format(bucket, prefix)\n\n# instantiate a pytorch estimator\nestimator = SKLearn(entry_point=\"train.py\",\n source_dir=\"source_sklearn\",\n role=role,\n train_instance_count=1,\n train_instance_type='ml.c4.xlarge'\n \n )", "This is not the latest supported version. If you would like to use version 0.23-1, please add framework_version=0.23-1 to your constructor.\n" ] ], [ [ "## EXERCISE: Train the estimator\n\nTrain your estimator on the training data stored in S3. This should create a training job that you can monitor in your SageMaker console.", "_____no_output_____" ] ], [ [ "%%time\n\n# Train your estimator on S3 training data\nestimator.fit({'train': input_data})\n", "'s3_input' class will be renamed to 'TrainingInput' in SageMaker Python SDK v2.\n" ] ], [ [ "## EXERCISE: Deploy the trained model\n\nAfter training, deploy your model to create a `predictor`. If you're using a PyTorch model, you'll need to create a trained `PyTorchModel` that accepts the trained `<model>.model_data` as an input parameter and points to the provided `source_pytorch/predict.py` file as an entry point. \n\nTo deploy a trained model, you'll use `<model>.deploy`, which takes in two arguments:\n* **initial_instance_count**: The number of deployed instances (1).\n* **instance_type**: The type of SageMaker instance for deployment.\n\nNote: If you run into an instance error, it may be because you chose the wrong training or deployment instance_type. It may help to refer to your previous exercise code to see which types of instances we used.", "_____no_output_____" ] ], [ [ "%%time\n#from sagemaker.sklearn.model import SKLearnModel\n# uncomment, if needed\n# from sagemaker.pytorch import PyTorchModel\n\n#model=SKLearnModel(model_data=estimator.model_data,\n # role = role,\n # framework_version='0.23-1',\n # entry_point='train.py',\n # source_dir='source_sklearn')\n# deploy your model to create a predictor\npredictor = estimator.deploy(initial_instance_count=1, instance_type='ml.m4.xlarge')\n", "Parameter image will be renamed to image_uri in SageMaker Python SDK v2.\n" ] ], [ [ "---\n# Evaluating Your Model\n\nOnce your model is deployed, you can see how it performs when applied to our test data.\n\nThe provided cell below, reads in the test data, assuming it is stored locally in `data_dir` and named `test.csv`. The labels and features are extracted from the `.csv` file.", "_____no_output_____" ] ], [ [ "\"\"\"\nDON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE\n\"\"\"\nimport os\n\n# read in test data, assuming it is stored locally\ntest_data = pd.read_csv(os.path.join(data_dir, \"test.csv\"), header=None, names=None)\n\n# labels are in the first column\ntest_y = test_data.iloc[:,0]\ntest_x = test_data.iloc[:,1:]", "_____no_output_____" ] ], [ [ "## EXERCISE: Determine the accuracy of your model\n\nUse your deployed `predictor` to generate predicted, class labels for the test data. Compare those to the *true* labels, `test_y`, and calculate the accuracy as a value between 0 and 1.0 that indicates the fraction of test data that your model classified correctly. You may use [sklearn.metrics](https://scikit-learn.org/stable/modules/classes.html#module-sklearn.metrics) for this calculation.\n\n**To pass this project, your model should get at least 90% test accuracy.**", "_____no_output_____" ] ], [ [ "import numpy as np\n# First: generate predicted, class labels\ntest_y_preds = np.squeeze(np.round(predictor.predict(test_x)))\n\n\n\"\"\"\nDON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE\n\"\"\"\n# test that your model generates the correct number of labels\nassert len(test_y_preds)==len(test_y), 'Unexpected number of predictions.'\nprint('Test passed!')", "Test passed!\n" ], [ "# Second: calculate the test accuracy\naccuracy = None\n# calculate true positives, false positives, true negatives, false negatives\ntp = np.logical_and(test_y.values, test_y_preds).sum()\nfp = np.logical_and(1-test_y.values, test_y_preds).sum()\ntn = np.logical_and(1-test_y.values, 1-test_y_preds).sum()\nfn = np.logical_and(test_y.values, 1-test_y_preds).sum()\n \n # calculate binary classification metrics\nrecall = tp / (tp + fn)\nprecision = tp / (tp + fp)\naccuracy = (tp + tn) / (tp + fp + tn + fn)\nprint(f'tp:{tp} fp:{fp} tn:{tn} fn:{fn}')\nprint(accuracy)\n\n\n## print out the array of predicted and true labels, if you want\nprint('\\nPredicted class labels: ')\nprint(test_y_preds)\nprint('\\nTrue class labels: ')\nprint(test_y.values)", "tp:15 fp:1 tn:9 fn:0\n0.96\n\nPredicted class labels: \n[1 1 1 1 1 1 0 0 0 0 0 0 1 1 1 1 1 1 0 1 1 1 1 0 0]\n\nTrue class labels: \n[1 1 1 1 1 1 0 0 0 0 0 0 1 1 1 1 1 1 0 1 0 1 1 0 0]\n" ] ], [ [ "### Question 1: How many false positives and false negatives did your model produce, if any? And why do you think this is?", "_____no_output_____" ], [ "** Answer**: Case 1, when use selected_features ['c_11', 'lcs_word'], and Classfier \"LinearSVC\", we have one false negative. \n Case 2, when use selected_features ['c_11', 'lcs_word'], and Classfier \"LogisticRegression\", we have one false positive.\n Case 3, when use selected_features ['c_1','c_11', 'lcs_word'], and Classfier \"LogisticRegression\", we have one false positive.\n The accuary is always 0.96. Because of the testset is only 25, not big data and not large amount of features, the performance between different classfier is not observed yet. If the noise is higher, maybe the accuray will be less than 0.96. we can't Use SKLearn Classifier like \"LinearSVC\", \"LogisticRegression\" have better accuary compared to Pytorch self-defined binary classifier. \n", "_____no_output_____" ] ], [ [ "#Case 1 Check:\npredict_df = pd.concat([pd.DataFrame(test_x), pd.DataFrame(test_y_preds), pd.DataFrame(test_y)], axis=1)\npredict_df.columns=['c_11', 'lcs_word', 'predicted class','true class']\npredict_df", "_____no_output_____" ], [ "#Case 2 Check:\npredict_df = pd.concat([pd.DataFrame(test_x), pd.DataFrame(test_y_preds), pd.DataFrame(test_y)], axis=1)\npredict_df.columns=['c_11', 'lcs_word', 'predicted class','true class']\npredict_df", "_____no_output_____" ], [ "#Case 3 Check:\npredict_df = pd.concat([pd.DataFrame(test_x), pd.DataFrame(test_y_preds), pd.DataFrame(test_y)], axis=1)\npredict_df.columns=['c_1','c_11', 'lcs_word', 'predicted class','true class']\npredict_df", "_____no_output_____" ] ], [ [ "### Question 2: How did you decide on the type of model to use? ", "_____no_output_____" ], [ "** Answer**: If the problem need only binary output (is 0 or 1)/(is true or false), binary classifier is a good choice. When the noise in the training data is minimized, then we can also use simpler classifier for good performance. Otherwise, we need use more complex classifiers like SVM or Neural Network for massive dataset and huge size of features.\n\n", "_____no_output_____" ] ], [ [ "a=pd.DataFrame(np.array([[0.765306, 0.394366, 0.621711]]),columns=[1, 2, 3])\n\ntest_file.getvalue()", "_____no_output_____" ], [ "import boto3\nimport io\nfrom io import StringIO\ntest_file = io.StringIO()\ncheck_data=test_x.iloc[:2,1:] #data.iloc[:2,1:]\ncheck_data.to_csv(test_file,header = None, index = None)\nruntime = boto3.Session().client('sagemaker-runtime')\nresponse = runtime.invoke_endpoint(EndpointName = predictor.endpoint, # The name of the endpoint we created\n ContentType = 'text/csv', # The data format that is expected\n Body ='0.0,0.7914438502673797,0.8207547169811321\\n0.0,0.0,0\\n' ) #test_file.getvalue() )\na=response['Body'].read().decode('utf-8')\neval(a)[0]", "_____no_output_____" ] ], [ [ "----\n## EXERCISE: Clean up Resources\n\nAfter you're done evaluating your model, **delete your model endpoint**. You can do this with a call to `.delete_endpoint()`. You need to show, in this notebook, that the endpoint was deleted. Any other resources, you may delete from the AWS console, and you will find more instructions on cleaning up all your resources, below.", "_____no_output_____" ] ], [ [ "# uncomment and fill in the line below!\n# <name_of_deployed_predictor>.delete_endpoint()\ndef delete_endpoint(predictor):\n try:\n boto3.client('sagemaker').delete_endpoint(EndpointName=predictor.endpoint)\n print('Deleted {}'.format(predictor.endpoint))\n except:\n print('Already deleted: {}'.format(predictor.endpoint))\ndelete_endpoint(predictor)", "Deleted sagemaker-scikit-learn-2020-10-23-18-36-19-083\n" ] ], [ [ "### Deleting S3 bucket\n\nWhen you are *completely* done with training and testing models, you can also delete your entire S3 bucket. If you do this before you are done training your model, you'll have to recreate your S3 bucket and upload your training data again.", "_____no_output_____" ] ], [ [ "# deleting bucket, uncomment lines below\n\nbucket_to_delete = boto3.resource('s3').Bucket(bucket)\nbucket_to_delete.objects.all().delete()", "_____no_output_____" ] ], [ [ "### Deleting all your models and instances\n\nWhen you are _completely_ done with this project and do **not** ever want to revisit this notebook, you can choose to delete all of your SageMaker notebook instances and models by following [these instructions](https://docs.aws.amazon.com/sagemaker/latest/dg/ex1-cleanup.html). Before you delete this notebook instance, I recommend at least downloading a copy and saving it, locally.", "_____no_output_____" ], [ "---\n## Further Directions\n\nThere are many ways to improve or add on to this project to expand your learning or make this more of a unique project for you. A few ideas are listed below:\n* Train a classifier to predict the *category* (1-3) of plagiarism and not just plagiarized (1) or not (0).\n* Utilize a different and larger dataset to see if this model can be extended to other types of plagiarism.\n* Use language or character-level analysis to find different (and more) similarity features.\n* Write a complete pipeline function that accepts a source text and submitted text file, and classifies the submitted text as plagiarized or not.\n* Use API Gateway and a lambda function to deploy your model to a web application.\n\nThese are all just options for extending your work. If you've completed all the exercises in this notebook, you've completed a real-world application, and can proceed to submit your project. Great job!", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown" ]

[ [ "markdown", "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown", "markdown" ], [ "code", "code", "code" ], [ "markdown", "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "## Exercise", "_____no_output_____" ], [ "check if x >10, return T/F", "_____no_output_____" ] ], [ [ "x = 12\nif x>10:\n print('True')\nelse:\n print('F')", "True\n" ] ], [ [ "define a function called square that returns the squared value of input x", "_____no_output_____" ] ], [ [ "def square(x):\n return(x**2)", "_____no_output_____" ], [ "square(12)", "_____no_output_____" ] ], [ [ "## Library Importing", "_____no_output_____" ] ], [ [ "##First way\nimport numpy", "_____no_output_____" ], [ "numpy.absolute(-7)", "_____no_output_____" ], [ "numpy.sqrt(8)", "_____no_output_____" ], [ "##Second way", "_____no_output_____" ], [ "from numpy import sqrt", "_____no_output_____" ], [ "sqrt(8)", "_____no_output_____" ], [ "absolute(-7)", "_____no_output_____" ], [ "import numpy as np", "_____no_output_____" ], [ "np.sqrt(8)", "_____no_output_____" ], [ "import random", "_____no_output_____" ], [ "random.randint(a= 0 ,b= 10 )", "_____no_output_____" ], [ "from random import randint", "_____no_output_____" ], [ "randint(0,10)", "_____no_output_____" ] ], [ [ "## Turtle", "_____no_output_____" ] ], [ [ "import turtle as t\nimport numpy as np\nimport random", "_____no_output_____" ], [ "##set up screen\nscreen = t.Screen()\n## set up background color\nscreen.bgcolor('lightgreen')\n## set screen title\nscreen.title(\"Rick's Program\")\n\n##set up a turtle\nrick = t.Turtle()\n\n## move forward\nrick.forward(100)\n# rick.fd(100)\n\n## move backward\n# rick.backward(100)\n# rick.bk(100)\n# rick.back(100)\n\n## move to the left\nrick.left(90)\nrick.forward(100)\n\n## move to the right\nrick.right(90)\nrick.forward(100)\n\n\nscreen.exitonclick()", "_____no_output_____" ] ], [ [ "Question: how do we draw a square?", "_____no_output_____" ] ], [ [ "for i in range(1,5):\n t.fd(90)\n t.left(90)\nt.exitonclick()", "_____no_output_____" ] ], [ [ "Question: how do we draw a circle?", "_____no_output_____" ] ], [ [ "t.circle(100)\nt.exitonclick()", "_____no_output_____" ] ], [ [ "Question: How to we draw this graph using turtle?", "_____no_output_____" ], [ "<img src=\"../tiltedsquares.png\">", "_____no_output_____" ] ], [ [ "##First Square\nt.left(20)\nfor i in range(1,5):\n t.fd(90)\n t.left(90)\n##Second Square\nt.left(20)\nfor i in range(1,5):\n t.fd(90)\n t.left(90)\n##Third Square\nt.left(20)\nfor i in range(1,5):\n t.fd(90)\n t.left(90)\n \nt.exitonclick()", "_____no_output_____" ], [ "for i in range(1,4):\n t.left(20)\n for i in range(1,5):\n t.fd(90)\n t.left(90)\nt.exitonclick()", "_____no_output_____" ], [ "## Adding motions into the screen\n##set up screen\nscreen = t.Screen()\n## set up background color\nscreen.bgcolor('lightgreen')\n## set screen title\nscreen.title(\"Rick's Program\")\n#set up a turtle\nrick = t.Turtle()\n#Change turtle shape\nrick.shape('turtle')\n## Change pen size\nrick.pensize(5)\nrick.forward(100)\n## Change pen color\nrick.pencolor('blue')\nrick.left(90)\nrick.forward(100)\n\n## change turtle color\nrick.color('red')\nrick.left(90)\nrick.forward(100)\n\n## Multiple Turtle\nrick2 = t.Turtle()\n# ##penup\n# rick2.penup()\n# rick2.bk(100)\n# ##pendown\n# rick2.pendown()\n# rick2.left(90)\n# rick2.fd(100)\n\n## Resize a turtle\n# rick.shapesize(2,2,0) #width, length, outline\n# rick.forward(100)\n# rick.shapesize(0.5,0.5,0)\n# rick.forward(100)\n\nrick2.penup()\nrick2.setposition(-100,-100)\n\n## hide turtle\nrick2.pendown()\nrick2.hideturtle()\nrick2.forward(100)\n\n## show turtle\nrick2.showturtle()\nrick2.left(90)\nrick2.forward(100)\n\n\nscreen.exitonclick()", "_____no_output_____" ] ], [ [ "### Game", "_____no_output_____" ] ], [ [ "screen = t.Screen()\n## set up background color\nscreen.bgcolor('lightgreen')\n## set screen title\nscreen.title(\"Rick's Program\")\n\n##Draw a Border\nborder = t.Turtle()\nborder.color('white')\nborder.penup()\nborder.setposition(-300,-300)\nborder.pendown()\nborder.pensize(3)\nfor side in range(4):\n border.fd(600)\n border.lt(90)\nborder.hideturtle()\n\n\n#set up player\nplayer = t.Turtle()\nplayer.color('blue')\nplayer.shape('triangle')\nplayer.penup()\n\n## set up Goal\ngoal = t.Turtle()\ngoal.color('red')\ngoal.shape('circle')\ngoal.penup()\ngoal.setpos(-100,100)\n\n\n\nspeed = 1\n## Define Function\ndef turnleft():\n player.lt(30)\n\ndef turnright():\n player.rt(30)\ndef speed5():\n global speed\n speed = 5\ndef speed1():\n global speed\n speed = 1\n\n##keyboard Binding\nt.listen()\nt.onkey(turnleft, 'Left')\nt.onkey(turnright, 'Right')\nt.onkeypress(speed5, 'Up')\nt.onkeyrelease(speed1, 'Up')\n\n\nwhile True:\n player.forward(speed)\n d = np.sqrt((player.xcor()-goal.xcor())**2\n +(player.ycor()-goal.ycor())**2)\n if d<20:\n goal.setpos(random.randint(-300,300)\n ,random.randint(-300,300))\n\n\n\n\n\n\n", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "raw", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "raw" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "import boto3\nimport cv2\nimport io\nimport os\nimport s3fs\nimport torch\n\nimport matplotlib.pyplot as plt\n%matplotlib inline\nimport numpy as np\nimport seaborn as sns\nimport torch.nn as nn\n\nfrom collections import OrderedDict\nfrom PIL import Image\nfrom torch import optim\nfrom torch.utils.data import Dataset, DataLoader \nfrom torchvision import datasets, models, transforms", "_____no_output_____" ] ], [ [ "#### s3 configuration", "_____no_output_____" ] ], [ [ "\ns3 = boto3.resource(\"s3\",\n endpoint_url = \"http://192.168.0.29\",\n aws_access_key_id=\"AKIAPo19vPR_TJaeVgleCiOSUw\",\n aws_secret_access_key=\"7cSWM1KCXvRpK4ICeDEAfuicEm+QQeuhqOi7cejZ\",\n region_name = 'eu-central-1',\n )\n\nkwargs = {'endpoint_url':\"http://192.168.0.29\",\n }\nclient = s3fs.S3FileSystem(key=\"AKIAPo19vPR_TJaeVgleCiOSUw\", \n secret=\"7cSWM1KCXvRpK4ICeDEAfuicEm+QQeuhqOi7cejZ\",\n use_ssl=False,\n \n client_kwargs=kwargs)\n", "_____no_output_____" ], [ "my_bucket = s3.Bucket(\"sample-dataset\")", "_____no_output_____" ], [ "map_labels = {\"Apple___Apple_scab\":0, \"Apple___Black_rot\":1, \n \"Apple___Cedar_apple_rust\":2, \"Apple___healthy\":3, \"Background_without_leaves\":4}", "_____no_output_____" ] ], [ [ "#### Create Custom Dataset to Load data\n- Pytorch do not have any existing Dataset Loader classes that fetch data from s3. Therefore we need to create a custom Dataset Loader that will fetch the data from Cortx-s3.", "_____no_output_____" ] ], [ [ "\nclass ImageDataset(Dataset):\n \n def __init__(self, path=\"s3://sample-dataset/sample_data/\", transform=None):\n self.path = path\n self.classes = [folder[\"name\"] for folder in client.listdir(path)][2:]\n self.files = []\n for directory in self.classes:\n self.files += [file for file in client.ls(directory)][1:]\n\n self.transform = transform\n\n def __len__(self):\n return len(self.files)\n \n\n def __getitem__(self, idx):\n img_name = self.files[idx]\n label = img_name.split(\"/\")[-2]\n label = map_labels[label]\n \n key = img_name.split(\"/\")\n key = \"/\".join(key[1:])\n \n img_name = my_bucket.Object(key).get().get('Body').read()\n \n image = cv2.imdecode(np.asarray(bytearray(img_name)), cv2.COLOR_BGR2RGB)\n image = Image.fromarray(image)\n if self.transform:\n image = self.transform(image)\n label = torch.tensor(label).long()\n return image, label", "_____no_output_____" ], [ "data_dir = client.glob(\"s3://sample-dataset/sample_data\")[0]\n\ntrain_transforms = transforms.Compose([transforms.RandomRotation(30),\n transforms.RandomResizedCrop(224),\n transforms.RandomHorizontalFlip(),\n transforms.ToTensor(),\n transforms.Normalize([0.485, 0.456, 0.406],\n [0.229, 0.224, 0.225])])\n\nval_transforms = transforms.Compose([transforms.Resize(255),\n transforms.CenterCrop(224),\n transforms.ToTensor(),\n transforms.Normalize([0.485, 0.456, 0.406],\n [0.229, 0.224, 0.225])])\n\ntrain_dataset = ImageDataset(path=\"s3://sample-dataset/sample_data/train\", transform=train_transforms)\n\n \nvalid_dataset = ImageDataset(path=\"s3://sample-dataset/sample_data/val\",\n transform=val_transforms)\n \ntrain_loader = DataLoader(train_dataset, \n batch_size=16, \n shuffle=True, \n num_workers=0)\nval_loader = DataLoader(valid_dataset, \n batch_size=16, \n shuffle=False, \n num_workers=0)", "_____no_output_____" ], [ "dataiter = iter(train_loader)\nimages, labels = dataiter.next()\nprint(type(images))\nprint(images.shape)\nprint(labels.shape)", "<class 'torch.Tensor'>\ntorch.Size([16, 3, 224, 224])\ntorch.Size([16])\n" ], [ "def imshow(image, ax=None, normalize=True):\n if ax is None:\n fig, ax = plt.subplots()\n image = image.numpy().transpose((1, 2, 0))\n\n if normalize:\n # if the data loader has transform.normalize\n # undo preprocessing\n mean = np.array([0.485, 0.456, 0.406])\n std = np.array([0.229, 0.224, 0.225])\n image = std * image + mean\n image = np.clip(image, 0, 1)\n\n ax.imshow(image)\n ax.spines['top'].set_visible(False)\n ax.spines['right'].set_visible(False)\n ax.spines['left'].set_visible(False)\n ax.spines['bottom'].set_visible(False)\n ax.tick_params(axis='both', length=0)\n ax.set_xticklabels('')\n ax.set_yticklabels('')\n \n return ax", "_____no_output_____" ], [ "imshow(images[1]);\nimshow(images[8]);\nimshow(images[12]);", "_____no_output_____" ], [ "model = models.densenet201(pretrained=True)", "_____no_output_____" ], [ "for param in model.parameters():\n param.required_grad = False", "_____no_output_____" ], [ "# change the classifier\nfrom collections import OrderedDict\n\nclassifier = nn.Sequential(OrderedDict([\n ('fc1', nn.Linear(1920, 500)),\n ('relu1', nn.ReLU()),\n ('dropout1', nn.Dropout(p=0.2)),\n ('fc2', nn.Linear(500, 256)),\n ('relu2', nn.ReLU()),\n ('dropout2', nn.Dropout(p=0.2)),\n ('fc3', nn.Linear(256, 5)),\n ('output', nn.LogSoftmax(dim=1))\n \n]))\nmodel.classifier = classifier", "_____no_output_____" ], [ "# Train either on GPU or CPU\ndevice = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")", "_____no_output_____" ], [ "\ncriterion = nn.NLLLoss()\n\noptimizer = optim.SGD(model.classifier.parameters(), lr = 0.01, momentum=0.9)\n\nmodel.to(device)", "_____no_output_____" ], [ "epochs = 1\n\nfor epoch in range(epochs):\n running_loss = 0\n for images, labels in train_loader:\n images, labels = images.to(device), labels.to(device)\n \n optimizer.zero_grad()\n \n outputs = model(images)\n loss = criterion(outputs, labels)\n loss.backward()\n optimizer.step()\n \n running_loss += loss.item()\n \n else:\n validation_loss = 0\n accuracy = 0\n \n with torch.no_grad():\n model.eval()\n for images, labels in val_loader:\n images, labels = images.to(device), labels.to(device)\n outputs = model(images)\n loss = criterion(outputs, labels)\n validation_loss += loss.item()\n \n ps = torch.exp(outputs)\n top_p, top_class = ps.topk(1, dim=1)\n equals = top_class == labels.view(*top_class.shape)\n accuracy += torch.mean(equals.type(torch.FloatTensor))\n \n model.train()\n\n print(\"Epoch: {}/{}.. \".format(epoch+1, epochs),\n \"Training Loss: {:.3f}.. \".format(running_loss/len(train_loader)),\n \"Valid Loss: {:.3f}.. \".format(validation_loss/len(val_loader)),\n \"Valid Accuracy: {:.3f}\".format(accuracy/len(val_loader)))", "Epoch: 1/1.. Training Loss: 0.717.. Valid Loss: 0.176.. Valid Accuracy: 0.946\n" ], [ "#map classes to indexes\nmodel.class_to_idx = map_labels", "_____no_output_____" ], [ "# save model\n\ns3_client = boto3.client(\"s3\",\n endpoint_url = \"http://192.168.0.29\",\n aws_access_key_id=\"AKIAPo19vPR_TJaeVgleCiOSUw\",\n aws_secret_access_key=\"7cSWM1KCXvRpK4ICeDEAfuicEm+QQeuhqOi7cejZ\",\n )\n \nbuffer = io.BytesIO()\ntorch.save({\"state_dict\":model.state_dict(),\n \"class_to_idx\":model.class_to_idx}, buffer)\ns3_client.put_object(Bucket=\"saved-models\", Key='classifier.pth', Body=buffer.getvalue())", "_____no_output_____" ] ] ]

[ "code", "markdown", "code", "markdown", "code" ]

[ [ "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "<a href=\"https://colab.research.google.com/github/jzzy-jeff/github-slideshow/blob/master/FlappyBird.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>", "_____no_output_____" ] ], [ [ "pip install pygame", "Requirement already satisfied: pygame in /usr/local/lib/python3.6/dist-packages (2.0.1)\n" ], [ "pip install neat-python\r\n", "Requirement already satisfied: neat-python in /usr/local/lib/python3.6/dist-packages (0.92)\n" ], [ "import pygame\r\nimport neat\r\nimport time\r\nimport os\r\nimport random", "_____no_output_____" ], [ "#Window and object images\r\nWIN_WIDTH = 600\r\nWIN_HEIGHT = 800\r\n\r\nBIRD_IMGS = [pygame.transorm.scale2x(pygame.image.load(os.path.join(\"imgs\", \"bird1.png\"))), [pygame.transorm.scale2x(pygame.image.load(os.path.join(\"imgs\", \"bird2.png\"))), [pygame.transorm.scale2x(pygame.image.load(os.path.join(\"imgs\", \"bird3.png\")))]\r\nPIPE_IMG = pygame.transform.scale2x(pygame.image.load(os.path.join(\"imgs\", \"pipe.png\")))\r\nBASE_IMG = pygame.transform.scale2x(pygame.image.load(os.path.join(\"imgs\", \"base.png\")))\r\nBG_IMG = pygame.transform.scale2x(pygame.image.load(os.path.join(\"imgs\", \"bg.png\")))\r\n\r\n\r\n ", "_____no_output_____" ], [ "#Object Classes\r\nclass Bird:\r\n IMGS = BIRD_IMGS\r\n MAX_ROTATION = 25\r\n ROT_VEL = 20\r\n ANIMATION_TIME = 5\r\n\r\n def __init__(self, x, y):\r\n self.x = x\r\n self.y = y\r\n self.tilt = 0\r\n self.tick_count = 0\r\n self.vel = 0\r\n self,height = self.y\r\n self.img_count = 0\r\n self.img = self.IMGS[0]\r\n\r\n def jump(self):\r\n self.vel = -10.5\r\n self.tick_count = 0\r\n self.height = self.y\r\n\r\n def move(self):\r\n self.tick_count += 1\r\n \r\n d = self.vel*self.tick_count + 1.5*self.tick_count**2\r\n \r\n if d >= 16:\r\n d = 16\r\n\r\n if d < 0:\r\n d-=2.4\r\n\r\n self.y = self.y + de\r\n\r\n if d < 0 or self.y < self.height + 50:\r\n if self.tilt < self.MAX_ROTATION:\r\n self.tilt = MAX_ROTATION\r\n else:\r\n if self.tilt> -90:\r\n self.tilt -= self.ROT_VEL\r\n\r\n def draw(self, win):\r\n self.img_count +=1\r\n\r\n if self.img_count < self.ANIMATION_TIME:\r\n self.img = self.IMGS[0]\r\n elif self.img_count < self.ANIMATION_TIME*2:\r\n self.img = self.IMGS[1]\r\n elif self.img_count < self.ANIMATION_TIME*3:\r\n self.img = self.IMGS[2]\r\n elif self.img_count < self.ANIMATION_TIME*4:\r\n self.img = self.IMGS[1]\r\n elif self.img_count < self.ANIMATION_TIME*4+1:\r\n self.img = self.IMGS[0]\r\n self.img_count = 0\r\n\r\n if self.tilt <= -80:\r\n self.img = self.IMGS[1]\r\n self.img_count = self.ANIMATION_TIME*2\r\n\r\n rotated_image = pygame.transorm.rotate(self.img, self.tilt)\r\n new_rect = rotated_image.get_rect(center=self.img.get_rect(topleft = (self.x, self.y)).center)\r\n win.blit(rotated_image, new_rect.topleft)\r\n\r\n def get_mask(self):\r\n return pygame.mask.from_surface(self.img)\r\n \r\ndef draw_window(win, bird):\r\n win.blit(BG_IMG, (0,0))\r\n bird.draw(win)\r\n pygame.display.update()\r\n\r\ndef main():\r\n bird = Bird(200,200)\r\n win = pygame.display.set_mode((WIN_WIDTH, WIN_HEIGHT))\r\n run = True\r\n while run:\r\n for event in pygame.event.get():\r\n if event.type == pygame.QUIT:\r\n run == False\r\n\r\n draw_window(win, bird)\r\n\r\n pygame.quit()\r\n quit()\r\n\r\n main()", "_____no_output_____" ] ] ]

[ "markdown", "code" ]

[ [ "markdown" ], [ "code", "code", "code", "code", "code" ] ]

[ "CC-BY-3.0" ]

[ "CC-BY-3.0" ]

[ "CC-BY-3.0" ]

[ [ [ "empty" ] ] ]

[ "empty" ]

[ [ "empty" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "# default_exp core", "_____no_output_____" ], [ "from nbdev.showdoc import *", "_____no_output_____" ], [ "# export\ndef power(a,b):\n # a^b\n if b==0:\n return a\n elif b%2==1:\n return power(a,b-1)*a\n else:\n return power(a,b/2)*power(a,b/2)", "_____no_output_____" ], [ "from nbdev.export import *\nnotebook2script()", "Converted 00_core.ipynb.\nConverted 99_index.ipynb.\n" ] ] ]

[ "code" ]

[ [ "code", "code", "code", "code" ] ]

[ "CC-BY-2.0" ]

[ "CC-BY-2.0" ]

[ "CC-BY-2.0" ]

[ [ [ "# Prediction API\n\nProgrammatically use OpenPifPaf to run multi-person pose estimation on an image.\nThe API is for more advanced use cases. Please read {doc}`predict_cli` as well.", "_____no_output_____" ] ], [ [ "import io\nimport numpy as np\nimport openpifpaf\nimport PIL\nimport requests\nimport torch\n\n%matplotlib inline\nopenpifpaf.show.Canvas.show = True\n\ndevice = torch.device('cpu')\n# device = torch.device('cuda') # if cuda is available\n\nprint(openpifpaf.__version__)\nprint(torch.__version__)", "_____no_output_____" ] ], [ [ "## Load an Example Image\n\nImage credit: \"[Learning to surf](https://www.flickr.com/photos/fotologic/6038911779/in/photostream/)\" by fotologic which is licensed under [CC-BY-2.0].\n\n[CC-BY-2.0]: https://creativecommons.org/licenses/by/2.0/", "_____no_output_____" ] ], [ [ "image_response = requests.get('https://raw.githubusercontent.com/vita-epfl/openpifpaf/master/docs/coco/000000081988.jpg')\npil_im = PIL.Image.open(io.BytesIO(image_response.content)).convert('RGB')\nim = np.asarray(pil_im)\n\nwith openpifpaf.show.image_canvas(im) as ax:\n pass", "_____no_output_____" ] ], [ [ "## Load a Trained Neural Network", "_____no_output_____" ] ], [ [ "net_cpu, _ = openpifpaf.network.Factory(checkpoint='shufflenetv2k16', download_progress=False).factory()\nnet = net_cpu.to(device)\n\nopenpifpaf.decoder.utils.CifSeeds.threshold = 0.5\nopenpifpaf.decoder.utils.nms.Keypoints.keypoint_threshold = 0.2\nopenpifpaf.decoder.utils.nms.Keypoints.instance_threshold = 0.2\nprocessor = openpifpaf.decoder.factory([hn.meta for hn in net_cpu.head_nets])", "_____no_output_____" ] ], [ [ "## Preprocessing, Dataset\n\nSpecify the image preprocossing. Beyond the default transforms, we also use `CenterPadTight(16)` which adds padding to the image such that both the height and width are multiples of 16 plus 1. With this padding, the feature map covers the entire image. Without it, there would be a gap on the right and bottom of the image that the feature map does not cover.", "_____no_output_____" ] ], [ [ "preprocess = openpifpaf.transforms.Compose([\n openpifpaf.transforms.NormalizeAnnotations(),\n openpifpaf.transforms.CenterPadTight(16),\n openpifpaf.transforms.EVAL_TRANSFORM,\n])\ndata = openpifpaf.datasets.PilImageList([pil_im], preprocess=preprocess)", "_____no_output_____" ] ], [ [ "## Dataloader, Visualizer", "_____no_output_____" ] ], [ [ "loader = torch.utils.data.DataLoader(\n data, batch_size=1, pin_memory=True, \n collate_fn=openpifpaf.datasets.collate_images_anns_meta)\n\nannotation_painter = openpifpaf.show.AnnotationPainter()", "_____no_output_____" ] ], [ [ "## Prediction", "_____no_output_____" ] ], [ [ "for images_batch, _, __ in loader:\n predictions = processor.batch(net, images_batch, device=device)[0]\n with openpifpaf.show.image_canvas(im) as ax:\n annotation_painter.annotations(ax, predictions)", "_____no_output_____" ] ], [ [ "Each prediction in the `predictions` list above is of type `Annotation`. You can access the joint coordinates in the `data` attribute. It is a numpy array that contains the $x$ and $y$ coordinates and the confidence for every joint:", "_____no_output_____" ] ], [ [ "predictions[0].data", "_____no_output_____" ] ], [ [ "## Fields\n\nBelow are visualizations of the fields.\nWhen using the API here, the visualization types are individually enabled.\nThen, the index for every field to visualize must be specified. In the example below, the fifth CIF (left shoulder) and the fifth CAF (left shoulder to left hip) are activated.\n\nThese plots are also accessible from the command line: use `--debug-indices cif:5 caf:5` to select which joints and connections to visualize.", "_____no_output_____" ] ], [ [ "openpifpaf.visualizer.Base.set_all_indices(['cif,caf:5:confidence'])\n\nfor images_batch, _, __ in loader:\n predictions = processor.batch(net, images_batch, device=device)[0]", "_____no_output_____" ], [ "openpifpaf.visualizer.Base.set_all_indices(['cif,caf:5:regression'])\n\nfor images_batch, _, __ in loader:\n predictions = processor.batch(net, images_batch, device=device)[0]", "_____no_output_____" ] ], [ [ "From the CIF field, a high resolution accumulation (in the code it's called `CifHr`) is generated.\nThis is also the basis for the seeds. Both are shown below.", "_____no_output_____" ] ], [ [ "openpifpaf.visualizer.Base.set_all_indices(['cif:5:hr', 'seeds'])\n\nfor images_batch, _, __ in loader:\n predictions = processor.batch(net, images_batch, device=device)[0]", "_____no_output_____" ] ], [ [ "Starting from a seed, the poses are constructed. At every joint position, an occupancy map marks whether a previous pose was already constructed here. This reduces the number of poses that are constructed from multiple seeds for the same person. The final occupancy map is below:", "_____no_output_____" ] ], [ [ "openpifpaf.visualizer.Base.set_all_indices(['occupancy:5'])\n\nfor images_batch, _, __ in loader:\n predictions = processor.batch(net, images_batch, device=device)[0]", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Ubuntu 16装机过程\n## 解决网络问题\n### DNS Server\n默认的DNS Server可能是无效的。 如果不能上网， 试试在network connection的IPV4 setting里面添加有效的DNS Server，如114.114.114.114, 并且把Method设置成Automatic(DHCP) address only\n\n\n\n配置后需要运行'''sudo systemctl restart network-manager.service'''让设置生效\n\n### 翻墙\nshadowsocks安装和配置看[这里](https://github.com/shadowsocks/shadowsocks/wiki/Shadowsocks-%E4%BD%BF%E7%94%A8%E8%AF%B4%E6%98%8E\n)\n\n注意安装shadowsocks的时候，要用sudo安装\n```\nsudo to install shadowsocks\n```\n\n为了开机的时候能够自动启动ss，可以将\n\n```sslocal -c /etc/shadowsocks.json -d start```\n\n这一命令添加至\n\n```/etc/rc.local```\n\n这样每次开机就能自动启动sslocal。\n\n启动了shadowsocks， 还得配置firefox， [见这里](https://aiguge.xyz/firefox-shadowsocks-foxyproxy-standard/)\n\ngfwlist可以使用```https://aiguge.xyz/firefox-shadowsocks-foxyproxy-standard/```\n\n\n## 升级ubuntu到较新的状态\n```sudo apt-get dist-upgrade```\n\n## python必备库安装\nubuntu 16自带了python3.5， 可以不需要自己安装\n\n### 安装pip\n```apt-get install python-pip```\n\n### virtualenv\n由于Ubuntu默认使用的python是2.7版本的,为了避免不必要的麻烦，可以使用virtualenv创造一个python3.5的环境。先安装virtualenv\n```sudo apt-get install virtualenv```\n\n然后创建虚拟环境，我将虚拟环境的文件夹放在了home下， 如果想放在别的地方请自己调整命令的参数。\n```\nvirtualenv -p /usr/bin/python3.5 ~/venv\n```\n\n然后激活创建的virtualenv,新建一个teminal后，输入。\n```\nsource ~/venv/bin/activate\n```\n可以看到terminal前面出现了一个(venv)字样，提示已经进入了虚拟环境。\n\n这个命令太长，每次输入很麻烦。我们可以为这个命令创建一个简化的命令'venv'。在```~/.bash_aliases```文件中输入\n```\nalias venv='source ~/venv/bin/activate'\n```\n运行\n```\nsource ~/bashrc\n```\n让刚才的设置生效。之后进入terminal后只要输入```venv```就可以进入我们的虚拟环境。\n\n**之后安装python库的时候都是先进入```venv```虚拟环境后安装的，后面不在重复说明。\n\n\n\n\n\n### cx_Oracle\t\n```sudo apt-get install unzip python3.5-dev libaio-dev```\n[具体安装方式见](https://gist.github.com/kimus/10012910)\n\ninstantclient-basic-linux.x64-12.1.0.2.0.zip[下载](http://www.oracle.com/technetwork/topics/linuxx86-64soft-092277.html)\n\ninstantclient-sdk-linux.x64-12.1.0.2.0.zip[下载](http://www.oracle.com/technetwork/topics/linuxx86-64soft-092277.html)\n\n### psycopg2\n```sudo apt-get install python-psycopg2```\n\n```pip install psycopg2```\n\n\n## Postgresql-9.5安装\n[见这里](https://www.postgresql.org/download/linux/ubuntu/)\n\n### 添加数据库用户\npostgreSQL会自动给ubuntu和数据库本身创建一个名为postgres的用户名。\n\n首先修改postgres用户(ubuntu)的密码\n```\nsudo passwd postgres\n```\n\n切换到postgres用户(ubuntu)，登陆数据库\n```\nsudo su postgres\npsql\n```\n然后建立一些普通的数据库用户，习惯上可以和ubunut的用户名保持一直。 注意这个账号如果要使用pl/python， 就必须得提供超级用户权限。\n\n```\ncreate user datascience superuser password 'xxxxxxx';\n```\n\n### 建立数据库\n可以根据需要为该账号建立数据库。\n```\ncreate database etl;\nalter database etl owner to datascience;\n\\q\n```\n\n之后如果要在非postgres用户状态下进入数据库\n\n```\npsql -U [dbuser] -d [db] -h 127.0.0.1 -p 5432\n```\n\n### plpython的安装\n```\nsudo apt-get install postgresql-plpython3\n```\n登陆想使用plpython3数据库后，运行\n```\nCREATE EXTENSION plpython3u;\n```\n\n如果要让plpython能够使用自己python库， 需要添加PYTHONPATH信息至etc/postgresql/9.5/main/environment, 如\n```\nPYTHONPATH='/home/cjj/jfds\n```\n\n## Wine\n如果要在ubuntu运行一些window程序的话，需要这个\n```\nsudo apt-get update\nsudo apt-get install wine\n```\n注意安装wine的界面要按tab才能切换到OK按钮\n\n\n## Git安装\n```sudo apt-get install git```\n\n\n## 安装常用软件\n### 搜狗拼音\n先用\n```\nsudo apt-get install -f\n```\n修复一些包依赖的问题。\n\n[download](http://pinyin.sogou.com/linux/help.php)很不错的中文输入法, 直接下载双击deb会失败，需要用```sudo dpkg -i xxx.deb```命令安装\n\n如果出现了下面的提示\n\nNo such key 'Gtk/IMModule' in schema 'org.gnome.settings-daemon.plugins.xsettings' as specified in override file '/usr/share/glib-2.0/schemas/50_sogoupinyin.gschema.override'; ignoring override for this key.\n\n并不用在意，这个不会影响搜狗输入法的使用。\n\n\n\n### download them all\n这个是firefox的插件， 可以有更好的下载体验\n\n### okular\n一个带笔记标注功能的pdf浏览器\n\n### startuml2\n很好的uml建模工具\n\n[下载依赖库](https://launchpad.net/ubuntu/+archive/primary/+files/libgcrypt11_1.5.3-2ubuntu4.2_amd64.deb\n\n```\nsudo dpkg -i libgcrypt11_1.5.3-2ubuntu4.2_amd64.deb\n\n```\n\n```\nsudo apt-get libpango1.0-0 libpangox-1.0-0\n```\n\n[官网](www.staruml.io)下载deb安装包安装\n```\nsudo dpkg -i StarUML-v2.7.0-64-bit.deb", "_____no_output_____" ] ] ]

[ "markdown" ]

[ [ "markdown" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "# 命令式和符号式混合编程\n\n本书到目前为止一直都在使用命令式编程，它使用编程语句改变程序状态。考虑下面这段简单的命令式编程代码。", "_____no_output_____" ] ], [ [ "def add(a, b):\n return a + b\n\ndef fancy_func(a, b, c, d):\n e = add(a, b)\n f = add(c, d)\n g = add(e, f)\n return g\n\nfancy_func(1, 2, 3, 4)", "_____no_output_____" ] ], [ [ "和我们预期的一样，在运行语句`e = add(a, b)`时，Python会做加法运算并将结果存储在变量`e`，从而令程序的状态发生了改变。类似地，后面的两个语句`f = add(c, d)`和`g = add(e, f)`会依次做加法运算并存储变量。\n\n虽然使用命令式编程很方便，但它的运行可能会慢。一方面，即使`fancy_func`函数中的`add`是被重复调用的函数，Python也会逐一执行这三个函数调用语句。另一方面，我们需要保存变量`e`和`f`的值直到`fancy_func`中所有语句执行结束。这是因为在执行`e = add(a, b)`和`f = add(c, d)`这两个语句之后我们并不知道变量`e`和`f`是否会被程序的其他部分使用。\n\n与命令式编程不同，符号式编程通常在计算流程完全定义好后才被执行。多个深度学习框架，例如Theano和TensorFlow，都使用了符号式编程。通常，符号式编程的程序需要下面三个步骤：\n\n1. 定义计算流程；\n2. 把计算流程编译成可执行的程序；\n3. 给定输入，调用编译好的程序执行。\n\n下面我们用符号式编程重新实现本节开头给出的命令式编程代码。", "_____no_output_____" ] ], [ [ "def add_str():\n return '''\ndef add(a, b):\n return a + b\n'''\n\ndef fancy_func_str():\n return '''\ndef fancy_func(a, b, c, d):\n e = add(a, b)\n f = add(c, d)\n g = add(e, f)\n return g\n'''\n\ndef evoke_str():\n return add_str() + fancy_func_str() + '''\nprint(fancy_func(1, 2, 3, 4))\n'''\n\nprog = evoke_str()\nprint(prog)\ny = compile(prog, '', 'exec')\nexec(y)", "\ndef add(a, b):\n return a + b\n\ndef fancy_func(a, b, c, d):\n e = add(a, b)\n f = add(c, d)\n g = add(e, f)\n return g\n\nprint(fancy_func(1, 2, 3, 4))\n\n10\n" ] ], [ [ "以上定义的三个函数都仅以字符串的形式返回计算流程。最后，我们通过`compile`函数编译完整的计算流程并运行。由于在编译时系统能够完整地看到整个程序，因此有更多空间优化计算。例如，编译的时候可以将程序改写成`print((1 + 2) + (3 + 4))`，甚至直接改写成`print(10)`。这样不仅减少了函数调用，还节省了内存。\n\n对比这两种编程方式，我们可以看到\n\n* 命令式编程更方便。当我们在Python里使用命令式编程时，大部分代码编写起来都很直观。同时，命令式编程更容易排错。这是因为我们可以很方便地获取并打印所有的中间变量值，或者使用Python的排错工具。\n\n* 符号式编程更高效并更容易移植。一方面，在编译的时候系统容易做更多优化；另一方面，符号式编程可以将程序变成一个与Python无关的格式，从而可以使程序在非Python环境下运行，以避开Python解释器的性能问题。\n\n\n## 混合式编程取两者之长\n\n大部分的深度学习框架在命令式编程和符号式编程之间二选一。例如Theano和受其启发的后来者TensorFlow使用了符号式编程；Chainer和它的追随者PyTorch使用了命令式编程。开发人员在设计Gluon时思考了这个问题：有没有可能既得到命令式编程的好处，又享受符号式编程的优势？开发者们认为，用户应该用纯命令式编程进行开发和调试；当需要产品级别的计算性能和部署时，用户可以将大部分程序转换成符号式来运行。Gluon通过提供混合式编程做到了这一点。\n\n在混合式编程中，我们可以通过使用HybridBlock类或者HybridSequential类构建模型。默认情况下，它们和Block或者Sequential类一样依据命令式编程的方式执行。当我们调用`hybridize`函数后，Gluon会转换成依据符号式编程的方式执行。事实上，绝大多数模型都可以享受这样的混合式编程的执行方式。\n\n本节将通过实验展示混合式编程的魅力。\n\n## 使用HybridSequential类构造模型\n\n我们之前学习了如何使用Sequential类来串联多个层。为了使用混合式编程，下面我们将Sequential类替换成HybridSequential类。", "_____no_output_____" ] ], [ [ "from mxnet import nd, sym\nfrom mxnet.gluon import nn\nimport time\n\ndef get_net():\n net = nn.HybridSequential() # 这里创建HybridSequential实例\n net.add(nn.Dense(256, activation='relu'),\n nn.Dense(128, activation='relu'),\n nn.Dense(2))\n net.initialize()\n return net\n\nx = nd.random.normal(shape=(1, 512))\nnet = get_net()\nnet(x)", "_____no_output_____" ] ], [ [ "我们可以通过调用`hybridize`函数来编译和优化HybridSequential实例中串联层的计算。模型的计算结果不变。", "_____no_output_____" ] ], [ [ "net.hybridize()\nnet(x)", "_____no_output_____" ] ], [ [ "需要注意的是，只有继承HybridBlock类的层才会被优化计算。例如，HybridSequential类和Gluon提供的`Dense`类都是HybridBlock类的子类，它们都会被优化计算。如果一个层只是继承自Block类而不是HybridBlock类，那么它将不会被优化。\n\n\n### 计算性能\n\n我们比较调用`hybridize`函数前后的计算时间来展示符号式编程的性能提升。这里我们计时1000次`net`模型计算。在`net`调用`hybridize`函数前后，它分别依据命令式编程和符号式编程做模型计算。", "_____no_output_____" ] ], [ [ "def benchmark(net, x):\n start = time.time()\n for i in range(1000):\n _ = net(x)\n nd.waitall() # 等待所有计算完成方便计时\n return time.time() - start\n\nnet = get_net()\nprint('before hybridizing: %.4f sec' % (benchmark(net, x)))\nnet.hybridize()\nprint('after hybridizing: %.4f sec' % (benchmark(net, x)))", "before hybridizing: 0.3017 sec\n" ] ], [ [ "由上面结果可见，在一个HybridSequential实例调用`hybridize`函数后，它可以通过符号式编程提升计算性能。\n\n\n### 获取符号式程序\n\n在模型`net`根据输入计算模型输出后，例如`benchmark`函数中的`net(x)`，我们就可以通过`export`函数来保存符号式程序和模型参数到硬盘。", "_____no_output_____" ] ], [ [ "net.export('my_mlp')", "_____no_output_____" ] ], [ [ "此时生成的.json和.params文件分别为符号式程序和模型参数。它们可以被Python或MXNet支持的其他前端语言读取，例如C++、R、Scala、Perl和其它语言。这样，我们就可以很方便地使用其他前端语言或在其他设备上部署训练好的模型。同时，由于部署时使用的是基于符号式编程的程序，计算性能往往比基于命令式编程时更好。\n\n在MXNet中，符号式程序指的是Symbol类型的程序。我们知道，当给`net`提供NDArray类型的输入`x`后，`net(x)`会根据`x`直接计算模型输出并返回结果。对于调用过`hybridize`函数后的模型，我们还可以给它输入一个Symbol类型的变量，`net(x)`会返回Symbol类型的结果。", "_____no_output_____" ] ], [ [ "x = sym.var('data')\nnet(x)", "_____no_output_____" ] ], [ [ "## 使用HybridBlock类构造模型\n\n和Sequential类与Block类之间的关系一样，HybridSequential类是HybridBlock类的子类。跟Block实例需要实现`forward`函数不太一样的是，对于HybridBlock实例我们需要实现`hybrid_forward`函数。\n\n前面我们展示了调用`hybridize`函数后的模型可以获得更好的计算性能和可移植性。另一方面，调用`hybridize`函数后的模型会影响灵活性。为了解释这一点，我们先使用HybridBlock类构造模型。", "_____no_output_____" ] ], [ [ "class HybridNet(nn.HybridBlock):\n def __init__(self, **kwargs):\n super(HybridNet, self).__init__(**kwargs)\n self.hidden = nn.Dense(10)\n self.output = nn.Dense(2)\n\n def hybrid_forward(self, F, x):\n print('F: ', F)\n print('x: ', x)\n x = F.relu(self.hidden(x))\n print('hidden: ', x)\n return self.output(x)", "_____no_output_____" ] ], [ [ "在继承HybridBlock类时，我们需要在`hybrid_forward`函数中添加额外的输入`F`。我们知道，MXNet既有基于命令式编程的NDArray类，又有基于符号式编程的Symbol类。由于这两个类的函数基本一致，MXNet会根据输入来决定`F`使用NDArray或Symbol。\n\n下面创建了一个HybridBlock实例。可以看到默认下`F`使用NDArray。而且，我们打印出了输入`x`和使用ReLU激活函数的隐藏层的输出。", "_____no_output_____" ] ], [ [ "net = HybridNet()\nnet.initialize()\nx = nd.random.normal(shape=(1, 4))\nnet(x)", "F: <module 'mxnet.ndarray' from '/var/lib/jenkins/miniconda2/envs/d2l-zh-build/lib/python3.6/site-packages/mxnet/ndarray/__init__.py'>\nx: \n[[-0.12225834 0.5429998 -0.9469352 0.59643304]]\n<NDArray 1x4 @cpu(0)>\nhidden: \n[[0.11134676 0.04770704 0.05341475 0. 0.08091211 0.\n 0. 0.04143535 0. 0. ]]\n<NDArray 1x10 @cpu(0)>\n" ] ], [ [ "再运行一次前向计算会得到同样的结果。", "_____no_output_____" ] ], [ [ "net(x)", "F: <module 'mxnet.ndarray' from '/var/lib/jenkins/miniconda2/envs/d2l-zh-build/lib/python3.6/site-packages/mxnet/ndarray/__init__.py'>\nx: \n[[-0.12225834 0.5429998 -0.9469352 0.59643304]]\n<NDArray 1x4 @cpu(0)>\nhidden: \n[[0.11134676 0.04770704 0.05341475 0. 0.08091211 0.\n 0. 0.04143535 0. 0. ]]\n<NDArray 1x10 @cpu(0)>\n" ] ], [ [ "接下来看看调用`hybridize`函数后会发生什么。", "_____no_output_____" ] ], [ [ "net.hybridize()\nnet(x)", "F: <module 'mxnet.symbol' from '/var/lib/jenkins/miniconda2/envs/d2l-zh-build/lib/python3.6/site-packages/mxnet/symbol/__init__.py'>\nx: <Symbol data>\nhidden: <Symbol hybridnet0_relu0>\n" ] ], [ [ "可以看到，`F`变成了Symbol。而且，虽然输入数据还是NDArray，但`hybrid_forward`函数里，相同输入和中间输出全部变成了Symbol类型。\n\n再运行一次前向计算看看。", "_____no_output_____" ] ], [ [ "net(x)", "_____no_output_____" ] ], [ [ "可以看到`hybrid_forward`函数里定义的三行打印语句都没有打印任何东西。这是因为上一次在调用`hybridize`函数后运行`net(x)`的时候，符号式程序已经得到。之后再运行`net(x)`的时候MXNet将不再访问Python代码，而是直接在C++后端执行符号式程序。这也是调用`hybridize`后模型计算性能会提升的一个原因。但它可能的问题在于我们损失了写程序的灵活性。在上面这个例子中，如果我们希望使用那三行打印语句调试代码，执行符号式程序时会跳过它们无法打印。此外，对于少数像`asnumpy`这样的Symbol所不支持的函数，以及像`a += b`和`a[:] = a + b`（需改写为`a = a + b`）这样的原地（in-place）操作，我们无法在`hybrid_forward`函数中使用并在调用`hybridize`函数后进行前向计算。\n\n\n## 小结\n\n* 命令式编程和符号式编程各有优劣。MXNet通过混合式编程取二者之长。\n* 通过HybridSequential类和HybridBlock类构建的模型可以调用`hybridize`函数将命令式程序转成符号式程序。我们建议大家使用这种方法获得计算性能的提升。\n\n\n## 练习\n\n* 在本节HybridNet类的`hybrid_forward`函数中第一行添加`x.asnumpy()`，运行本节全部代码，观察报错的位置和错误类型。\n* 如果在`hybrid_forward`函数中加入Python的`if`和`for`语句会怎么样？\n* 回顾前面几章中你感兴趣的模型，改用HybridBlock类或HybridSequential类实现。\n\n\n## 扫码直达[讨论区](https://discuss.gluon.ai/t/topic/1665)\n\n", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown" ]

[ [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "from selenium import webdriver\nfrom selenium.webdriver.common.keys import Keys\nimport time\nimport os\nimport warnings\nwarnings.filterwarnings(\"ignore\")\n\nchromedriver = \"/Applications/chromedriver\"\nos.environ[\"webdriver.chrome.driver\"] = chromedriver", "_____no_output_____" ], [ "# Set url to Five Star Alliance page for New York City\nurl = 'https://www.fivestaralliance.com/luxury-hotels/271/north-america/united-states-northeast/new-york-ny/'\ndriver = webdriver.Chrome(chromedriver)\ndriver.get(url)\n# manually click \"View more Hotels\" (x3)", "_____no_output_____" ], [ "# Find links to hotels and save to list\nhotel_links=driver.find_elements_by_xpath(\"//a[@style='cursor: pointer;']\")", "_____no_output_____" ], [ "hotel_list=[]\nfor hotel in hotel_links:\n hotel_list.append(hotel.text)", "_____no_output_____" ], [ "# Check that hotel list pulled in correctly\nfor hotel in hotel_list:\n print (hotel)", "The Greenwich Hotel\nThe Soho Grand Hotel\nThe James New York - NoMad\nThe Chatwal New York\nMandarin Oriental, New York\nThe Pierre Hotel New York\nThe Wagner at the Battery\nThe Mark New York\nBlakely New York\nInterContinental The Barclay New York\nHotel Plaza Athenee New York\nLangham Place Fifth Avenue\nThe Carlyle\nGramercy Park Hotel\nInterContinental New York Times Square\nThe London NYC\nThe Beekman\nSofitel New York Hotel\nPark Hyatt New York\nThe Plaza Hotel\nThe Quin Hotel\nThe Lotte New York Palace\nGansevoort Meatpacking NYC\nThe Dominick\nTrump International Hotel & Tower New York\nThe Peninsula New York\nThe St Regis New York\nTopping Rose House\nLoews Regency Hotel\nThe Surrey New York\nThe Ritz-Carlton New York, Central Park\nJW Marriott Essex House New York\nBaccarat Hotel New York\nW New York Union Square\nThe Benjamin Hotel\nHotel 50 Bowery\nRoyalton Park Avenue\nThe NoMad Hotel\nThe High Line Hotel\nThe New York EDITION\nFour Seasons New York Downtown\nViceroy New York\nWesthouse Hotel New York\nThe Knickerbocker\nThe Broome Hotel\nHotel Americano\nMarmara Park Avenue\nThe Mercer Hotel\nRefinery Hotel New York\nNoMo Soho\nConrad New York\nRoyalton Hotel\nThe Standard High Line\nWestin New York Times Square\nHotel 48 Lex\nDream Downtown\nONE UN New York\nNolitan Hotel New York\nThe Iroquois New York\nThe Algonquin\nHotel 373 Fifth Avenue\nThe Westin New York Grand Central\nHyatt Regency Jersey City On the Hudson\nSoho House New York\nThe City Club Hotel\nThe Garden City Hotel\nThe Library Hotel\nArcher Hotel New York\nMillenium Hilton\nHamilton Park Hotel\nAndaz 5th Avenue\nW New York Times Square\nCastle Hotel and Spa\nThe Lowell\nParker New York\nFour Seasons New York\nThe James New York - SoHo\nW New York Downtown\nCrosby Street Hotel\nThe Maxwell New York\nAndaz Wall Street\nSt Giles Tuscany New York\nThe Marcel At Gramercy\nCassa Hotel 45th Street\nChambers Hotel\nThe Muse Hotel New York\nMillennium Broadway New York\nHotel Metro New York\n6 Columbus\nWarwick New York Hotel\nPark Lane Hotel New York\nEventi Hotel New York\nThe Kimberly Hotel\nSt Giles The Court New York\nSIXTY SoHo\nBryant Park\nHotel Mela New York\nInk48, A Kimpton Hotel\n70 Park Avenue Hotel\nDuane Street Hotel\nOmni Berkshire Place New York\nGild Hall\nThe Roger New York\nThe Kitano New York\nMorgans\nSmyth\nThe Roxy Hotel Tribeca\nOheka Castle Hotel and Estate\nW Hoboken\nDylan Hotel New York\nThe Premier Hotel New York\nThe Michelangelo Hotel New York\nThe Waldorf-Astoria\nWyndham Midtown 45\n" ], [ "# Remove hotels without ratings\nhotel_list.remove('The Beekman')\nhotel_list.remove('Baccarat Hotel New York')\nhotel_list.remove('The Algonquin')\nprint (hotel_list)", "['The Greenwich Hotel', 'The Soho Grand Hotel', 'The James New York - NoMad', 'The Chatwal New York', 'Mandarin Oriental, New York', 'The Pierre Hotel New York', 'The Wagner at the Battery', 'The Mark New York', 'Blakely New York', 'InterContinental The Barclay New York', 'Hotel Plaza Athenee New York', 'Langham Place Fifth Avenue', 'The Carlyle', 'Gramercy Park Hotel', 'InterContinental New York Times Square', 'The London NYC', 'Sofitel New York Hotel', 'Park Hyatt New York', 'The Plaza Hotel', 'The Quin Hotel', 'The Lotte New York Palace', 'Gansevoort Meatpacking NYC', 'The Dominick', 'Trump International Hotel & Tower New York', 'The Peninsula New York', 'The St Regis New York', 'Topping Rose House', 'Loews Regency Hotel', 'The Surrey New York', 'The Ritz-Carlton New York, Central Park', 'JW Marriott Essex House New York', 'W New York Union Square', 'The Benjamin Hotel', 'Hotel 50 Bowery', 'Royalton Park Avenue', 'The NoMad Hotel', 'The High Line Hotel', 'The New York EDITION', 'Four Seasons New York Downtown', 'Viceroy New York', 'Westhouse Hotel New York', 'The Knickerbocker', 'The Broome Hotel', 'Hotel Americano', 'Marmara Park Avenue', 'The Mercer Hotel', 'Refinery Hotel New York', 'NoMo Soho', 'Conrad New York', 'Royalton Hotel', 'The Standard High Line', 'Westin New York Times Square', 'Hotel 48 Lex', 'Dream Downtown', 'ONE UN New York', 'Nolitan Hotel New York', 'The Iroquois New York', 'Hotel 373 Fifth Avenue', 'The Westin New York Grand Central', 'Hyatt Regency Jersey City On the Hudson', 'Soho House New York', 'The City Club Hotel', 'The Garden City Hotel', 'The Library Hotel', 'Archer Hotel New York', 'Millenium Hilton', 'Hamilton Park Hotel', 'Andaz 5th Avenue', 'W New York Times Square', 'Castle Hotel and Spa', 'The Lowell', 'Parker New York', 'Four Seasons New York', 'The James New York - SoHo', 'W New York Downtown', 'Crosby Street Hotel', 'The Maxwell New York', 'Andaz Wall Street', 'St Giles Tuscany New York', 'The Marcel At Gramercy', 'Cassa Hotel 45th Street', 'Chambers Hotel', 'The Muse Hotel New York', 'Millennium Broadway New York', 'Hotel Metro New York', '6 Columbus', 'Warwick New York Hotel', 'Park Lane Hotel New York', 'Eventi Hotel New York', 'The Kimberly Hotel', 'St Giles The Court New York', 'SIXTY SoHo', 'Bryant Park', 'Hotel Mela New York', 'Ink48, A Kimpton Hotel', '70 Park Avenue Hotel', 'Duane Street Hotel', 'Omni Berkshire Place New York', 'Gild Hall', 'The Roger New York', 'The Kitano New York', 'Morgans', 'Smyth', 'The Roxy Hotel Tribeca', 'Oheka Castle Hotel and Estate', 'W Hoboken', 'Dylan Hotel New York', 'The Premier Hotel New York', 'The Michelangelo Hotel New York', 'The Waldorf-Astoria', 'Wyndham Midtown 45']\n" ], [ "# Initialize an empty dictionary for storing scraped hotel data\nHotel_Data_Dict={}", "_____no_output_____" ], [ "# Loop through hotel names to click on each and scrape ratings, reviews, amenities (& activities)\nfor hotel in hotel_list:\n Hotel_Data_Dict[hotel]=[]\n driver.find_element_by_link_text(hotel).send_keys(Keys.COMMAND,Keys.RETURN)\n time.sleep(1)\n driver.switch_to_window(driver.window_handles[-1])\n rating=driver.find_element_by_class_name('value')\n Hotel_Data_Dict[hotel].append(rating.text.strip())\n reviews=driver.find_element_by_xpath('//a[@id=\"trustyou_review\"]')\n Hotel_Data_Dict[hotel].append(reviews.text)\n amenities=driver.find_element_by_xpath('//li[@rel=\"tab4\"]')\n amenities.click()\n amenities_data=driver.find_elements_by_xpath('//div[@class=\"col-md-4\"]')\n for amenity in amenities_data:\n Hotel_Data_Dict[hotel].append(amenity.text)\n driver.close()\n driver.switch_to_window(driver.window_handles[0])", "/anaconda3/lib/python3.6/site-packages/ipykernel_launcher.py:5: DeprecationWarning: use driver.switch_to.window instead\n \"\"\"\n/anaconda3/lib/python3.6/site-packages/ipykernel_launcher.py:16: DeprecationWarning: use driver.switch_to.window instead\n app.launch_new_instance()\n" ], [ "# Check scraped hotel data and include count of hotels (should be 111)\nprint (len(Hotel_Data_Dict), Hotel_Data_Dict)", "111 {'The Greenwich Hotel': ['94', '271 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed\\n- Pool Indoor\\n- Spa Facility', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'The Soho Grand Hotel': ['86', '2972 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'The James New York - NoMad': ['83', '2296 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- High Speed Internet', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'The Chatwal New York': ['90', '334 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Spa Facility', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'Mandarin Oriental, New York': ['89', '594 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Disabled Access\\n- Executive Retreat\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed\\n- Pool Indoor\\n- Spa Facility', 'Available Activities\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'The Pierre Hotel New York': ['91', '3331 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Executive Retreat\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed\\n- Spa Facility', 'Available Activities\\n- Biking Touring\\n- Horseback Riding\\n- Ice Skating\\n- Jogging Running\\n- Shopping\\n- Tennis Courts Nearby\\n- Theatre & Museums'], 'The Wagner at the Battery': ['89', '424 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Executive Retreat\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed\\n- Spa Facility', 'Available Activities\\n- Biking Touring\\n- Dining\\n- Golfing\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'The Mark New York': ['92', '791 Reviews', 'Hotel Amenities\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed\\n- Spa Facility', 'Available Activities\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'Blakely New York': ['86', '3065 Reviews', 'Hotel Amenities\\n- Business Center\\n- Disabled Access\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'InterContinental The Barclay New York': ['88', '4210 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Executive Retreat\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- High Speed Internet\\n- Pets Allowed', 'Available Activities\\n- Dining\\n- Ice Skating\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'Hotel Plaza Athenee New York': ['86', '1048 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Spa Facility', 'Available Activities\\n- Biking Touring\\n- Dining\\n- Ice Skating\\n- Shopping\\n- Tennis Courts Nearby\\n- Theatre & Museums'], 'Langham Place Fifth Avenue': ['92', '2006 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed\\n- Spa Facility', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'The Carlyle': ['86', '1796 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Executive Retreat\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed\\n- Spa Facility', 'Available Activities\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'Gramercy Park Hotel': ['87', '1301 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Child Programs\\n- Pets Allowed', 'Available Activities\\n- Dining\\n- Horseback Riding\\n- Ice Skating\\n- Jogging Running\\n- Shopping\\n- Tennis Courts Nearby\\n- Theatre & Museums'], 'InterContinental New York Times Square': ['83', '5147 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'The London NYC': ['83', '2634 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- High Speed Internet', 'Available Activities\\n- Dining\\n- Shopping\\n- Theatre & Museums'], 'Sofitel New York Hotel': ['87', '10434 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed', 'Available Activities\\n- Biking Touring\\n- Dining\\n- Hiking\\n- Horseback Riding\\n- Ice Skating\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'Park Hyatt New York': ['93', '931 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pool Indoor\\n- Spa Facility', 'Available Activities\\n- Ice Skating\\n- Jogging Running\\n- Shopping'], 'The Plaza Hotel': ['86', '1048 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Child Programs\\n- Pets Allowed\\n- Spa Facility', 'Available Activities\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'The Quin Hotel': ['88', '2191 Reviews', 'Hotel Amenities\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- High Speed Internet', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'The Lotte New York Palace': ['89', '4233 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Child Programs\\n- Pets Allowed\\n- Spa Facility', 'Available Activities\\n- Biking Touring\\n- Dining\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'Gansevoort Meatpacking NYC': ['84', '3278 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Child Programs\\n- Pets Allowed\\n- Pool Outdoor\\n- Spa Facility', 'Available Activities\\n- Biking Touring\\n- Shopping\\n- Theatre & Museums'], 'The Dominick': ['85', '1377 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pool Outdoor\\n- Spa Facility', 'Available Activities\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'Trump International Hotel & Tower New York': ['78', '2194 Reviews', 'Hotel Amenities\\n- Business Center\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Child Programs\\n- Pool Indoor\\n- Spa Facility', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'The Peninsula New York': ['92', '1321 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Disabled Access\\n- Executive Retreat\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed\\n- Pool Indoor\\n- Spa Facility', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'The St Regis New York': ['93', '691 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Disabled Access\\n- Executive Retreat\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- High Speed Internet', 'Available Activities\\n- Golfing\\n- Shopping\\n- Theatre & Museums'], 'Topping Rose House': ['77', '60 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- High Speed Internet\\n- Pool Outdoor\\n- Spa Facility', 'Available Activities\\n- Beach\\n- Biking Touring\\n- Boating\\n- Fishing Fly\\n- Fishing Ocean\\n- Golfing\\n- Horseback Riding\\n- Sailing\\n- Shopping\\n- Tennis Courts Nearby\\n- Theatre & Museums\\n- Winery Tours'], 'Loews Regency Hotel': ['89', '1429 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Executive Retreat\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Child Programs\\n- Pets Allowed\\n- Spa Facility', 'Available Activities\\n- Dining\\n- Ice Skating\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'The Surrey New York': ['86', '1819 Reviews', 'Hotel Amenities\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Spa Facility', 'Available Activities\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'The Ritz-Carlton New York, Central Park': ['89', '581 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Executive Retreat\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Child Programs\\n- Pets Allowed\\n- Spa Facility', 'Available Activities\\n- Golf Driving Range\\n- Golfing\\n- Jogging Running\\n- Shopping'], 'JW Marriott Essex House New York': ['86', '1568 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Executive Retreat\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed', 'Available Activities\\n- Biking Touring\\n- Dining\\n- Horseback Riding\\n- Ice Skating\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'W New York Union Square': ['86', '918 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Disabled Access\\n- Executive Retreat\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'The Benjamin Hotel': ['85', '3251 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed\\n- Spa Facility', 'Available Activities\\n- Dining\\n- Shopping\\n- Theatre & Museums'], 'Hotel 50 Bowery': ['91', '2115 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed', 'Available Activities\\n- Biking Touring\\n- Dining\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'Royalton Park Avenue': ['85', '1901 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed\\n- Pool Outdoor\\n- Spa Facility'], 'The NoMad Hotel': ['91', '1954 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed', 'Available Activities\\n- Ice Skating\\n- Jogging Running\\n- Shopping\\n- Tennis Courts Nearby\\n- Theatre & Museums'], 'The High Line Hotel': ['87', '892 Reviews', 'Hotel Amenities\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining Nearby\\n- High Speed Internet\\n- Pets Allowed', 'Available Activities\\n- Dining'], 'The New York EDITION': ['90', '1591 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Spa Facility', 'Available Activities\\n- Dining\\n- Shopping\\n- Theatre & Museums'], 'Four Seasons New York Downtown': ['94', '833 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Child Programs\\n- Pool Indoor\\n- Spa Facility', 'Available Activities\\n- Boating\\n- Dining\\n- Ice Skating\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'Viceroy New York': ['80', '2111 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Executive Retreat\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed\\n- Pool Indoor', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'Westhouse Hotel New York': ['82', '1410 Reviews', 'Hotel Amenities\\n- Disabled Access\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Spa Facility', 'Available Activities\\n- Dining\\n- Shopping\\n- Theatre & Museums'], 'The Knickerbocker': ['89', '3714 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Spa Facility', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'The Broome Hotel': ['91', '336 Reviews', 'Hotel Amenities\\n- Disabled Access\\n- Fine Dining Nearby\\n- High Speed Internet', 'Available Activities\\n- Dining\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'Hotel Americano': ['80', '667 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- High Speed Internet\\n- Pool Outdoor', 'Available Activities\\n- Biking Touring\\n- Shopping\\n- Theatre & Museums'], 'Marmara Park Avenue': ['86', '649 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Executive Retreat\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed\\n- Pool Indoor\\n- Spa Facility', 'Available Activities\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'The Mercer Hotel': ['88', '555 Reviews', 'Hotel Amenities\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'Refinery Hotel New York': ['89', '2359 Reviews', 'Hotel Amenities\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'NoMo Soho': ['83', '2992 Reviews', 'Hotel Amenities\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- High Speed Internet', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'Conrad New York': ['93', '2925 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed', 'Available Activities\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'Royalton Hotel': ['83', '2016 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'The Standard High Line': ['84', '4550 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed', 'Available Activities\\n- Dining\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'Westin New York Times Square': ['84', '4470 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed\\n- Spa Facility'], 'Hotel 48 Lex': ['84', '1366 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet', 'Available Activities\\n- Shopping'], 'Dream Downtown': ['78', '3537 Reviews', 'Hotel Amenities\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Pool Outdoor\\n- Spa Facility', 'Available Activities\\n- Dining\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'ONE UN New York': ['84', '5541 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Tennis Courts On Site', 'Available Activities\\n- Theatre & Museums'], 'Nolitan Hotel New York': ['87', '1057 Reviews', 'Hotel Amenities\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed', 'Available Activities\\n- Biking Touring\\n- Shopping\\n- Theatre & Museums'], 'The Iroquois New York': ['90', '1292 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet', 'Available Activities\\n- Ice Skating\\n- Shopping\\n- Theatre & Museums'], 'Hotel 373 Fifth Avenue': ['82', '2454 Reviews', 'Hotel Amenities\\n- Disabled Access\\n- Fine Dining On Site\\n- High Speed Internet\\n- Pets Allowed'], 'The Westin New York Grand Central': ['83', '3648 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'Hyatt Regency Jersey City On the Hudson': ['88', '2752 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pool Indoor', 'Available Activities\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'Soho House New York': ['90', '573 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- High Speed Internet\\n- Pool Outdoor\\n- Spa Facility', 'Available Activities\\n- Dining\\n- Ice Skating\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'The City Club Hotel': ['83', '1926 Reviews', 'Hotel Amenities\\n- Business Center\\n- Fine Dining On Site\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'The Garden City Hotel': ['92', '439 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed\\n- Pool Outdoor', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'The Library Hotel': ['93', '939 Reviews', 'Hotel Amenities\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- High Speed Internet', 'Available Activities\\n- Dining\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'Archer Hotel New York': ['92', '4336 Reviews', 'Hotel Amenities\\n- Business Center\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'Millenium Hilton': ['84', '5062 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pool Outdoor', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'Hamilton Park Hotel': ['85', '771 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pool Outdoor', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'Andaz 5th Avenue': ['88', '1279 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'W New York Times Square': ['81', '3415 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Executive Retreat\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Child Programs\\n- Pets Allowed\\n- Spa Facility', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'Castle Hotel and Spa': ['87', '98 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Executive Retreat\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pool Outdoor\\n- Spa Facility\\n- Tennis Courts On Site', 'Available Activities\\n- Biking Mountain\\n- Boating\\n- Dining\\n- Ecological Tourism\\n- Hiking\\n- Jogging Running\\n- Tennis Courts Nearby\\n- Theatre & Museums\\n- Winery Tours'], 'The Lowell': ['93', '174 Reviews', 'Hotel Amenities\\n- Business Center\\n- Disabled Access\\n- Executive Retreat\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'Parker New York': ['83', '2069 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pool Outdoor\\n- Spa Facility', 'Available Activities\\n- Biking Touring\\n- Dining\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'Four Seasons New York': ['92', '772 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Child Programs\\n- Pets Allowed\\n- Spa Facility', 'Available Activities\\n- Ballooning Hot Air\\n- Biking Touring\\n- Boating\\n- Dining\\n- Golf Driving Range\\n- Golfing\\n- Horseback Riding\\n- Ice Skating\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'The James New York - SoHo': ['82', '1216 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining Nearby\\n- High Speed Internet', 'Available Activities\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'W New York Downtown': ['84', '1116 Reviews', 'Hotel Amenities\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'Crosby Street Hotel': ['93', '586 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'The Maxwell New York': ['78', '2791 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Executive Retreat\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed\\n- Spa Facility', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'Andaz Wall Street': ['88', '1670 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed\\n- Spa Facility', 'Available Activities\\n- Jogging Running'], 'St Giles Tuscany New York': ['89', '1326 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Disabled Access\\n- Executive Retreat\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Child Programs\\n- Spa Facility', 'Available Activities\\n- Ecological Tourism\\n- Shopping\\n- Theatre & Museums'], 'The Marcel At Gramercy': ['83', '2562 Reviews', 'Hotel Amenities\\n- Fine Dining On Site\\n- High Speed Internet'], 'Cassa Hotel 45th Street': ['86', '2897 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'Chambers Hotel': ['88', '1369 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Disabled Access\\n- Fine Dining On Site\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed'], 'The Muse Hotel New York': ['90', '1687 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Disabled Access\\n- Fine Dining On Site\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'Millennium Broadway New York': ['76', '7293 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet', 'Available Activities\\n- Ice Skating\\n- Shopping\\n- Theatre & Museums'], 'Hotel Metro New York': ['85', '2510 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Disabled Access\\n- Fine Dining On Site\\n- Fitness Center\\n- High Speed Internet'], '6 Columbus': ['78', '744 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- High Speed Internet'], 'Warwick New York Hotel': ['83', '5729 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet'], 'Park Lane Hotel New York': ['77', '8867 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Child Programs\\n- Pets Allowed', 'Available Activities\\n- Ecological Tourism\\n- Shopping\\n- Theatre & Museums'], 'Eventi Hotel New York': ['88', '2164 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Fine Dining On Site\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed'], 'The Kimberly Hotel': ['93', '1905 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Fine Dining On Site\\n- High Speed Internet'], 'St Giles The Court New York': ['77', '2562 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- High Speed Internet', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'SIXTY SoHo': ['88', '1126 Reviews', 'Hotel Amenities\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed', 'Available Activities\\n- Dining\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'Bryant Park': ['91', '1621 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Executive Retreat\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet', 'Available Activities\\n- Biking Touring\\n- Dining\\n- Ice Skating\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'Hotel Mela New York': ['78', '3888 Reviews', 'Hotel Amenities\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Executive Retreat\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed', 'Available Activities\\n- Biking Touring\\n- Dining\\n- Horseback Riding\\n- Ice Skating\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'Ink48, A Kimpton Hotel': ['88', '2204 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Disabled Access\\n- Fine Dining On Site\\n- High Speed Internet\\n- Pets Allowed'], '70 Park Avenue Hotel': ['84', '3386 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet', 'Available Activities\\n- Dining\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'Duane Street Hotel': ['85', '514 Reviews', 'Hotel Amenities\\n- Business Center\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- High Speed Internet\\n- Pets Allowed'], 'Omni Berkshire Place New York': ['88', '1780 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Executive Retreat\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Child Programs\\n- Pets Allowed', 'Available Activities\\n- Dining\\n- Ice Skating\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'Gild Hall': ['90', '1408 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet', 'Available Activities\\n- Dining\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'The Roger New York': ['84', '2206 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet', 'Available Activities\\n- Shopping'], 'The Kitano New York': ['87', '1687 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet', 'Available Activities\\n- Shopping\\n- Theatre & Museums'], 'Morgans': ['84', '480 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Executive Retreat\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- High Speed Internet', 'Available Activities\\n- Dining\\n- Ice Skating\\n- Shopping\\n- Theatre & Museums'], 'Smyth': ['87', '760 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed', 'Available Activities\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'The Roxy Hotel Tribeca': ['89', '2661 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed', 'Available Activities\\n- Dining\\n- Shopping\\n- Theatre & Museums'], 'Oheka Castle Hotel and Estate': ['90', '84 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet', 'Available Activities\\n- Dining\\n- Golfing\\n- Hiking\\n- Jogging Running\\n- Tennis Courts Nearby\\n- Theatre & Museums\\n- Winery Tours'], 'W Hoboken': ['89', '554 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed\\n- Spa Facility', 'Available Activities\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'Dylan Hotel New York': ['79', '2324 Reviews', 'Hotel Amenities\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet', 'Available Activities\\n- Dining\\n- Shopping\\n- Theatre & Museums'], 'The Premier Hotel New York': ['82', '1591 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet', 'Available Activities\\n- Ice Skating\\n- Jogging Running\\n- Shopping\\n- Tennis Courts Nearby\\n- Theatre & Museums'], 'The Michelangelo Hotel New York': ['88', '2589 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet', 'Available Activities\\n- Theatre & Museums'], 'The Waldorf-Astoria': ['84', '2063 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Meeting Space\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- High Speed Internet', 'Available Activities\\n- Biking Touring\\n- Dining\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums'], 'Wyndham Midtown 45': ['86', '849 Reviews', 'Hotel Amenities\\n- Banquets & Meetings\\n- Business Center\\n- Disabled Access\\n- Fine Dining On Site\\n- Fine Dining Nearby\\n- Fitness Center\\n- High Speed Internet\\n- Pets Allowed', 'Available Activities\\n- Dining\\n- Hiking\\n- Ice Skating\\n- Jogging Running\\n- Shopping\\n- Theatre & Museums']}\n" ], [ "import pandas as pd\nimport numpy as np\nimport matplotlib.pyplot as plt\nimport seaborn as sns\n%matplotlib inline", "_____no_output_____" ], [ "# Put scraped data into data frame\ndf=pd.DataFrame.from_dict(Hotel_Data_Dict, orient='index')", "_____no_output_____" ], [ "df", "_____no_output_____" ], [ "# Create new columns for each amenity/activity and code as 1 for having it, and 0 otherwise\n\namenities_list=['Banquets & Meetings','Business Center','Child Programs','Disabled Access',\n 'Executive Retreat','Fine Dining Nearby','Fine Dining On Site','Fitness Center',\n 'High Speed Internet','Meeting Space','Pets Allowed','Pool Indoor','Pool Outdoor',\n 'Spa Facility','Tennis Courts On Site']\n\nfor amenity in amenities_list:\n df[amenity]=0\n for hotel in hotel_list:\n if amenity in str(df.loc[hotel,2]):\n df.loc[hotel,amenity]=1\n else:\n df.loc[hotel,amenity]=0\n\nactivities_list=['Beach','Biking Touring','Boating','Dining','Ecological Tourism','Fishing Ocean',\n 'Fishing Fly','Golf Driving Range','Golfing','Hiking','Horseback Riding','Ice Skating',\n 'Jogging Running','Shopping','Tennis Courts Nearby','Theatre & Museums','Winery Tours']\n\nfor activity in activities_list:\n df[activity]=0\n for hotel in hotel_list:\n if activity in str(df.loc[hotel,3]):\n df.loc[hotel,activity]=1\n else:\n df.loc[hotel,activity]=0", "_____no_output_____" ], [ "# Strip text and whitespace from reviews, leaving only number\ndf[1]=df[1].str.replace(' Reviews','')", "_____no_output_____" ], [ "# Convert ratings and # of reviews to integer\ndf[0]=df[0].astype(int)\ndf[1]=df[1].astype(int)", "_____no_output_____" ], [ "df=df.rename(columns={0:'Rating',1:'Number of Reviews'})", "_____no_output_____" ], [ "# Delete scraped-text columns now that dummy variables have been created based on them\ndel df[2]\ndel df[3]", "_____no_output_____" ], [ "df", "_____no_output_____" ], [ "df.to_pickle('my_df.pkl')\ndel df\n", "_____no_output_____" ], [ "df = pd.read_pickle('my_df.pkl')\ndf", "_____no_output_____" ], [ "df_corr=df.corr()", "_____no_output_____" ], [ "df_corr", "_____no_output_____" ], [ "df_corr['Rating'].sort_values(ascending=False)", "_____no_output_____" ], [ "import os\nprint(os.getcwd())", "_____no_output_____" ], [ "import csv\ndf_corr.to_csv('correlation_matrix.csv')", "_____no_output_____" ], [ "\"\"\"\"Look at csv to find multicollinearity. Beach, Fishing Ocean, and Fishing Fly are perfectly correlated, \nso dropping the latter two. Shopping and Theatre & Museums are highly correlated at 0.73, but IMO this isn't \nhigh enough to exclude one or the other, especially since they aren't the most noteworthy variables in the \nfinal analysis.", "_____no_output_____" ], [ "del df['Fishing Fly']\ndel df['Fishing Ocean']", "_____no_output_____" ], [ "df.columns", "_____no_output_____" ], [ "plt.hist(df.Rating,bins=10)\nplt.title('Distribution of Hotel Ratings')\nplt.xlabel('Hotel Rating')\nplt.ylabel('Number of Hotels')\nplt.xticks([75,80,85,90,95])\nplt.yticks([0,5,10,15,20])\nplt.show", "_____no_output_____" ], [ "# Examine number of reviews by rating\n\nfig, ax = plt.subplots(1, 1, figsize=(10, 8))\n\nax.scatter(df.Rating, df['Number of Reviews'])\nax.set_xlabel('Hotel Rating')\nax.set_ylabel('Number of Reviews')\nax.set_title('Number of Reviews by Hotel Rating')", "_____no_output_____" ], [ "from sklearn.linear_model import LinearRegression\nfrom sklearn import metrics\nfrom sklearn.preprocessing import PolynomialFeatures\nfrom sklearn.pipeline import make_pipeline\nfrom sklearn.cross_validation import train_test_split\nfrom sklearn.cross_validation import KFold\nfrom sklearn.metrics import mean_squared_error\nfrom sklearn.cross_validation import cross_val_score\n\n# Create two new data frames, one with the feature variables and the other with the target column\ndfX=df.drop(columns=['Rating','Number of Reviews']) # Rating is the target, and we're not going to use # of reviews\n# since they are neither discretionary nor do they tell us about customer preferences\nX=dfX\ny=df['Rating']\n\n# Split data into training and holdout/test\nX_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.25, random_state=43) # b/c 42 is overrated", "_____no_output_____" ], [ "# 3-fold cross-validation with our data\nmodel = LinearRegression()\nscores = cross_val_score(model, X_train, y_train, cv=3, scoring='mean_squared_error')\n\nprint(scores)", "[-56.7078362 -50.65953325 -52.6792722 ]\n" ], [ "##### from sklearn.model_selection import train_test_split, GridSearchCV\nfrom sklearn.model_selection import train_test_split, GridSearchCV\nfrom sklearn.linear_model import Ridge, Lasso\nfrom sklearn.preprocessing import StandardScaler\n\n# Scale X-axis data for Ridge and Lasso regressions\nssX = StandardScaler()\nX_train_scaled = ssX.fit_transform(X_train)", "_____no_output_____" ], [ "model = Ridge()\nparameters = {'alpha': [1e-3,1e-2,1e-1,1,1e1,1e2,1e3], 'fit_intercept': [True]}\ngrid = GridSearchCV(model, parameters, cv=3, scoring='mean_squared_error', n_jobs=1)\ngrid.fit(X_train_scaled, y_train)\ngrid.cv_results_", "_____no_output_____" ], [ "model = Lasso()\nparameters = {'alpha': [1e-3,1e-2,1e-1,1,1e1,1e2,1e3], 'fit_intercept': [True]}\ngrid = GridSearchCV(model, parameters, cv=3, scoring='mean_squared_error', n_jobs=1)\ngrid.fit(X_train_scaled, y_train)\ngrid.cv_results_", "_____no_output_____" ], [ "type(grid.best_estimator_)", "_____no_output_____" ], [ "# Check predicted values and residuals on best lasso, and compute sum of squared residuals.\nX_test_scaled = ssX.transform(X_test)\nbest_lasso = grid.best_estimator_\nlasso_pred = best_lasso.predict(X_test_scaled)\nresid_list = []\nSSR = 0\nfor true,pred in zip(y_test, lasso_pred):\n resid = true - pred\n resid_list.append(resid)\n SSR += resid**2\n print(\"pred, resid:\", str(pred) + \", $\"+ str(resid))\n \nprint(grid.best_params_, grid.best_score_)", "pred, resid: 86.77009813929656, $-1.7700981392965645\npred, resid: 86.77009813929656, $2.2299018607034355\npred, resid: 85.58809516865138, $2.4119048313486218\npred, resid: 85.58809516865138, $-2.5880951686513782\npred, resid: 86.77009813929656, $5.2299018607034355\npred, resid: 85.58809516865138, $-2.5880951686513782\npred, resid: 85.58809516865138, $7.411904831348622\npred, resid: 86.77009813929656, $0.22990186070343555\npred, resid: 86.77009813929656, $-2.7700981392965645\npred, resid: 85.58809516865138, $-5.588095168651378\npred, resid: 86.77009813929656, $1.2299018607034355\npred, resid: 85.58809516865138, $2.4119048313486218\npred, resid: 85.58809516865138, $-2.5880951686513782\npred, resid: 85.58809516865138, $6.411904831348622\npred, resid: 85.58809516865138, $4.411904831348622\npred, resid: 85.58809516865138, $3.4119048313486218\npred, resid: 85.58809516865138, $-1.5880951686513782\npred, resid: 85.58809516865138, $2.4119048313486218\npred, resid: 86.77009813929656, $1.2299018607034355\npred, resid: 86.77009813929656, $-8.770098139296564\npred, resid: 85.58809516865138, $8.411904831348622\npred, resid: 85.58809516865138, $0.41190483134862177\npred, resid: 86.77009813929656, $0.22990186070343555\npred, resid: 86.77009813929656, $4.2299018607034355\npred, resid: 85.58809516865138, $0.41190483134862177\npred, resid: 85.58809516865138, $4.411904831348622\npred, resid: 85.58809516865138, $3.4119048313486218\npred, resid: 85.58809516865138, $1.4119048313486218\n{'alpha': 1, 'fit_intercept': True} -19.65991145906881\n" ], [ "mean_squared_error=SSR/len(resid_list)\nprint (mean_squared_error)", "15.84707025439869\n" ], [ "# With a decent MSE on the holdout (lower than our CV set!), let's try using this lasso model (lambda=1) as our \n# final model.\n\nfrom sklearn.linear_model import LassoCV\nfinal = LassoCV(alphas=[1],cv=4)\nfinal.fit(X, y)\nfinal.score(X, y)", "_____no_output_____" ], [ "# Lambda=1 is giving an R^2 of 0 on the full set. Looks like it may be collapsing the coefficients to baseline.\n# So let's go with lambda=0.1 instead.\n\nfrom sklearn.linear_model import LassoCV\nfinal = LassoCV(alphas=[0.1],cv=4)\nfinal.fit(X, y)\nfinal.score(X, y)", "_____no_output_____" ], [ "# Now we have something to work with. Time to examine the coefficients.\nlabeled_coef=list(zip(X.columns,final.coef_))\n\nfor e in labeled_coef:\n print (e)", "('Banquets & Meetings', 0.0)\n('Business Center', -0.0)\n('Child Programs', -0.0)\n('Disabled Access', 0.4625503144988509)\n('Executive Retreat', -0.0)\n('Fine Dining Nearby', -0.0)\n('Fine Dining On Site', 0.0)\n('Fitness Center', 0.7602233015751865)\n('High Speed Internet', 0.0)\n('Meeting Space', -0.0)\n('Pets Allowed', 0.6237839406670173)\n('Pool Indoor', 0.0)\n('Pool Outdoor', -1.181228854210307)\n('Spa Facility', 0.15021986607349797)\n('Tennis Courts On Site', -0.0)\n('Beach', -0.0)\n('Biking Touring', -0.12874010618420537)\n('Boating', 0.0)\n('Dining', 0.0)\n('Ecological Tourism', -0.0)\n('Golf Driving Range', 0.0)\n('Golfing', 0.0)\n('Hiking', 0.0)\n('Horseback Riding', -0.28365682430377764)\n('Ice Skating', 0.0)\n('Jogging Running', 1.6772959749304803)\n('Shopping', -0.0)\n('Tennis Courts Nearby', -0.0)\n('Theatre & Museums', -0.0)\n('Winery Tours', -0.0)\n" ] ] ]

[ "code" ]

[ [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "## Comprehensions:\n#### Documentation: https://python-3-patterns-idioms-test.readthedocs.io/en/latest/Comprehensions.html\n\n#### Situation: \n - We have one or more sources of iterable data.\n \n#### Need:\n - We want to do something with that data, and output it into a list, dictionary or generator format.\n \n#### Solution:\n - Python offers a cleaner/faster way of working without using traditional for loops.\n\n-----\n\n \n#### Example:\n - Lets take a traditional for loop", "_____no_output_____" ] ], [ [ "even_squares = []\nfor num in range(11):\n if num%2 == 0:\n even_squares.append(num * num)\n\neven_squares\n", "_____no_output_____" ], [ "# Can we do better than the above?\n\neven_squares = [num*num for num in range(11) if num%2 == 0]\n\neven_squares", "_____no_output_____" ] ], [ [ "### List comprehension Pattern:\n\n##### [Figure reference](https://towardsdatascience.com/comprehending-the-concept-of-comprehensions-in-python-c9dafce5111)\n \n", "_____no_output_____" ], [ "### We can do the same with dictionaries, or generators:", "_____no_output_____" ] ], [ [ "first_names = ['Mark', 'Demmis', 'Elon', 'Jeff', 'Lex']\nlast_names = ['Zuckerberg','Hasabis', 'Musk','Bezos','Fridman']\n\nfull_names = {}\nfor first, last in zip(first_names, last_names):\n full_names[first] = last\n \nfull_names", "_____no_output_____" ], [ "full_names = {first: last for first, last in zip(first_names, last_names)}\n\nfull_names\n# len(full_names)", "_____no_output_____" ] ], [ [ "## How about a generator?\nLike a comprehension but waits, and yields each item out of the expression, one by one.", "_____no_output_____" ] ], [ [ "# even_squares was [0, 4, 16, 36, 64, 100] with the list comprehension\n\n# generator equivallent\neven_squares = (num*num for num in range(11) if num%2 == 0)", "_____no_output_____" ], [ "next(even_squares)", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown" ], [ "code", "code" ], [ "markdown", "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "%load_ext autoreload\n%autoreload 2\n%env CUDA_VISIBLE_DEVICES=0\nimport os, sys\nimport time\nsys.path.insert(0, '..')\nimport numpy as np\nimport matplotlib.pyplot as plt\n%matplotlib inline\nimport lib\nimport torch, torch.nn as nn\nimport torch.nn.functional as F\n\ndevice = 'cuda' if torch.cuda.is_available() else 'cpu'\n\nexperiment_name = 'year_node_8layers'\nexperiment_name = '{}_{}.{:0>2d}.{:0>2d}_{:0>2d}:{:0>2d}'.format(experiment_name, *time.gmtime()[:5])\nprint(\"experiment:\", experiment_name)", "env: CUDA_VISIBLE_DEVICES=0\nexperiment: year_node_8layers_2019.08.27_17:11\n" ], [ "data = lib.Dataset(\"YEAR\", random_state=1337, quantile_transform=True, quantile_noise=1e-3)\nin_features = data.X_train.shape[1]\n\nmu, std = data.y_train.mean(), data.y_train.std()\nnormalize = lambda x: ((x - mu) / std).astype(np.float32)\ndata.y_train, data.y_valid, data.y_test = map(normalize, [data.y_train, data.y_valid, data.y_test])\n\nprint(\"mean = %.5f, std = %.5f\" % (mu, std))", "Downloading https://www.dropbox.com/s/l09pug0ywaqsy0e/YearPredictionMSD.txt?dl=1 > ./data/YEAR/data.csv\n" ], [ "model = nn.Sequential(\n lib.DenseBlock(in_features, 128, num_layers=8, tree_dim=3, depth=6, flatten_output=False,\n choice_function=lib.entmax15, bin_function=lib.entmoid15),\n lib.Lambda(lambda x: x[..., 0].mean(dim=-1)), # average first channels of every tree\n \n).to(device)\n\nwith torch.no_grad():\n res = model(torch.as_tensor(data.X_train[:5000], device=device))\n # trigger data-aware init\n \nif torch.cuda.device_count() > 1:\n model = nn.DataParallel(model)", "_____no_output_____" ], [ "from qhoptim.pyt import QHAdam\noptimizer_params = { 'nus':(0.7, 1.0), 'betas':(0.95, 0.998) }", "_____no_output_____" ], [ "trainer = lib.Trainer(\n model=model, loss_function=F.mse_loss,\n experiment_name=experiment_name,\n warm_start=False,\n Optimizer=QHAdam,\n optimizer_params=optimizer_params,\n verbose=True,\n n_last_checkpoints=5\n)", "_____no_output_____" ], [ "from tqdm import tqdm\nfrom IPython.display import clear_output\nloss_history, mse_history = [], []\nbest_mse = float('inf')\nbest_step_mse = 0\nearly_stopping_rounds = 5000\nreport_frequency = 100", "_____no_output_____" ], [ "for batch in lib.iterate_minibatches(data.X_train, data.y_train, batch_size=1024, \n shuffle=True, epochs=float('inf')):\n metrics = trainer.train_on_batch(*batch, device=device)\n \n loss_history.append(metrics['loss'])\n\n if trainer.step % report_frequency == 0:\n trainer.save_checkpoint()\n trainer.average_checkpoints(out_tag='avg')\n trainer.load_checkpoint(tag='avg')\n mse = trainer.evaluate_mse(\n data.X_valid, data.y_valid, device=device, batch_size=16384)\n\n if mse < best_mse:\n best_mse = mse\n best_step_mse = trainer.step\n trainer.save_checkpoint(tag='best_mse')\n mse_history.append(mse)\n \n trainer.load_checkpoint() # last\n trainer.remove_old_temp_checkpoints()\n\n clear_output(True)\n plt.figure(figsize=[18, 6])\n plt.subplot(1, 2, 1)\n plt.plot(loss_history)\n plt.title('Loss')\n plt.grid()\n plt.subplot(1, 2, 2)\n plt.plot(mse_history)\n plt.title('MSE')\n plt.grid()\n plt.show()\n print(\"Loss %.5f\" % (metrics['loss']))\n print(\"Val MSE: %0.5f\" % (mse))\n if trainer.step > best_step_mse + early_stopping_rounds:\n print('BREAK. There is no improvment for {} steps'.format(early_stopping_rounds))\n print(\"Best step: \", best_step_mse)\n print(\"Best Val MSE: %0.5f\" % (best_mse))\n break", "_____no_output_____" ], [ "trainer.load_checkpoint(tag='best_mse')\nmse = trainer.evaluate_mse(data.X_test, data.y_test, device=device)\nprint('Best step: ', trainer.step)\nprint(\"Test MSE: %0.5f\" % (mse))", "Loaded logs/year_node_8layers_2019.08.27_17:11/checkpoint_best_mse.pth\nBest step: 3400\nTest MSE: 0.63787\n" ], [ "mse * std ** 2", "_____no_output_____" ] ] ]

[ "code" ]

[ [ "code", "code", "code", "code", "code", "code", "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "<a href=\"https://colab.research.google.com/github/skimaza/assist/blob/main/perceptron_Iris.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>", "_____no_output_____" ], [ "# AI 전략경영MBA 경영자를 위한 딥러닝 원리의 이해\n# Perceptron 실습 예제\n# 붓꽃 분류 문제", "_____no_output_____" ], [ "The original code comes from Sebastian Reschka's blog (http://sebastianraschka.com/Articles/2015_singlelayer_neurons.html).<br/>\nSlightly modified for the lecture. -skimaza", "_____no_output_____" ], [ "# 라이브러리 import\n- numpy: number, 특히 다차원 배열을 다루는 라이브러리(패키지)\n- pandas: 데이터를 다양한 표 형태로 취급할 수 있는 패키지\n- matplotlib: 이미지와 그래프 표시", "_____no_output_____" ] ], [ [ "import numpy as np\nimport pandas as pd\nimport matplotlib.pyplot as plt", "_____no_output_____" ] ], [ [ "# Colab으로 배정된 가상머신 확인", "_____no_output_____" ], [ "### 현재 디렉토리(폴더)\n### '!'로 시작하는 명령은 가상머신의 명령을 실행하라는 의미", "_____no_output_____" ] ], [ [ "!pwd", "/content\n" ] ], [ [ "### 현재 디렉토리의 내용", "_____no_output_____" ] ], [ [ "!ls -l", "total 12\n-rw-r--r-- 1 root root 4551 Sep 22 01:24 iris.dat\ndrwxr-xr-x 1 root root 4096 Sep 16 13:40 sample_data\n" ] ], [ [ "### sample_data directory에는 Google Colab에서 기본으로 제공하는 데이터가 있음\n### (이번 특강에서 사용할 데이터는 아님)", "_____no_output_____" ] ], [ [ "!ls sample_data", "anscombe.json\t\t mnist_test.csv\ncalifornia_housing_test.csv mnist_train_small.csv\ncalifornia_housing_train.csv README.md\n" ] ], [ [ "# 예제 코드", "_____no_output_____" ] ], [ [ "weights = []\nerrors_log = []\nepochs = 20\neta = 0.01\n\nIRIS_DATA = \"iris.dat\" # Iris 데이터셋을 저장할 파일이름", "_____no_output_____" ] ], [ [ "### os는 운영체제 관련 기능, urllib는 인터넷으로 데이터를 다운로드받기 위한 패키지\n### 인터넷에서 Iris 데이터셋을 다운로드하여 IRIS_DATA 파일에 저장", "_____no_output_____" ] ], [ [ "import os\nfrom urllib.request import urlopen\n\nif not os.path.exists(IRIS_DATA):\n raw = urlopen('https://archive.ics.uci.edu/ml/machine-learning-databases/iris/iris.data').read()\n with open(IRIS_DATA, \"wb\") as f:\n f.write(raw)", "_____no_output_____" ], [ "!ls -l", "total 12\n-rw-r--r-- 1 root root 4551 Sep 22 01:24 iris.dat\ndrwxr-xr-x 1 root root 4096 Sep 16 13:40 sample_data\n" ] ], [ [ "# pandas의 read_csv 명령을 사용하여 데이터를 pandas DataFrame 구조로 읽어들임", "_____no_output_____" ] ], [ [ "df = pd.read_csv(IRIS_DATA, header=None)", "_____no_output_____" ], [ "df", "_____no_output_____" ] ], [ [ "꽃받침 길이, 꽃받침 너비, 꽃잎 길이, 꽃잎 너비 (cm), 붓꽃 종류", "_____no_output_____" ] ], [ [ "df[4].values", "_____no_output_____" ], [ "df.iloc[0:100, 4]", "_____no_output_____" ], [ "df.iloc[0:100, 4].values", "_____no_output_____" ], [ "# setosa and versicolor\ny = np.asarray(df.iloc[0:100, 4].values)\ny = np.where(y == 'Iris-setosa', -1, 1)\n\n# sepal length and petal length\nX = np.asarray(df.iloc[0:100, [0,2]].values)", "_____no_output_____" ], [ "print(y)", "[-1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1\n -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1\n -1 -1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1\n 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1\n 1 1 1 1]\n" ], [ "print(X)", "[[5.1 1.4]\n [4.9 1.4]\n [4.7 1.3]\n [4.6 1.5]\n [5. 1.4]\n [5.4 1.7]\n [4.6 1.4]\n [5. 1.5]\n [4.4 1.4]\n [4.9 1.5]\n [5.4 1.5]\n [4.8 1.6]\n [4.8 1.4]\n [4.3 1.1]\n [5.8 1.2]\n [5.7 1.5]\n [5.4 1.3]\n [5.1 1.4]\n [5.7 1.7]\n [5.1 1.5]\n [5.4 1.7]\n [5.1 1.5]\n [4.6 1. ]\n [5.1 1.7]\n [4.8 1.9]\n [5. 1.6]\n [5. 1.6]\n [5.2 1.5]\n [5.2 1.4]\n [4.7 1.6]\n [4.8 1.6]\n [5.4 1.5]\n [5.2 1.5]\n [5.5 1.4]\n [4.9 1.5]\n [5. 1.2]\n [5.5 1.3]\n [4.9 1.5]\n [4.4 1.3]\n [5.1 1.5]\n [5. 1.3]\n [4.5 1.3]\n [4.4 1.3]\n [5. 1.6]\n [5.1 1.9]\n [4.8 1.4]\n [5.1 1.6]\n [4.6 1.4]\n [5.3 1.5]\n [5. 1.4]\n [7. 4.7]\n [6.4 4.5]\n [6.9 4.9]\n [5.5 4. ]\n [6.5 4.6]\n [5.7 4.5]\n [6.3 4.7]\n [4.9 3.3]\n [6.6 4.6]\n [5.2 3.9]\n [5. 3.5]\n [5.9 4.2]\n [6. 4. ]\n [6.1 4.7]\n [5.6 3.6]\n [6.7 4.4]\n [5.6 4.5]\n [5.8 4.1]\n [6.2 4.5]\n [5.6 3.9]\n [5.9 4.8]\n [6.1 4. ]\n [6.3 4.9]\n [6.1 4.7]\n [6.4 4.3]\n [6.6 4.4]\n [6.8 4.8]\n [6.7 5. ]\n [6. 4.5]\n [5.7 3.5]\n [5.5 3.8]\n [5.5 3.7]\n [5.8 3.9]\n [6. 5.1]\n [5.4 4.5]\n [6. 4.5]\n [6.7 4.7]\n [6.3 4.4]\n [5.6 4.1]\n [5.5 4. ]\n [5.5 4.4]\n [6.1 4.6]\n [5.8 4. ]\n [5. 3.3]\n [5.6 4.2]\n [5.7 4.2]\n [5.7 4.2]\n [6.2 4.3]\n [5.1 3. ]\n [5.7 4.1]]\n" ], [ "# Versicolor\npos = X[[y == 1]]\n# Setosa\nneg = X[[y == -1]]", "/usr/local/lib/python3.7/dist-packages/ipykernel_launcher.py:2: FutureWarning: Using a non-tuple sequence for multidimensional indexing is deprecated; use `arr[tuple(seq)]` instead of `arr[seq]`. In the future this will be interpreted as an array index, `arr[np.array(seq)]`, which will result either in an error or a different result.\n \n/usr/local/lib/python3.7/dist-packages/ipykernel_launcher.py:4: FutureWarning: Using a non-tuple sequence for multidimensional indexing is deprecated; use `arr[tuple(seq)]` instead of `arr[seq]`. In the future this will be interpreted as an array index, `arr[np.array(seq)]`, which will result either in an error or a different result.\n after removing the cwd from sys.path.\n" ], [ "print(pos)", "[[7. 4.7]\n [6.4 4.5]\n [6.9 4.9]\n [5.5 4. ]\n [6.5 4.6]\n [5.7 4.5]\n [6.3 4.7]\n [4.9 3.3]\n [6.6 4.6]\n [5.2 3.9]\n [5. 3.5]\n [5.9 4.2]\n [6. 4. ]\n [6.1 4.7]\n [5.6 3.6]\n [6.7 4.4]\n [5.6 4.5]\n [5.8 4.1]\n [6.2 4.5]\n [5.6 3.9]\n [5.9 4.8]\n [6.1 4. ]\n [6.3 4.9]\n [6.1 4.7]\n [6.4 4.3]\n [6.6 4.4]\n [6.8 4.8]\n [6.7 5. ]\n [6. 4.5]\n [5.7 3.5]\n [5.5 3.8]\n [5.5 3.7]\n [5.8 3.9]\n [6. 5.1]\n [5.4 4.5]\n [6. 4.5]\n [6.7 4.7]\n [6.3 4.4]\n [5.6 4.1]\n [5.5 4. ]\n [5.5 4.4]\n [6.1 4.6]\n [5.8 4. ]\n [5. 3.3]\n [5.6 4.2]\n [5.7 4.2]\n [5.7 4.2]\n [6.2 4.3]\n [5.1 3. ]\n [5.7 4.1]]\n" ], [ "print(neg)", "[[5.1 1.4]\n [4.9 1.4]\n [4.7 1.3]\n [4.6 1.5]\n [5. 1.4]\n [5.4 1.7]\n [4.6 1.4]\n [5. 1.5]\n [4.4 1.4]\n [4.9 1.5]\n [5.4 1.5]\n [4.8 1.6]\n [4.8 1.4]\n [4.3 1.1]\n [5.8 1.2]\n [5.7 1.5]\n [5.4 1.3]\n [5.1 1.4]\n [5.7 1.7]\n [5.1 1.5]\n [5.4 1.7]\n [5.1 1.5]\n [4.6 1. ]\n [5.1 1.7]\n [4.8 1.9]\n [5. 1.6]\n [5. 1.6]\n [5.2 1.5]\n [5.2 1.4]\n [4.7 1.6]\n [4.8 1.6]\n [5.4 1.5]\n [5.2 1.5]\n [5.5 1.4]\n [4.9 1.5]\n [5. 1.2]\n [5.5 1.3]\n [4.9 1.5]\n [4.4 1.3]\n [5.1 1.5]\n [5. 1.3]\n [4.5 1.3]\n [4.4 1.3]\n [5. 1.6]\n [5.1 1.9]\n [4.8 1.4]\n [5.1 1.6]\n [4.6 1.4]\n [5.3 1.5]\n [5. 1.4]]\n" ], [ "# versicolor with blue dots and setosa with red dots\nplt.scatter(pos[:,0], pos[:, 1], color='blue', label=\"pos\")\nplt.scatter(neg[:,0], neg[:, 1], color='red', label=\"neg\")\nplt.xlabel(\"x1\")\nplt.ylabel(\"x2\")\nplt.legend(loc=2, scatterpoints=1, fontsize=10)", "_____no_output_____" ], [ "def train(X, y, epochs=epochs, eta=eta):\n global weights\n global errors_log\n weights = np.zeros(1 + X.shape[1])\n print(\"Initial weights\", weights)\n errors_log = []\n\n for i in range(epochs):\n errors = 0\n print(\"EPOCHS\", i+1)\n for xi, target in zip(X, y):\n update = eta * (target - predict(xi))\n #print(xi, \"target\", target, \"sum\", net_input(xi), \"update\", update)\n if update != 0:\n weights[1:] += update * xi\n weights[0] += update\n print(\"Updated WEIGHTS\", weights)\n errors += int(update != 0.0)\n errors_log.append(errors)\n return\n\ndef net_input(X):\n global weights\n return np.dot(X, weights[1:]) + weights[0]\n\ndef predict(X):\n return np.where(net_input(X) > 0.0, 1, -1)", "_____no_output_____" ], [ "train(X, y)", "Initial weights [0. 0. 0.]\nEPOCHS 1\nUpdated WEIGHTS [0.02 0.14 0.094]\nEPOCHS 2\nUpdated WEIGHTS [0. 0.038 0.066]\nUpdated WEIGHTS [-0.02 -0.06 0.038]\nUpdated WEIGHTS [0. 0.08 0.132]\nEPOCHS 3\nUpdated WEIGHTS [-0.02 -0.022 0.104]\nUpdated WEIGHTS [-0.04 -0.12 0.076]\nUpdated WEIGHTS [-0.02 0.02 0.17]\nEPOCHS 4\nUpdated WEIGHTS [-0.04 -0.082 0.142]\nUpdated WEIGHTS [-0.02 0.032 0.212]\nEPOCHS 5\nUpdated WEIGHTS [-0.04 -0.07 0.184]\nEPOCHS 6\nEPOCHS 7\nEPOCHS 8\nEPOCHS 9\nEPOCHS 10\nEPOCHS 11\nEPOCHS 12\nEPOCHS 13\nEPOCHS 14\nEPOCHS 15\nEPOCHS 16\nEPOCHS 17\nEPOCHS 18\nEPOCHS 19\nEPOCHS 20\n" ], [ "print(errors_log)", "[1, 3, 3, 2, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]\n" ], [ "print(weights)", "[-0.04 -0.07 0.184]\n" ] ], [ [ "$w_{1}x_{1} + w_{2}x_{2} + w_{0} = 0$ \n$x_{2} = - \\frac{w_{1}}{w_{2}}x_{1} - \\frac{w_{0}}{w_{2}}$", "_____no_output_____" ] ], [ [ "\nfig = plt.figure()\nax = fig.add_subplot(111)\n\n# draw between 4 and 7 of x1\npoint_x = np.array([4, 7])\n# x2 = -(w0 + w1 * x1) / w2\npoint_y = np.array([- (weights[0] + weights[1] * 4) / weights[2], - (weights[0] + weights[1] * 7) / weights[2]])\nline, = ax.plot(point_x, point_y, 'b-', picker=5)\n\nax.scatter(pos[:,0], pos[:, 1], color='blue', label=\"pos\")\nax.scatter(neg[:,0], neg[:, 1], color='red', label=\"neg\")\nplt.xlabel(\"x1\")\nplt.ylabel(\"x2\")\nplt.legend(loc=2, scatterpoints=1, fontsize=10)\n\nplt.show()", "_____no_output_____" ], [ "", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown", "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "_Lambda School Data Science — Classification & Validation_ \n\n# Baselines & Validation\n\nObjectives\n- Train/Validate/Test split\n- Cross-Validation\n- Begin with baselines", "_____no_output_____" ], [ "## Weather data — mean baseline\n\nLet's try baselines for regression.\n\nYou can [get Past Weather by Zip Code from Climate.gov](https://www.climate.gov/maps-data/dataset/past-weather-zip-code-data-table). I downloaded the data for my town: Normal, Illinois.", "_____no_output_____" ] ], [ [ "%matplotlib inline\nimport matplotlib.pyplot as plt\nimport pandas as pd \n\nurl = 'https://raw.githubusercontent.com/LambdaSchool/DS-Unit-2-Sprint-3-Classification-Validation/master/module2-baselines-validation/weather-normal-il.csv'\nweather = pd.read_csv(url, parse_dates=['DATE']).set_index('DATE')\nweather['2014-05':'2019-05'].plot(y='TMAX')\nplt.title('Daily high temperature in Normal, IL');", "_____no_output_____" ] ], [ [ "Over the years, across the seasons, the average daily high temperature in my town is about 63 degrees.", "_____no_output_____" ] ], [ [ "weather['TMAX'].mean()", "_____no_output_____" ] ], [ [ "Remember from [the preread:](https://github.com/LambdaSchool/DS-Unit-2-Sprint-3-Classification-Validation/blob/master/module2-baselines-validation/model-validation-preread.md#what-does-baseline-mean) \"A baseline for regression can be the mean of the training labels.\"", "_____no_output_____" ], [ "If I predicted that every day, the high will be 63 degrees, I'd be off by about 19 degrees on average.", "_____no_output_____" ] ], [ [ "from sklearn.metrics import mean_absolute_error\npredicted = [weather['TMAX'].mean()] * len(weather) \nmean_absolute_error(weather['TMAX'], predicted)", "_____no_output_____" ] ], [ [ "But, we can get a better baseline here: \"A baseline for time-series regressions can be the value from the previous timestep.\"\n\n*Data Science for Business* explains, \n\n> Weather forecasters have two simple—but not simplistic—baseline models that they compare against. ***One (persistence) predicts that the weather tomorrow is going to be whatever it was today.*** The other (climatology) predicts whatever the average historical weather has been on this day from prior years. Each model performs considerably better than random guessing, and both are so easy to compute that they make natural baselines of comparison. Any new, more complex model must beat these.", "_____no_output_____" ], [ "Let's predict that the weather tomorrow is going to be whatever it was today. Which is another way of saying that the weather today is going to be whatever it was yesterday.\n\nWe can engineer this feature with one line of code, using the pandas [`shift`](https://pandas.pydata.org/pandas-docs/stable/generated/pandas.DataFrame.shift.html) function.\n\nThis new baseline is off by less than 6 degress on average.", "_____no_output_____" ] ], [ [ "weather['TMAX_yesterday'] = weather.TMAX.shift(1)\nweather = weather.dropna() # Drops the first date, because it doesn't have a \"yesterday\"\nmean_absolute_error(weather.TMAX, weather.TMAX_yesterday)", "_____no_output_____" ] ], [ [ "I applied this same concept for [my first submission to the Kaggle Instacart competition.](https://github.com/rrherr/springboard/blob/master/Kaggle%20Instacart%20first%20submission.ipynb)", "_____no_output_____" ], [ "## Bank Marketing — majority class baseline\n\nhttps://archive.ics.uci.edu/ml/datasets/Bank+Marketing\n\n>The data is related with direct marketing campaigns of a Portuguese banking institution. The marketing campaigns were based on phone calls. Often, more than one contact to the same client was required, in order to access if the product (bank term deposit) would be ('yes') or not ('no') subscribed. \n\n>Output variable (desired target): \n>y - has the client subscribed a term deposit? (binary: 'yes','no')\n\n>bank-additional-full.csv with all examples (41188) and 20 inputs, ordered by date (from May 2008 to November 2010)", "_____no_output_____" ], [ "Get and read the data", "_____no_output_____" ] ], [ [ "!wget https://archive.ics.uci.edu/ml/machine-learning-databases/00222/bank-additional.zip", "--2019-05-09 19:00:23-- https://archive.ics.uci.edu/ml/machine-learning-databases/00222/bank-additional.zip\nResolving archive.ics.uci.edu (archive.ics.uci.edu)... 128.195.10.252\nConnecting to archive.ics.uci.edu (archive.ics.uci.edu)|128.195.10.252|:443... connected.\nHTTP request sent, awaiting response... 200 OK\nLength: 444572 (434K) [application/x-httpd-php]\nSaving to: ‘bank-additional.zip.1’\n\nbank-additional.zip 100%[===================>] 434.15K 2.43MB/s in 0.2s \n\n2019-05-09 19:00:23 (2.43 MB/s) - ‘bank-additional.zip.1’ saved [444572/444572]\n\n" ], [ "!unzip bank-additional.zip", "Archive: bank-additional.zip\nreplace bank-additional/.DS_Store? [y]es, [n]o, [A]ll, [N]one, [r]ename: " ], [ "bank = pd.read_csv('bank-additional/bank-additional-full.csv', sep=';')", "_____no_output_____" ] ], [ [ "Assign to X and y", "_____no_output_____" ] ], [ [ "X = bank.drop(columns='y')\ny = bank['y'] == 'yes'", "_____no_output_____" ] ], [ [ "## 3-way split: Train / Validation / Test ", "_____no_output_____" ], [ "We know how to do a _two-way split_, with the [**`sklearn.model_selection.train_test_split`**](https://scikit-learn.org/stable/modules/generated/sklearn.model_selection.train_test_split.html) function:", "_____no_output_____" ] ], [ [ "from sklearn.model_selection import train_test_split\n\nX_train, X_test, y_train, y_test = train_test_split(\n X, y, test_size=0.2, random_state=42, stratify=y)", "_____no_output_____" ] ], [ [ "How can we get from a two-way split, to a three-way split?\n\nWe can use the same function again, to split the training data into training and validation data.", "_____no_output_____" ] ], [ [ "X_train, X_val, y_train, y_val = train_test_split(\n X_train, y_train, test_size=0.3, random_state=42, stratify=y_train)", "_____no_output_____" ], [ "X_train.shape, X_val.shape, X_test.shape, y_train.shape, y_val.shape, y_test.shape", "_____no_output_____" ] ], [ [ "## Majority class baseline", "_____no_output_____" ], [ "Determine the majority class:", "_____no_output_____" ] ], [ [ "y_train.value_counts(normalize=True)", "_____no_output_____" ] ], [ [ "What if we guessed the majority class for every prediction?", "_____no_output_____" ] ], [ [ "majority_class = y_train.mode()[0]\ny_pred = [majority_class] * len(y_val)", "_____no_output_____" ] ], [ [ "#### [`sklearn.metrics.accuracy_score`](https://scikit-learn.org/stable/modules/generated/sklearn.metrics.accuracy_score.html)\n\nBaseline accuracy by guessing the majority class for every prediction:", "_____no_output_____" ] ], [ [ "from sklearn.metrics import accuracy_score\naccuracy_score(y_val, y_pred)", "_____no_output_____" ] ], [ [ "#### [`sklearn.metrics.roc_auc_score`](https://scikit-learn.org/stable/modules/generated/sklearn.metrics.roc_auc_score.html)\n\nBaseline \"ROC AUC\" score by guessing the majority class for every prediction:", "_____no_output_____" ] ], [ [ "from sklearn.metrics import roc_auc_score\nroc_auc_score(y_val, y_pred)", "_____no_output_____" ] ], [ [ "## Fast first models", "_____no_output_____" ], [ "### Ignore rows/columns with nulls", "_____no_output_____" ], [ "Does this dataset have nulls?", "_____no_output_____" ] ], [ [ "X_train.isnull().sum()", "_____no_output_____" ] ], [ [ "### Ignore nonnumeric features", "_____no_output_____" ], [ "Here are the numeric features:", "_____no_output_____" ] ], [ [ "X_train.describe(include='number')", "_____no_output_____" ] ], [ [ "Here are the nonnumeric features:", "_____no_output_____" ] ], [ [ "X_train.describe(exclude='number')", "_____no_output_____" ] ], [ [ "Just select the nonnumeric features:", "_____no_output_____" ] ], [ [ "X_train_numeric = X_train.select_dtypes('number')\nX_val_numeric = X_val.select_dtypes('number')", "_____no_output_____" ] ], [ [ "### Shallow trees are good for fast, first baselines, and to look for \"leakage\"", "_____no_output_____" ], [ "#### Shallow trees", "_____no_output_____" ], [ "After naive baselines, *Data Science for Business* suggests [\"decision stumps.\"](https://en.wikipedia.org/wiki/Decision_stump)\n\n> A slightly more complex alternative is a model that only considers a very small amount of feature information. ...\n\n> One example is to build a \"decision stump\"—a decision tree with only one internal node, the root node. A tree limited to one internal node simply means that the tree induction selects the single most informative feature to make a decision. In a well-known paper in machine learning, [Robert Holte (1993)](https://link.springer.com/article/10.1023/A:1022631118932) showed that ***decision stumps often produce quite good baseline performance*** ...\n\n> A decision stump is an example of the strategy of ***choosing the single most informative piece of information*** available and basing all decisions on it. In some cases most of the leverage may be coming from a single feature, and this method assesses whether and to what extent this is the case.\n\nTo fit a \"decision stump\" we could use a [`DecisionTreeClassifier`](http://scikit-learn.org/stable/modules/generated/sklearn.tree.DecisionTreeClassifier.html) model with parameter `max_depth=1`.\n\nIn this case, we'll let our tree grow a little deeper, and use the parameter `max_depth=2`\n\nIn the previous code cell, we selected only the numeric features, to avoid data wrangling and save time. For now, we'll use only the numeric features.", "_____no_output_____" ], [ "#### Looking for leakage", "_____no_output_____" ], [ "[Xavier Amatriain recommends,](https://www.quora.com/What-are-some-best-practices-for-training-machine-learning-models/answer/Xavier-Amatriain)\n\n\"Make sure your training features do not contain data from the “future” (aka time traveling). While this might be easy and obvious in some cases, it can get tricky. ... If your test metric becomes really good all of the sudden, ask yourself what you might be doing wrong. Chances are you are time travelling or overfitting in some way.\"", "_____no_output_____" ], [ "We can test this with the [UCI repository's Bank Marketing dataset](https://archive.ics.uci.edu/ml/datasets/Bank+Marketing). It has a feature which leaks information from the future and should be dropped:\n\n>11 - duration: last contact duration, in seconds (numeric). Important note: this attribute highly affects the output target (e.g., if duration=0 then y='no'). Yet, the duration is not known before a call is performed. Also, after the end of the call y is obviously known. Thus, this input ... should be discarded if the intention is to have a realistic predictive model.", "_____no_output_____" ], [ "#### Let's train a shallow tree basline\n\n... without dropping the leaky `duration` feature.", "_____no_output_____" ] ], [ [ "\nfrom sklearn.tree import DecisionTreeClassifier\n\ntree = DecisionTreeClassifier(max_depth=2)\ntree.fit(X_train_numeric,y_train)\ny_pred_proba = tree.predict_proba(X_val_numeric)[:,1]\nroc_auc_score(y_val, y_pred_proba)", "_____no_output_____" ] ], [ [ "Then we can visualize the tree to see which feature(s) were the \"most informative\":", "_____no_output_____" ] ], [ [ "import graphviz\nfrom sklearn.tree import export_graphviz\n\ndot_data = export_graphviz(tree, out_file=None, feature_names=X_train_numeric.columns, \n class_names=['No', 'Yes'], filled=True, impurity=False, proportion=True)\n\ngraphviz.Source(dot_data)", "_____no_output_____" ] ], [ [ "This baseline has a ROC AUC score above 0.85, and it uses the `duration` feature, as well as `nr.employed`, a \"social and economic context attribute\" for \"number of employees - quarterly indicator.\"", "_____no_output_____" ], [ "#### Let's drop the `duration` feature", "_____no_output_____" ] ], [ [ "\nX_train = X_train.drop(columns='duration')\nX_val = X_val.drop(columns='duration')\nX_test = X_test.drop(columns='duration')\n\nX_train_numeric = X_train.select_dtypes('number')\nX_val_numeric = X_val_numeric.select_dtypes('number')", "_____no_output_____" ] ], [ [ "When the `duration` feature is dropped, then the ROC AUC score drops. Which is what we expect, it's not a bad thing in this situation!", "_____no_output_____" ] ], [ [ "tree = DecisionTreeClassifier(max_depth=2)\ntree.fit(X_train_numeric,y_train)\ny_pred_proba = tree.predict_proba(X_val_numeric)[:,1]\nroc_auc_score(y_val, y_pred_proba)", "_____no_output_____" ], [ "dot_data = export_graphviz(tree, out_file=None, feature_names=X_train_numeric.columns, \n class_names=['No', 'Yes'], filled=True, impurity=False, proportion=True)\n\ngraphviz.Source(dot_data)", "_____no_output_____" ] ], [ [ "### Logistic Regression\n\nLogistic Regression is another great option for fast, first baselines!", "_____no_output_____" ] ], [ [ "from sklearn.linear_model import LogisticRegression\n\nmodel = LogisticRegression(solver='lbfgs', max_iter=1000)\nmodel.fit(X_train_numeric,y_train)\ny_pred_proba = model.predict_proba(X_val_numeric)[:,1]\nroc_auc_score(y_val,y_pred_proba)", "_____no_output_____" ] ], [ [ "### With Scaler\nhttps://scikit-learn.org/stable/modules/preprocessing.html", "_____no_output_____" ] ], [ [ "import warnings\nfrom sklearn.exceptions import DataConversionWarning\nwarnings.filterwarnings(action='ignore', category=DataConversionWarning)", "_____no_output_____" ], [ "from sklearn.preprocessing import StandardScaler\n\nscaler = StandardScaler()\nX_train_scaled = scaler.fit_transform(X_train_numeric)\nX_val_scaled = scaler.transform(X_val_numeric)\n\nmodel = LogisticRegression(solver='lbfgs', max_iter=1000)\nmodel.fit(X_train_scaled, y_train)\ny_pred_proba = model.predict_proba(X_val_scaled)[:,1]\nroc_auc_score(y_val,y_pred_proba)", "_____no_output_____" ] ], [ [ "### Same, as a pipeline", "_____no_output_____" ] ], [ [ "\nfrom sklearn.pipeline import make_pipeline\npipeline = make_pipeline(\n StandardScaler(),\n LogisticRegression(solver='lbfgs',max_iter=1000))\n\npipeline.fit(X_train_numeric,y_train)\ny_pred_proba = pipeline.predict_proba(X_val_numeric)[:,1]", "_____no_output_____" ] ], [ [ "### Encode \"low cardinality\" categoricals", "_____no_output_____" ], [ "[Cardinality](https://simple.wikipedia.org/wiki/Cardinality) means the number of unique values that a feature has:\n> In mathematics, the cardinality of a set means the number of its elements. For example, the set A = {2, 4, 6} contains 3 elements, and therefore A has a cardinality of 3. \n\nOne-hot encoding adds a dimension for each unique value of each categorical feature. So, it may not be a good choice for \"high cardinality\" categoricals that have dozens, hundreds, or thousands of unique values. \n\nIn this dataset, all the categoricals seem to be \"low cardinality\", so we can use one-hot encoding.", "_____no_output_____" ] ], [ [ "!pip install category_encoders\n\nimport category_encoders as ce\npipeline = make_pipeline(\n ce.OneHotEncoder(use_cat_names=True),\n StandardScaler(),\n LogisticRegression(solver='lbfgs', max_iter=1000))\n\npipeline.fit(X_train,y_train)\ny_pred_proba = pipeline.predict_proba(X_val)[:,1]\nroc_auc_score(y_val,y_pred_proba)", "Collecting category_encoders\n\u001b[?25l Downloading https://files.pythonhosted.org/packages/6e/a1/f7a22f144f33be78afeb06bfa78478e8284a64263a3c09b1ef54e673841e/category_encoders-2.0.0-py2.py3-none-any.whl (87kB)\n\u001b[K |████████████████████████████████| 92kB 5.9MB/s \n\u001b[?25hRequirement already satisfied: numpy>=1.11.3 in /usr/local/lib/python3.6/dist-packages (from category_encoders) (1.16.3)\nRequirement already satisfied: scikit-learn>=0.20.0 in /usr/local/lib/python3.6/dist-packages (from category_encoders) (0.20.3)\nRequirement already satisfied: statsmodels>=0.6.1 in /usr/local/lib/python3.6/dist-packages (from category_encoders) (0.9.0)\nRequirement already satisfied: patsy>=0.4.1 in /usr/local/lib/python3.6/dist-packages (from category_encoders) (0.5.1)\nRequirement already satisfied: scipy>=0.19.0 in /usr/local/lib/python3.6/dist-packages (from category_encoders) (1.2.1)\nRequirement already satisfied: pandas>=0.21.1 in /usr/local/lib/python3.6/dist-packages (from category_encoders) (0.24.2)\nRequirement already satisfied: six in /usr/local/lib/python3.6/dist-packages (from patsy>=0.4.1->category_encoders) (1.12.0)\nRequirement already satisfied: python-dateutil>=2.5.0 in /usr/local/lib/python3.6/dist-packages (from pandas>=0.21.1->category_encoders) (2.5.3)\nRequirement already satisfied: pytz>=2011k in /usr/local/lib/python3.6/dist-packages (from pandas>=0.21.1->category_encoders) (2018.9)\nInstalling collected packages: category-encoders\nSuccessfully installed category-encoders-2.0.0\n" ] ], [ [ "#### Install the [Category Encoders](https://github.com/scikit-learn-contrib/categorical-encoding) library\n\nIf you're running on Google Colab:\n\n```\n!pip install category_encoders\n```\n\nIf you're running locally with Anaconda:\n\n```\n!conda install -c conda-forge category_encoders\n```", "_____no_output_____" ] ], [ [ "", "_____no_output_____" ] ], [ [ "# Baseline with cross-validation + independent test set\nA complete example, as an alternative to Train/Validate/Test\n\n\n#### scikit-learn documentation\n- [`sklearn.model_selection.cross_val_score`](https://scikit-learn.org/stable/modules/generated/sklearn.model_selection.cross_val_score.html)\n- [ The `scoring` parameter: defining model evaluation rules](https://scikit-learn.org/stable/modules/model_evaluation.html#the-scoring-parameter-defining-model-evaluation-rules)", "_____no_output_____" ] ], [ [ "# Imports\n%matplotlib inline\nimport warnings\nimport category_encoders as ce\nimport matplotlib.pyplot as plt\nimport pandas as pd\nfrom sklearn.linear_model import LogisticRegression\nfrom sklearn.model_selection import cross_val_score\nfrom sklearn.pipeline import make_pipeline\nfrom sklearn.exceptions import DataConversionWarning\nfrom sklearn.preprocessing import StandardScaler\nwarnings.filterwarnings(action='ignore', category=DataConversionWarning)\n\n# Load data\nbank = pd.read_csv('bank-additional/bank-additional-full.csv', sep=';')\n\n# Assign to X, y\nX = bank.drop(columns='y')\ny = bank['y'] == 'yes'\n\n# Drop leaky & random features\nX = X.drop(columns='duration')\n\n# Split Train, Test\nX_train, X_test, y_train, y_test = train_test_split(\n X, y, test_size=0.2, random_state=42, stratify=y)\n\n# Make pipeline\npipeline = make_pipeline(\n ce.OneHotEncoder(use_cat_names=True), \n StandardScaler(), \n LogisticRegression(solver='lbfgs', max_iter=1000)\n)\n\n# Cross-validate with training data\nscores = cross_val_score(pipeline, X_train, y_train, scoring='roc_auc', cv=10, n_jobs=-1, verbose=10)", "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n[Parallel(n_jobs=-1)]: Done 1 tasks | elapsed: 3.9s\n[Parallel(n_jobs=-1)]: Done 4 tasks | elapsed: 6.7s\n[Parallel(n_jobs=-1)]: Done 10 out of 10 | elapsed: 12.8s finished\n" ] ], [ [ "This is the baseline score that more sophisticated models must beat. ", "_____no_output_____" ] ], [ [ "print('Cross-Validation ROC AUC scores:', scores)\nprint('Average:', scores.mean())", "Cross-Validation ROC AUC scores: [0.82042478 0.79227573 0.79162088 0.762977 0.78662274 0.78877613\n 0.76414311 0.79607284 0.80670867 0.77968487]\nAverage: 0.7889306746390174\n" ] ], [ [ "Is more effort justified? It depends. The blogpost [\"Always start with a stupid model\"](https://blog.insightdatascience.com/always-start-with-a-stupid-model-no-exceptions-3a22314b9aaa) explains,\n\n> Here is a very common story: a team wants to implement a model to predict something like the probability of a user clicking an ad. They start with a logistic regression and quickly (after some minor tuning) reach 90% accuracy.\n\n> From there, the question is: Should the team focus on getting the accuracy up to 95%, or should they solve other problems 90% of the way?\n\n> ***If a baseline does well, then you’ve saved yourself the headache of setting up a more complex model. If it does poorly, the kind of mistakes it makes are very instructive*** ...\n\nSo what else can we learn from this baseline? \n\n[\"Always start with a stupid model\"](https://blog.insightdatascience.com/always-start-with-a-stupid-model-no-exceptions-3a22314b9aaa) suggests to look at\n\n> **What type of signal your model picks up on.** Most baselines will allow you to extract ***feature importances***, revealing which aspects of the input are most predictive. Analyzing feature importance is a great way to realize how your model is making decisions, and what it might be missing.\n\nWe can do that:", "_____no_output_____" ] ], [ [ "# (Re)fit on training data\npipeline.fit(X_train, y_train)\n\n# Visualize coefficients\nplt.figure(figsize=(10,30))\nplt.title('Coefficients')\ncoefficients = pipeline.named_steps['logisticregression'].coef_[0]\nfeature_names = pipeline.named_steps['onehotencoder'].transform(X_train).columns\npd.Series(coefficients, feature_names).sort_values().plot.barh(color='gray');", "_____no_output_____" ] ], [ [ "[The post](https://blog.insightdatascience.com/always-start-with-a-stupid-model-no-exceptions-3a22314b9aaa) also recommends we consider, \n\n> **What signal your model is missing.** If there is a certain aspect of the data that seems intuitively important but that your model is ignoring, ***a good next step is to engineer a feature*** or pick a different model that could better leverage this particular aspect of your data.", "_____no_output_____" ], [ "### Look at your data (you still need to do it!)\n\nCautionary tales\n- [Exploring the ChestXray14 dataset: problems](https://lukeoakdenrayner.wordpress.com/2017/12/18/the-chestxray14-dataset-problems/)\n- [The Quartz guide to bad data](https://github.com/Quartz/bad-data-guide)\n\nIncomplete list of issues to address\n- Categoricals (text, dates/times, high cardinality)\n- Feature Engineering (extraction, interaction, transformations)\n- Missing Values\n- Outliers", "_____no_output_____" ], [ "# ASSIGNMENT options\n\n- **Replicate the lesson code.** [Do it \"the hard way\" or with the \"Benjamin Franklin method.\"](https://docs.google.com/document/d/1ubOw9B3Hfip27hF2ZFnW3a3z9xAgrUDRReOEo-FHCVs/edit)\n- Apply the lesson to other datasets you've worked with before, and compare results.\n- Iterate and improve your **Bank Marketing** model. Engineer new features.\n- Get **weather** data for your own area and calculate both baselines. _\"One (persistence) predicts that the weather tomorrow is going to be whatever it was today. The other (climatology) predicts whatever the average historical weather has been on this day from prior years.\"_ What is the mean absolute error for each baseline? What if you average the two together? \n- [This example from scikit-learn documentation](https://scikit-learn.org/stable/auto_examples/compose/plot_column_transformer_mixed_types.html) demonstrates its improved `OneHotEncoder` and new `ColumnTransformer` objects, which can replace functionality from [third-party libraries](https://github.com/scikit-learn-contrib) like category_encoders and sklearn-pandas. Adapt this example, which uses Titanic data, to work with Bank Marketing or another dataset.\n- When would this notebook's pipelines fail? How could you fix them? Add more [preprocessing](https://scikit-learn.org/stable/modules/preprocessing.html) and [imputation](https://scikit-learn.org/stable/modules/impute.html) to your [pipelines](https://scikit-learn.org/stable/modules/compose.html) with scikit-learn.", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown" ]

[ [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown", "markdown", "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown", "markdown", "markdown", "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown", "markdown" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "# Kubeflow pipelines\n\n**Learning Objectives:**\n 1. Learn how to deploy a Kubeflow cluster on GCP\n 1. Learn how to create a experiment in Kubeflow\n 1. Learn how to package you code into a Kubeflow pipeline\n 1. Learn how to run a Kubeflow pipeline in a repeatable and traceable way\n\n\n## Introduction\n\nIn this notebook, we will first setup a Kubeflow cluster on GCP.\nThen, we will create a Kubeflow experiment and a Kubflow pipeline from our taxifare machine learning code. At last, we will run the pipeline on the Kubeflow cluster, providing us with a reproducible and traceable way to execute machine learning code.", "_____no_output_____" ] ], [ [ "!pip3 install --user kfp --upgrade", "Requirement already satisfied: kfp in /home/jupyter/.local/lib/python3.7/site-packages (1.7.2)\nRequirement already satisfied: google-auth<2,>=1.6.1 in /opt/conda/lib/python3.7/site-packages (from kfp) (1.34.0)\nRequirement already satisfied: jsonschema<4,>=3.0.1 in /opt/conda/lib/python3.7/site-packages (from kfp) (3.2.0)\nRequirement already satisfied: PyYAML<6,>=5.3 in /opt/conda/lib/python3.7/site-packages (from kfp) (5.4.1)\nRequirement already satisfied: kfp-server-api<2.0.0,>=1.1.2 in /home/jupyter/.local/lib/python3.7/site-packages (from kfp) (1.7.0)\nRequirement already satisfied: pydantic<2,>=1.8.2 in /opt/conda/lib/python3.7/site-packages (from kfp) (1.8.2)\nRequirement already satisfied: kfp-pipeline-spec<0.2.0,>=0.1.9 in /home/jupyter/.local/lib/python3.7/site-packages (from kfp) (0.1.9)\nRequirement already satisfied: cloudpickle<2,>=1.3.0 in /opt/conda/lib/python3.7/site-packages (from kfp) (1.6.0)\nRequirement already satisfied: requests-toolbelt<1,>=0.8.0 in /home/jupyter/.local/lib/python3.7/site-packages (from kfp) (0.9.1)\nRequirement already satisfied: kubernetes<13,>=8.0.0 in /home/jupyter/.local/lib/python3.7/site-packages (from kfp) (12.0.1)\nRequirement already satisfied: protobuf<4,>=3.13.0 in /opt/conda/lib/python3.7/site-packages (from kfp) (3.16.0)\nRequirement already satisfied: Deprecated<2,>=1.2.7 in /home/jupyter/.local/lib/python3.7/site-packages (from kfp) (1.2.12)\nRequirement already satisfied: tabulate<1,>=0.8.6 in /home/jupyter/.local/lib/python3.7/site-packages (from kfp) (0.8.9)\nRequirement already satisfied: google-cloud-storage<2,>=1.20.0 in /opt/conda/lib/python3.7/site-packages (from kfp) (1.41.1)\nRequirement already satisfied: google-api-python-client<2,>=1.7.8 in /home/jupyter/.local/lib/python3.7/site-packages (from kfp) (1.12.8)\nRequirement already satisfied: click<8,>=7.1.1 in /home/jupyter/.local/lib/python3.7/site-packages (from kfp) (7.1.2)\nRequirement already satisfied: fire<1,>=0.3.1 in /home/jupyter/.local/lib/python3.7/site-packages (from kfp) (0.4.0)\nRequirement already satisfied: absl-py<=0.11,>=0.9 in /home/jupyter/.local/lib/python3.7/site-packages (from kfp) (0.11.0)\nRequirement already satisfied: docstring-parser<1,>=0.7.3 in /home/jupyter/.local/lib/python3.7/site-packages (from kfp) (0.10)\nRequirement already satisfied: strip-hints<1,>=0.1.8 in /home/jupyter/.local/lib/python3.7/site-packages (from kfp) (0.1.10)\nRequirement already satisfied: six in /opt/conda/lib/python3.7/site-packages (from absl-py<=0.11,>=0.9->kfp) (1.16.0)\nRequirement already satisfied: wrapt<2,>=1.10 in /opt/conda/lib/python3.7/site-packages (from Deprecated<2,>=1.2.7->kfp) (1.12.1)\nRequirement already satisfied: termcolor in /opt/conda/lib/python3.7/site-packages (from fire<1,>=0.3.1->kfp) (1.1.0)\nRequirement already satisfied: httplib2<1dev,>=0.15.0 in /opt/conda/lib/python3.7/site-packages (from google-api-python-client<2,>=1.7.8->kfp) (0.19.1)\nRequirement already satisfied: google-auth-httplib2>=0.0.3 in /opt/conda/lib/python3.7/site-packages (from google-api-python-client<2,>=1.7.8->kfp) (0.1.0)\nRequirement already satisfied: google-api-core<2dev,>=1.21.0 in /opt/conda/lib/python3.7/site-packages (from google-api-python-client<2,>=1.7.8->kfp) (1.31.1)\nRequirement already satisfied: uritemplate<4dev,>=3.0.0 in /opt/conda/lib/python3.7/site-packages (from google-api-python-client<2,>=1.7.8->kfp) (3.0.1)\nRequirement already satisfied: requests<3.0.0dev,>=2.18.0 in /opt/conda/lib/python3.7/site-packages (from google-api-core<2dev,>=1.21.0->google-api-python-client<2,>=1.7.8->kfp) (2.25.1)\nRequirement already satisfied: googleapis-common-protos<2.0dev,>=1.6.0 in /opt/conda/lib/python3.7/site-packages (from google-api-core<2dev,>=1.21.0->google-api-python-client<2,>=1.7.8->kfp) (1.53.0)\nRequirement already satisfied: setuptools>=40.3.0 in /opt/conda/lib/python3.7/site-packages (from google-api-core<2dev,>=1.21.0->google-api-python-client<2,>=1.7.8->kfp) (49.6.0.post20210108)\nRequirement already satisfied: packaging>=14.3 in /opt/conda/lib/python3.7/site-packages (from google-api-core<2dev,>=1.21.0->google-api-python-client<2,>=1.7.8->kfp) (21.0)\nRequirement already satisfied: pytz in /opt/conda/lib/python3.7/site-packages (from google-api-core<2dev,>=1.21.0->google-api-python-client<2,>=1.7.8->kfp) (2021.1)\nRequirement already satisfied: rsa<5,>=3.1.4 in /opt/conda/lib/python3.7/site-packages (from google-auth<2,>=1.6.1->kfp) (4.7.2)\nRequirement already satisfied: pyasn1-modules>=0.2.1 in /opt/conda/lib/python3.7/site-packages (from google-auth<2,>=1.6.1->kfp) (0.2.7)\nRequirement already satisfied: cachetools<5.0,>=2.0.0 in /opt/conda/lib/python3.7/site-packages (from google-auth<2,>=1.6.1->kfp) (4.2.2)\nRequirement already satisfied: google-resumable-media<3.0dev,>=1.3.0 in /opt/conda/lib/python3.7/site-packages (from google-cloud-storage<2,>=1.20.0->kfp) (1.3.2)\nRequirement already satisfied: google-cloud-core<3.0dev,>=1.6.0 in /opt/conda/lib/python3.7/site-packages (from google-cloud-storage<2,>=1.20.0->kfp) (1.7.2)\nRequirement already satisfied: google-crc32c<2.0dev,>=1.0 in /opt/conda/lib/python3.7/site-packages (from google-resumable-media<3.0dev,>=1.3.0->google-cloud-storage<2,>=1.20.0->kfp) (1.1.2)\nRequirement already satisfied: cffi>=1.0.0 in /opt/conda/lib/python3.7/site-packages (from google-crc32c<2.0dev,>=1.0->google-resumable-media<3.0dev,>=1.3.0->google-cloud-storage<2,>=1.20.0->kfp) (1.14.6)\nRequirement already satisfied: pycparser in /opt/conda/lib/python3.7/site-packages (from cffi>=1.0.0->google-crc32c<2.0dev,>=1.0->google-resumable-media<3.0dev,>=1.3.0->google-cloud-storage<2,>=1.20.0->kfp) (2.20)\nRequirement already satisfied: pyparsing<3,>=2.4.2 in /opt/conda/lib/python3.7/site-packages (from httplib2<1dev,>=0.15.0->google-api-python-client<2,>=1.7.8->kfp) (2.4.7)\nRequirement already satisfied: pyrsistent>=0.14.0 in /opt/conda/lib/python3.7/site-packages (from jsonschema<4,>=3.0.1->kfp) (0.17.3)\nRequirement already satisfied: importlib-metadata in /opt/conda/lib/python3.7/site-packages (from jsonschema<4,>=3.0.1->kfp) (4.6.3)\nRequirement already satisfied: attrs>=17.4.0 in /opt/conda/lib/python3.7/site-packages (from jsonschema<4,>=3.0.1->kfp) (21.2.0)\nRequirement already satisfied: python-dateutil in /opt/conda/lib/python3.7/site-packages (from kfp-server-api<2.0.0,>=1.1.2->kfp) (2.8.2)\nRequirement already satisfied: urllib3>=1.15 in /opt/conda/lib/python3.7/site-packages (from kfp-server-api<2.0.0,>=1.1.2->kfp) (1.26.6)\nRequirement already satisfied: certifi in /opt/conda/lib/python3.7/site-packages (from kfp-server-api<2.0.0,>=1.1.2->kfp) (2021.5.30)\nRequirement already satisfied: requests-oauthlib in /opt/conda/lib/python3.7/site-packages (from kubernetes<13,>=8.0.0->kfp) (1.3.0)\nRequirement already satisfied: websocket-client!=0.40.0,!=0.41.*,!=0.42.*,>=0.32.0 in /opt/conda/lib/python3.7/site-packages (from kubernetes<13,>=8.0.0->kfp) (0.57.0)\nRequirement already satisfied: pyasn1<0.5.0,>=0.4.6 in /opt/conda/lib/python3.7/site-packages (from pyasn1-modules>=0.2.1->google-auth<2,>=1.6.1->kfp) (0.4.8)\nRequirement already satisfied: typing-extensions>=3.7.4.3 in /opt/conda/lib/python3.7/site-packages (from pydantic<2,>=1.8.2->kfp) (3.10.0.0)\nRequirement already satisfied: idna<3,>=2.5 in /opt/conda/lib/python3.7/site-packages (from requests<3.0.0dev,>=2.18.0->google-api-core<2dev,>=1.21.0->google-api-python-client<2,>=1.7.8->kfp) (2.10)\nRequirement already satisfied: chardet<5,>=3.0.2 in /opt/conda/lib/python3.7/site-packages (from requests<3.0.0dev,>=2.18.0->google-api-core<2dev,>=1.21.0->google-api-python-client<2,>=1.7.8->kfp) (4.0.0)\nRequirement already satisfied: wheel in /opt/conda/lib/python3.7/site-packages (from strip-hints<1,>=0.1.8->kfp) (0.36.2)\nRequirement already satisfied: zipp>=0.5 in /opt/conda/lib/python3.7/site-packages (from importlib-metadata->jsonschema<4,>=3.0.1->kfp) (3.5.0)\nRequirement already satisfied: oauthlib>=3.0.0 in /opt/conda/lib/python3.7/site-packages (from requests-oauthlib->kubernetes<13,>=8.0.0->kfp) (3.1.1)\n" ] ], [ [ "### Restart the kernel\n\nAfter you install the additional packages, you need to restart the notebook kernel so it can find the packages.", "_____no_output_____" ], [ "### Import libraries and define constants", "_____no_output_____" ] ], [ [ "from os import path\n\nimport kfp\nimport kfp.compiler as compiler\nimport kfp.components as comp\nimport kfp.dsl as dsl\nimport kfp.gcp as gcp\nimport kfp.notebook", "_____no_output_____" ] ], [ [ "## Setup a Kubeflow cluster on GCP", "_____no_output_____" ], [ "**TODO 1**", "_____no_output_____" ], [ "To deploy a [Kubeflow](https://www.kubeflow.org/) cluster\nin your GCP project, use the [AI Platform pipelines](https://console.cloud.google.com/ai-platform/pipelines):\n\n1. Go to [AI Platform Pipelines](https://console.cloud.google.com/ai-platform/pipelines) in the GCP Console.\n1. Create a new instance\n2. Hit \"Configure\"\n3. Check the box \"Allow access to the following Cloud APIs\"\n1. Hit \"Create Cluster\"\n4. Hit \"Deploy\"\n\nWhen the cluster is ready, go back to the AI Platform pipelines page and click on \"SETTINGS\" entry for your cluster.\nThis will bring up a pop up with code snippets on how to access the cluster \nprogrammatically. \n\nCopy the \"host\" entry and set the \"HOST\" variable below with that.\n", "_____no_output_____" ] ], [ [ "HOST = \"\" # TODO: fill in the HOST information for the cluster", "_____no_output_____" ] ], [ [ "### Authenticate your KFP cluster with a Kubernetes secret\n\nIf you run pipelines that requires calling any GCP services, you need to set the application default credential to a pipeline step by mounting the proper GCP service account token as a Kubernetes secret.\n\nFirst point your kubectl current context to your cluster. Go back to your [Kubeflow cluster dashboard](https://console.cloud.google.com/ai-platform/pipelines/clusters) or navigate to `Navigation menu > AI Platform > Pipelines` and look to see the cluster name, zone and namespace for the pipeline you deployed above. It's likely called `cluster-1` if this is the first AI Pipelines you've created. ", "_____no_output_____" ] ], [ [ "# Change below if necessary\nPROJECT = !gcloud config get-value project # noqa: E999\nPROJECT = PROJECT[0]\nBUCKET = PROJECT # change if needed\nCLUSTER = \"cluster-1\" # change if needed\nZONE = \"us-central1-a\" # change if needed\nNAMESPACE = \"default\" # change if needed\n\n%env PROJECT=$PROJECT\n%env CLUSTER=$CLUSTER\n%env ZONE=$ZONE\n%env NAMESPACE=$NAMESPACE", "env: PROJECT=dsparing-sandbox\nenv: CLUSTER=cluster-1\nenv: ZONE=us-central1-a\nenv: NAMESPACE=default\n" ], [ "# Configure kubectl to connect with the cluster\n!gcloud container clusters get-credentials \"$CLUSTER\" --zone \"$ZONE\" --project \"$PROJECT\"", "Fetching cluster endpoint and auth data.\nkubeconfig entry generated for cluster-1.\n" ] ], [ [ "We'll create a service account called `kfpdemo` with the necessary IAM permissions for our cluster secret. We'll give this service account permissions for any GCP services it might need. This `taxifare` pipeline needs access to Cloud Storage, so we'll give it the `storage.admin` role and `ml.admin`. Open a Cloud Shell and copy/paste this code in the terminal there.\n\n```bash\nPROJECT=$(gcloud config get-value project)\n\n# Create service account\ngcloud iam service-accounts create kfpdemo \\\n --display-name kfpdemo --project $PROJECT\n\n# Grant permissions to the service account by binding roles\ngcloud projects add-iam-policy-binding $PROJECT \\\n --member=serviceAccount:kfpdemo@$PROJECT.iam.gserviceaccount.com \\\n --role=roles/storage.admin\n \ngcloud projects add-iam-policy-binding $PROJECT \\\n --member=serviceAccount:kfpdemo@$PROJECT.iam.gserviceaccount.com \\\n --role=roles/ml.admin \n```", "_____no_output_____" ], [ "Then, we'll create and download a key for this service account and store the service account credential as a Kubernetes secret called `user-gcp-sa` in the cluster.", "_____no_output_____" ] ], [ [ "%%bash\ngcloud iam service-accounts keys create application_default_credentials.json \\\n --iam-account kfpdemo@$PROJECT.iam.gserviceaccount.com", "_____no_output_____" ], [ "# Check that the key was downloaded correctly.\n!ls application_default_credentials.json", "application_default_credentials.json\n" ], [ "# Create a k8s secret. If already exists, override.\n!kubectl create secret generic user-gcp-sa \\\n --from-file=user-gcp-sa.json=application_default_credentials.json \\\n -n $NAMESPACE --dry-run=client -o yaml | kubectl apply -f -", "secret/user-gcp-sa configured\n" ] ], [ [ "## Create an experiment", "_____no_output_____" ], [ "**TODO 2**", "_____no_output_____" ], [ "We will start by creating a Kubeflow client to pilot the Kubeflow cluster:", "_____no_output_____" ] ], [ [ "client = kfp.Client(host=HOST)", "_____no_output_____" ] ], [ [ "Let's look at the experiments that are running on this cluster. Since you just launched it, you should see only a single \"Default\" experiment:", "_____no_output_____" ] ], [ [ "client.list_experiments()", "_____no_output_____" ] ], [ [ "Now let's create a 'taxifare' experiment where we could look at all the various runs of our taxifare pipeline:", "_____no_output_____" ] ], [ [ "exp = client.create_experiment(name=\"taxifare\")", "_____no_output_____" ] ], [ [ "Let's make sure the experiment has been created correctly:", "_____no_output_____" ] ], [ [ "client.list_experiments()", "_____no_output_____" ] ], [ [ "## Packaging your code into Kubeflow components", "_____no_output_____" ], [ "We have packaged our taxifare ml pipeline into three components:\n* `./components/bq2gcs` that creates the training and evaluation data from BigQuery and exports it to GCS\n* `./components/trainjob` that launches the training container on AI-platform and exports the model\n* `./components/deploymodel` that deploys the trained model to AI-platform as a REST API\n\nEach of these components has been wrapped into a Docker container, in the same way we did with the taxifare training code in the previous lab.\n\nIf you inspect the code in these folders, you'll notice that the `main.py` or `main.sh` files contain the code we previously executed in the notebooks (loading the data to GCS from BQ, or launching a training job to AI-platform, etc.). The last line in the `Dockerfile` tells you that these files are executed when the container is run. \nSo we just packaged our ml code into light container images for reproducibility. \n\nWe have made it simple for you to build the container images and push them to the Google Cloud image registry gcr.io in your project:", "_____no_output_____" ] ], [ [ "# Builds the taxifare trainer container in case you skipped the optional part\n# of lab 1\n!taxifare/scripts/build.sh", "Sending build context to Docker daemon 157.2kB\nStep 1/4 : FROM gcr.io/deeplearning-platform-release/tf2-cpu.2-5:latest\n ---> 19875ee1008d\nStep 2/4 : COPY . /code\n ---> Using cache\n ---> 9c12e76129be\nStep 3/4 : WORKDIR /code\n ---> Using cache\n ---> 0b77723599c9\nStep 4/4 : ENTRYPOINT [\"python3\", \"-m\", \"trainer.task\"]\n ---> Using cache\n ---> ec727c7b2e63\nSuccessfully built ec727c7b2e63\nSuccessfully tagged gcr.io/dsparing-sandbox/taxifare_training_container:latest\n" ], [ "# Pushes the taxifare trainer container to gcr/io\n!taxifare/scripts/push.sh", "Using default tag: latest\nThe push refers to repository [gcr.io/dsparing-sandbox/taxifare_training_container]\n\n\u001b[1B5a90caa8: Preparing \n\u001b[1B0326dd85: Preparing \n\u001b[1Bbe2a94be: Preparing \n\u001b[1B2667b401: Preparing \n\u001b[1B97d991c7: Preparing \n\u001b[1Bbdf9b557: Preparing \n\u001b[1Bdbc2b748: Preparing \n\u001b[1Bb8f29c2e: Preparing \n\u001b[1B7b2f7486: Preparing \n\u001b[1B506c54ba: Preparing \n\u001b[1B3dc8d38f: Preparing \n\u001b[1Bd7ce97e4: Preparing \n\u001b[1B97e5c777: Preparing \n\u001b[1B5dfd94f2: Preparing \n\u001b[1Bafebd1ec: Preparing \n\u001b[1Bbf18a086: Preparing \n\u001b[1B7318f223: Preparing \n\u001b[1B3d35a813: Preparing \n\u001b[1Ba1af4c10: Preparing \n\u001b[1B9b09744f: Layer already exists \u001b[18A\u001b[2K\u001b[14A\u001b[2K\u001b[10A\u001b[2K\u001b[8A\u001b[2K\u001b[3A\u001b[2K\u001b[2A\u001b[2Klatest: digest: sha256:90f6400640d2e3c16d7e3d5bbeb4974ef1e9c99155fa2a67032608bab735d002 size: 4507\n" ], [ "# Builds the KF component containers and push them to gcr/io\n!cd pipelines && make components", "make[1]: Entering directory '/home/jupyter/asl-ml-immersion/notebooks/building_production_ml_systems/solutions/pipelines/components/bq2gcs'\nrm: cannot remove './venv': No such file or directory\nOK\nSending build context to Docker daemon 21.5kB\nStep 1/6 : FROM google/cloud-sdk:latest\n ---> 915a516535e8\nStep 2/6 : RUN apt-get update && apt-get install --yes python3-pip\n ---> Using cache\n ---> 0f653294e07c\nStep 3/6 : COPY . /code\n ---> Using cache\n ---> 7d8f8d185c30\nStep 4/6 : WORKDIR /code\n ---> Using cache\n ---> 20d39822bdb4\nStep 5/6 : RUN pip3 install google-cloud-bigquery\n ---> Using cache\n ---> a1baf2091090\nStep 6/6 : ENTRYPOINT [\"python3\", \"./main.py\"]\n ---> Using cache\n ---> 05e9191c9619\nSuccessfully built 05e9191c9619\nSuccessfully tagged gcr.io/dsparing-sandbox/taxifare-bq2gcs:latest\nUsing default tag: latest\nThe push refers to repository [gcr.io/dsparing-sandbox/taxifare-bq2gcs]\n\n\u001b[1B038e5a12: Preparing \n\u001b[1B7ee6f1b4: Preparing \n\u001b[1Bced31aad: Preparing \n\u001b[1Be268b455: Preparing \n\u001b[1B15a7c280: Preparing \n\u001b[1B9ae3a881: Preparing \n\u001b[1B24ad8c63: Preparing \n\u001b[1Bd95c5384: Preparing \n\u001b[1Bd7a1159c: Preparing \n\u001b[1Bd1217615: Preparing \n\u001b[1Bc1bc2645: Layer already exists \u001b[6A\u001b[2K\u001b[5A\u001b[2K\u001b[1A\u001b[2Klatest: digest: sha256:2b9f25d58c03019983d785b6fd7910e61d886d9d84e96781aade0ae05a50e020 size: 2633\nmake[1]: Leaving directory '/home/jupyter/asl-ml-immersion/notebooks/building_production_ml_systems/solutions/pipelines/components/bq2gcs'\nmake[1]: Entering directory '/home/jupyter/asl-ml-immersion/notebooks/building_production_ml_systems/solutions/pipelines/components/trainjob'\nrm: cannot remove './venv': No such file or directory\nOK\nSending build context to Docker daemon 14.85kB\nStep 1/5 : FROM google/cloud-sdk:latest\n ---> 915a516535e8\nStep 2/5 : COPY . /code\n ---> Using cache\n ---> f598ddbd44a2\nStep 3/5 : WORKDIR /code\n ---> Using cache\n ---> 79f9d21c1bcf\nStep 4/5 : RUN pip3 install cloudml-hypertune\n ---> Using cache\n ---> 9edc5e05ae65\nStep 5/5 : ENTRYPOINT [\"./main.sh\"]\n ---> Using cache\n ---> ae939b43e795\nSuccessfully built ae939b43e795\nSuccessfully tagged gcr.io/dsparing-sandbox/taxifare-trainjob:latest\nUsing default tag: latest\nThe push refers to repository [gcr.io/dsparing-sandbox/taxifare-trainjob]\n\n\u001b[1B9e8dbe19: Preparing \n\u001b[1Bf0a19019: Preparing \n\u001b[1Be268b455: Preparing \n\u001b[1B15a7c280: Preparing \n\u001b[1B9ae3a881: Preparing \n\u001b[1B24ad8c63: Preparing \n\u001b[1Bd95c5384: Preparing \n\u001b[1Bd7a1159c: Preparing \n\u001b[1Bd1217615: Preparing \n\u001b[2Bd1217615: Layer already exists \u001b[6A\u001b[2K\u001b[5A\u001b[2Klatest: digest: sha256:4d28b7a097e81f911782a0101b395b2c3d56f83b7a55bddd8a3a83d2ab3e18b6 size: 2420\nmake[1]: Leaving directory '/home/jupyter/asl-ml-immersion/notebooks/building_production_ml_systems/solutions/pipelines/components/trainjob'\nmake[1]: Entering directory '/home/jupyter/asl-ml-immersion/notebooks/building_production_ml_systems/solutions/pipelines/components/deploymodel'\nrm: cannot remove './venv': No such file or directory\nOK\nSending build context to Docker daemon 14.85kB\nStep 1/4 : FROM google/cloud-sdk:latest\n ---> 915a516535e8\nStep 2/4 : COPY . /code\n ---> Using cache\n ---> 8ce1e25d9ba8\nStep 3/4 : WORKDIR /code\n ---> Using cache\n ---> 9b585070839b\nStep 4/4 : ENTRYPOINT [\"./main.sh\"]\n ---> Using cache\n ---> 0559c7f14028\nSuccessfully built 0559c7f14028\nSuccessfully tagged gcr.io/dsparing-sandbox/taxifare-deploymodel:latest\nUsing default tag: latest\nThe push refers to repository [gcr.io/dsparing-sandbox/taxifare-deploymodel]\n\n\u001b[1B539f9c38: Preparing \n\u001b[1Be268b455: Preparing \n\u001b[1B15a7c280: Preparing \n\u001b[1B9ae3a881: Preparing \n\u001b[1B24ad8c63: Preparing \n\u001b[1Bd95c5384: Preparing \n\u001b[1Bd7a1159c: Preparing \n\u001b[1Bd1217615: Preparing \n\u001b[1Bc1bc2645: Layer already exists \u001b[5A\u001b[2K\u001b[3A\u001b[2Klatest: digest: sha256:6028a2f793dc35ee033376f1c43294cfdb0e3128443b6dbea7836393d6f53fd2 size: 2211\nmake[1]: Leaving directory '/home/jupyter/asl-ml-immersion/notebooks/building_production_ml_systems/solutions/pipelines/components/deploymodel'\n" ] ], [ [ "Now that the container images are pushed to the [registry in your project](https://console.cloud.google.com/gcr), we need to create yaml files describing to Kubeflow how to use these containers. It boils down essentially to\n* describing what arguments Kubeflow needs to pass to the containers when it runs them\n* telling Kubeflow where to fetch the corresponding Docker images\n\nIn the cells below, we have three of these \"Kubeflow component description files\", one for each of our components.", "_____no_output_____" ], [ "**TODO 3**", "_____no_output_____" ], [ "**IMPORTANT: Modify the image URI in the cell \nbelow to reflect that you pushed the images into the gcr.io associated with your project.**", "_____no_output_____" ] ], [ [ "%%writefile bq2gcs.yaml\n\nname: bq2gcs\n \ndescription: |\n This component creates the training and\n validation datasets as BiqQuery tables and export\n them into a Google Cloud Storage bucket at\n gs://qwiklabs-gcp-00-568a75dfa3e1/taxifare/data.\n \ninputs:\n - {name: Input Bucket , type: String, description: 'GCS directory path.'}\n\nimplementation:\n container:\n image: gcr.io/qwiklabs-gcp-00-568a75dfa3e1/taxifare-bq2gcs\n args: [\"--bucket\", {inputValue: Input Bucket}]", "Overwriting bq2gcs.yaml\n" ], [ "%%writefile trainjob.yaml\n\nname: trainjob\n \ndescription: |\n This component trains a model to predict that taxi fare in NY.\n It takes as argument a GCS bucket and expects its training and\n eval data to be at gs://<BUCKET>/taxifare/data/ and will export\n the trained model at gs://<BUCKET>/taxifare/model/.\n \ninputs:\n - {name: Input Bucket , type: String, description: 'GCS directory path.'}\n\nimplementation:\n container:\n image: gcr.io/qwiklabs-gcp-00-568a75dfa3e1/taxifare-trainjob\n args: [{inputValue: Input Bucket}]", "Overwriting trainjob.yaml\n" ], [ "%%writefile deploymodel.yaml\n\nname: deploymodel\n \ndescription: |\n This component deploys a trained taxifare model on GCP as taxifare:dnn.\n It takes as argument a GCS bucket and expects the model to deploy \n to be found at gs://<BUCKET>/taxifare/model/export/savedmodel/\n \ninputs:\n - {name: Input Bucket , type: String, description: 'GCS directory path.'}\n\nimplementation:\n container:\n image: gcr.io/qwiklabs-gcp-00-568a75dfa3e1/taxifare-deploymodel\n args: [{inputValue: Input Bucket}]", "Overwriting deploymodel.yaml\n" ] ], [ [ "## Create a Kubeflow pipeline", "_____no_output_____" ], [ "The code below creates a kubeflow pipeline by decorating a regular function with the\n`@dsl.pipeline` decorator. Now the arguments of this decorated function will be\nthe input parameters of the Kubeflow pipeline.\n\nInside the function, we describe the pipeline by\n* loading the yaml component files we created above into a Kubeflow `op`\n* specifying the order into which the Kubeflow ops should be run", "_____no_output_____" ] ], [ [ "# TODO 3\nPIPELINE_TAR = \"taxifare.tar.gz\"\nBQ2GCS_YAML = \"./bq2gcs.yaml\"\nTRAINJOB_YAML = \"./trainjob.yaml\"\nDEPLOYMODEL_YAML = \"./deploymodel.yaml\"\n\n\[email protected](\n name=\"Taxifare\",\n description=\"Train a ml model to predict the taxi fare in NY\",\n)\ndef pipeline(gcs_bucket_name=\"<bucket where data and model will be exported>\"):\n\n bq2gcs_op = comp.load_component_from_file(BQ2GCS_YAML)\n bq2gcs = bq2gcs_op(\n input_bucket=gcs_bucket_name,\n )\n\n\n\"\"\"\n trainjob_op = comp.load_component_from_file(TRAINJOB_YAML)\n trainjob = trainjob_op(\n input_bucket=gcs_bucket_name,\n )\n\n deploymodel_op = comp.load_component_from_file(DEPLOYMODEL_YAML)\n deploymodel = deploymodel_op(\n input_bucket=gcs_bucket_name,\n )\n\n trainjob.after(bq2gcs)\n deploymodel.after(trainjob)\n\"\"\"", "_____no_output_____" ] ], [ [ "The pipeline function above is then used by the Kubeflow compiler to create a Kubeflow pipeline artifact that can be either uploaded to the Kubeflow cluster from the UI, or programatically, as we will do below:", "_____no_output_____" ] ], [ [ "compiler.Compiler().compile(pipeline, PIPELINE_TAR)", "/home/jupyter/.local/lib/python3.7/site-packages/kfp/components/_components.py:175: FutureWarning: Container component must specify command to be compatible with KFP v2 compatible mode and emissary executor, which will be the default executor for KFP v2.https://www.kubeflow.org/docs/components/pipelines/installation/choose-executor/\n category=FutureWarning,\n" ], [ "ls $PIPELINE_TAR", "taxifare.tar.gz\n" ] ], [ [ "If you untar and uzip this pipeline artifact, you'll see that the compiler has transformed the\nPython description of the pipeline into yaml description!\n\nNow let's feed Kubeflow with our pipeline and run it using our client:", "_____no_output_____" ] ], [ [ "# TODO 4\nrun = client.run_pipeline(\n experiment_id=exp.id,\n job_name=\"taxifare\",\n pipeline_package_path=\"taxifare.tar.gz\",\n params={\n \"gcs_bucket_name\": BUCKET,\n },\n)", "_____no_output_____" ] ], [ [ "Have a look at the link to monitor the run. ", "_____no_output_____" ], [ "Now all the runs are nicely organized under the experiment in the UI, and new runs can be either manually launched or scheduled through the UI in a completely repeatable and traceable way!", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown" ]

[ [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown", "markdown" ], [ "code", "code", "code" ], [ "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code", "code", "code" ], [ "markdown", "markdown", "markdown" ], [ "code", "code", "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "<a href=\"https://colab.research.google.com/github/amittal27/course-v3/blob/master/planet_experimental2.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>", "_____no_output_____" ], [ "### **Setup**", "_____no_output_____" ] ], [ [ "%reload_ext autoreload\n%autoreload 2\n%matplotlib inline", "_____no_output_____" ], [ "from fastai.vision import *", "_____no_output_____" ] ], [ [ "### **Configure data**", "_____no_output_____" ] ], [ [ "path = Path.cwd()/'planet'\npath.mkdir(parents=True, exist_ok=True)\npath", "_____no_output_____" ] ], [ [ "### **Multiclassification**", "_____no_output_____" ] ], [ [ "# read the csv file using the pandas library (popular way of dealing with tabular data in python), print the first five rows\ndf = pd.read_csv(path/'train_classes.csv')\ndf.head()", "_____no_output_____" ], [ "# data augmentation\ntfms = get_transforms(flip_vert=True, max_lighting=0.1, max_zoom=1.05, max_warp=0) # do not want warp on the satellite images", "_____no_output_____" ], [ "# ensure we have the same validation set each time\nnp.random.seed(42)\n# get images, get labels\nsrc = (ImageList.from_csv(path, 'train_classes.csv', folder='train-jpg', suffix='.jpg')\n .split_by_rand_pct(0.2) # set aside 20% of the training set (the current set of images) for the validation set\n .label_from_df(label_delim=' '))", "_____no_output_____" ], [ "# apply transforms, construct dataset\ndata = (src.transform(tfms, size=128) # standardize images to be a little smaller, 128 x 128\n .databunch().normalize(imagenet_stats)) # use databunch to bind training and validation datasets", "_____no_output_____" ], [ "data.show_batch(rows=3, figsize=(12,9))", "_____no_output_____" ], [ "# base artchitecture\narch = models.resnet50", "_____no_output_____" ] ], [ [ "### **Metrics**\nmetrics to print out during training (NOTE: they do not impact how our model trains); just shows us how we're doing", "_____no_output_____" ] ], [ [ "# however, instead of picking just one of the classes in len(data.classes) as our prediction label, we want to pick out n of those classes\n# anything higher than a desired threshold will be assumed to be a label for the input image\n# accuracy uses argmax to find the category with the maximum probability of being represented by the image/data given\n# it compared it to the actual accuracy and took the average; this method can't be implemented when we have multiple labels per image\n# in this case, our threshold value is 0.2 (experimentally found to be pretty good)\n\n# a partial function takes in a function and a list of keywords & values and creates a new func that's exactly the same but with the arguments provided; new function was generated\nacc_02 = partial(accuracy_thresh, thresh=0.2) \ndata.c # number of outputs we want our model to create = len(data.classes); given one probability for each of these classes", "_____no_output_____" ], [ "f_score = partial(fbeta, thresh=0.2) # a metric used by Kaggle to weigh false positive and false negatives\nlearn = cnn_learner(data, arch, metrics=[acc_02, f_score])", "_____no_output_____" ], [ "# find a good learning rate\nlearn.lr_find()", "_____no_output_____" ], [ "# plot results\nlearn.recorder.plot()", "_____no_output_____" ], [ "# pick learning rate\nlr = 1e-2", "_____no_output_____" ], [ "# fit_one_cycle five times with that learning rate\nlearn.fit_one_cycle(5, slice(lr))", "_____no_output_____" ], [ "# save\nlearn.save('stage-1-rn50')", "_____no_output_____" ], [ "learn.unfreeze()", "_____no_output_____" ], [ "learn.lr_find()\nlearn.recorder.plot()", "_____no_output_____" ], [ "# fitting with original dataset, though we could create a new databunch with just the misclassified instances\nlearn.fit_one_cycle(5, slice(1e-5, lr/5))", "_____no_output_____" ], [ "learn.save('stage-2-rn50')", "_____no_output_____" ] ], [ [ "### **Now, create a whole \"new\" dataset where the images are 256 x 256 to hopefully increase our fbeta metric score**\nno concern of overfitting then", "_____no_output_____" ] ], [ [ "# create new databunch with 256 x 256 images (higher res images)\ndata = (src.transform(tfms, size=256), # same transforms as before\n .databunch.normalize(imagenet_stats))", "_____no_output_____" ], [ "# start with our pre-trained model\nlearn.data = data # replace learner data with new databunch\ndata.train_ds[0][0].shape", "_____no_output_____" ], [ "# freeze to just train the last few layers\nlearn.lr_find()\nlearn.recorder.plot()", "_____no_output_____" ], [ "# new learning rate\nlr=1e-2/2", "_____no_output_____" ], [ "# just training the last few layers\nlearn.fit_one_cycle(5, slice(lr))", "_____no_output_____" ], [ "learn.recorder.plot_losses()", "_____no_output_____" ], [ "learn.save('stage-2-256-rn50')", "_____no_output_____" ], [ "learn.export()", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown", "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "#install.packages(\"BiocManager\")\n#install.packages(\"forcats\")\n#install.packages(\"stringr\")\n#install.packages(\"ggplot2\")\n#install.packages(\"ggrepel\")\n#install.packages(\"readr\")\n#install.packages(\"tidyr\")\n#install.packages(\"survminer\")\n#BiocManager::install(\"GEOquery\")\n#BiocManager::install(\"limma\")\n#BiocManager::install(\"pheatmap\")\n#BiocManager::install(\"org.Hs.eg.db\")\nSys.setenv(\"VROOM_CONNECTION_SIZE\" = 131072 * 2)\noptions(timeout = max(10000, getOption(\"timeout\")))\noptions(warn=0)", "_____no_output_____" ], [ "library(GEOquery)\nsetwd('Documents/')", "_____no_output_____" ], [ "ids = c('GSE190540','GSE147040','GSE174818','GSE152204')\ngse = c()\nfor (i in ids){\n gse = append (gse,getGEO(i,destdir = '.'))\n}", "_____no_output_____" ], [ "library(dplyr)", "_____no_output_____" ], [ "sampleInfo1 <- pData(gse[[1]])\nsampleInfo2 <- pData(gse[[2]])\nsampleInfo3 <- pData(gse[[3]])\nsampleInfo4 <- pData(gse[[4]])", "_____no_output_____" ], [ "sampleInfo1[,c('age:ch1','disease status:ch1','gender:ch1')] -> sampleInfo1", "_____no_output_____" ], [ "colnames(sampleInfo1) = c('Age','Status','Gender')", "_____no_output_____" ], [ "sampleInfo1[\"Status\"][sampleInfo1[\"Status\"] == \"Case\"] <- \"MCI\"", "_____no_output_____" ], [ "sampleInfo2[,c('age at death:ch1','current smoking status:ch1','sex (m, male; f, female):ch1')] -> sampleInfo2", "_____no_output_____" ], [ "colnames(sampleInfo2) = c('Age','Status','Gender')", "_____no_output_____" ], [ "sampleInfo3[,c('age:ch1','covid status:ch1','Sex:ch1')] -> sampleInfo3", "_____no_output_____" ], [ "colnames(sampleInfo3) = c('Age','Status','Gender')", "_____no_output_____" ], [ "sampleInfo3$Status = as.character(sampleInfo3$Status)", "_____no_output_____" ], [ "sampleInfo3[\"Status\"][sampleInfo3[\"Status\"] == '0'] <- \"Control\"", "_____no_output_____" ], [ "sampleInfo4[,c('epigenetic age:ch1','source_name_ch1','Sex:ch1')] -> sampleInfo4", "_____no_output_____" ], [ "colnames(sampleInfo4) = c('Age','Status','Gender')", "_____no_output_____" ], [ "sampleInfo4[\"Status\"][sampleInfo4[\"Status\"] == 'Case'] <- \"OAV\"", "_____no_output_____" ], [ "metadata <- rbind(sampleInfo1, sampleInfo2,sampleInfo3,sampleInfo4)", "_____no_output_____" ], [ "metadata$Age = as.numeric(metadata$Age)", "_____no_output_____" ], [ "metadata$Status=as.factor(metadata$Status)", "_____no_output_____" ], [ "summary(metadata)", "_____no_output_____" ] ] ]

[ "code" ]

[ [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Convolutional Neural Networks: Step by Step\n\nWelcome to Course 4's first assignment! In this assignment, you will implement convolutional (CONV) and pooling (POOL) layers in numpy, including both forward propagation and (optionally) backward propagation. \n\nBy the end of this notebook, you'll be able to: \n\n* Explain the convolution operation\n* Apply two different types of pooling operation\n* Identify the components used in a convolutional neural network (padding, stride, filter, ...) and their purpose\n* Build a convolutional neural network \n\n**Notation**:\n- Superscript $[l]$ denotes an object of the $l^{th}$ layer. \n - Example: $a^{[4]}$ is the $4^{th}$ layer activation. $W^{[5]}$ and $b^{[5]}$ are the $5^{th}$ layer parameters.\n\n\n- Superscript $(i)$ denotes an object from the $i^{th}$ example. \n - Example: $x^{(i)}$ is the $i^{th}$ training example input.\n \n \n- Subscript $i$ denotes the $i^{th}$ entry of a vector.\n - Example: $a^{[l]}_i$ denotes the $i^{th}$ entry of the activations in layer $l$, assuming this is a fully connected (FC) layer.\n \n \n- $n_H$, $n_W$ and $n_C$ denote respectively the height, width and number of channels of a given layer. If you want to reference a specific layer $l$, you can also write $n_H^{[l]}$, $n_W^{[l]}$, $n_C^{[l]}$. \n- $n_{H_{prev}}$, $n_{W_{prev}}$ and $n_{C_{prev}}$ denote respectively the height, width and number of channels of the previous layer. If referencing a specific layer $l$, this could also be denoted $n_H^{[l-1]}$, $n_W^{[l-1]}$, $n_C^{[l-1]}$. \n\nYou should be familiar with `numpy` and/or have completed the previous courses of the specialization. Let's get started!", "_____no_output_____" ], [ "## Table of Contents\n\n- [1 - Packages](#1)\n- [2 - Outline of the Assignment](#2)\n- [3 - Convolutional Neural Networks](#3)\n - [3.1 - Zero-Padding](#3-1)\n - [Exercise 1 - zero_pad](#ex-1)\n - [3.2 - Single Step of Convolution](#3-2)\n - [Exercise 2 - conv_single_step](#ex-2)\n - [3.3 - Convolutional Neural Networks - Forward Pass](#3-3)\n - [Exercise 3 - conv_forward](#ex-3)\n- [4 - Pooling Layer](#4)\n - [4.1 - Forward Pooling](#4-1)\n - [Exercise 4 - pool_forward](#ex-4)\n- [5 - Backpropagation in Convolutional Neural Networks (OPTIONAL / UNGRADED)](#5)\n - [5.1 - Convolutional Layer Backward Pass](#5-1)\n - [5.1.1 - Computing dA](#5-1-1)\n - [5.1.2 - Computing dW](#5-1-2)\n - [5.1.3 - Computing db](#5-1-3)\n - [Exercise 5 - conv_backward](#ex-5)\n - [5.2 Pooling Layer - Backward Pass](#5-2)\n - [5.2.1 Max Pooling - Backward Pass](#5-2-1)\n - [Exercise 6 - create_mask_from_window](#ex-6)\n - [5.2.2 - Average Pooling - Backward Pass](#5-2-2)\n - [Exercise 7 - distribute_value](#ex-7)\n - [5.2.3 Putting it Together: Pooling Backward](#5-2-3)\n - [Exercise 8 - pool_backward](#ex-8)", "_____no_output_____" ], [ "<a name='1'></a>\n## 1 - Packages\n\nLet's first import all the packages that you will need during this assignment. \n- [numpy](www.numpy.org) is the fundamental package for scientific computing with Python.\n- [matplotlib](http://matplotlib.org) is a library to plot graphs in Python.\n- np.random.seed(1) is used to keep all the random function calls consistent. This helps to grade your work.", "_____no_output_____" ] ], [ [ "import numpy as np\nimport h5py\nimport matplotlib.pyplot as plt\nfrom public_tests import *\n\n%matplotlib inline\nplt.rcParams['figure.figsize'] = (5.0, 4.0) # set default size of plots\nplt.rcParams['image.interpolation'] = 'nearest'\nplt.rcParams['image.cmap'] = 'gray'\n\n%load_ext autoreload\n%autoreload 2\n\nnp.random.seed(1)", "_____no_output_____" ] ], [ [ "<a name='2'></a>\n## 2 - Outline of the Assignment\n\nYou will be implementing the building blocks of a convolutional neural network! Each function you will implement will have detailed instructions to walk you through the steps:\n\n- Convolution functions, including:\n - Zero Padding\n - Convolve window \n - Convolution forward\n - Convolution backward (optional)\n- Pooling functions, including:\n - Pooling forward\n - Create mask \n - Distribute value\n - Pooling backward (optional)\n \nThis notebook will ask you to implement these functions from scratch in `numpy`. In the next notebook, you will use the TensorFlow equivalents of these functions to build the following model:\n\n<img src=\"images/model.png\" style=\"width:800px;height:300px;\">\n\n**Note**: For every forward function, there is a corresponding backward equivalent. Hence, at every step of your forward module you will store some parameters in a cache. These parameters are used to compute gradients during backpropagation. ", "_____no_output_____" ], [ "<a name='3'></a>\n## 3 - Convolutional Neural Networks\n\nAlthough programming frameworks make convolutions easy to use, they remain one of the hardest concepts to understand in Deep Learning. A convolution layer transforms an input volume into an output volume of different size, as shown below. \n\n<img src=\"images/conv_nn.png\" style=\"width:350px;height:200px;\">\n\nIn this part, you will build every step of the convolution layer. You will first implement two helper functions: one for zero padding and the other for computing the convolution function itself. ", "_____no_output_____" ], [ "<a name='3-1'></a>\n### 3.1 - Zero-Padding\n\nZero-padding adds zeros around the border of an image:\n\n<img src=\"images/PAD.png\" style=\"width:600px;height:400px;\">\n<caption><center> <u> <font color='purple'> <b>Figure 1</b> </u><font color='purple'> : <b>Zero-Padding</b><br> Image (3 channels, RGB) with a padding of 2. </center></caption>\n\nThe main benefits of padding are:\n\n- It allows you to use a CONV layer without necessarily shrinking the height and width of the volumes. This is important for building deeper networks, since otherwise the height/width would shrink as you go to deeper layers. An important special case is the \"same\" convolution, in which the height/width is exactly preserved after one layer. \n\n- It helps us keep more of the information at the border of an image. Without padding, very few values at the next layer would be affected by pixels at the edges of an image.\n\n<a name='ex-1'></a>\n### Exercise 1 - zero_pad\nImplement the following function, which pads all the images of a batch of examples X with zeros. [Use np.pad](https://docs.scipy.org/doc/numpy/reference/generated/numpy.pad.html). Note if you want to pad the array \"a\" of shape $(5,5,5,5,5)$ with `pad = 1` for the 2nd dimension, `pad = 3` for the 4th dimension and `pad = 0` for the rest, you would do:\n```python\na = np.pad(a, ((0,0), (1,1), (0,0), (3,3), (0,0)), mode='constant', constant_values = (0,0))\n```", "_____no_output_____" ] ], [ [ "# GRADED FUNCTION: zero_pad\n\ndef zero_pad(X, pad):\n \"\"\"\n Pad with zeros all images of the dataset X. The padding is applied to the height and width of an image, \n as illustrated in Figure 1.\n \n Argument:\n X -- python numpy array of shape (m, n_H, n_W, n_C) representing a batch of m images\n pad -- integer, amount of padding around each image on vertical and horizontal dimensions\n \n Returns:\n X_pad -- padded image of shape (m, n_H + 2 * pad, n_W + 2 * pad, n_C)\n \"\"\"\n \n #(≈ 1 line)\n X_pad = np.pad(X, ((0,0), (pad,pad), (pad,pad), (0,0)), mode='constant', constant_values = (0,0))\n # YOUR CODE STARTS HERE\n \n \n # YOUR CODE ENDS HERE\n \n return X_pad", "_____no_output_____" ], [ "np.random.seed(1)\nx = np.random.randn(4, 3, 3, 2)\nx_pad = zero_pad(x, 3)\nprint (\"x.shape =\\n\", x.shape)\nprint (\"x_pad.shape =\\n\", x_pad.shape)\nprint (\"x[1,1] =\\n\", x[1, 1])\nprint (\"x_pad[1,1] =\\n\", x_pad[1, 1])\n\nassert type(x_pad) == np.ndarray, \"Output must be a np array\"\nassert x_pad.shape == (4, 9, 9, 2), f\"Wrong shape: {x_pad.shape} != (4, 9, 9, 2)\"\nprint(x_pad[0, 0:2,:, 0])\nassert np.allclose(x_pad[0, 0:2,:, 0], [[0, 0, 0, 0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0, 0, 0, 0]], 1e-15), \"Rows are not padded with zeros\"\nassert np.allclose(x_pad[0, :, 7:9, 1].transpose(), [[0, 0, 0, 0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0, 0, 0, 0]], 1e-15), \"Columns are not padded with zeros\"\nassert np.allclose(x_pad[:, 3:6, 3:6, :], x, 1e-15), \"Internal values are different\"\n\nfig, axarr = plt.subplots(1, 2)\naxarr[0].set_title('x')\naxarr[0].imshow(x[0, :, :, 0])\naxarr[1].set_title('x_pad')\naxarr[1].imshow(x_pad[0, :, :, 0])\nzero_pad_test(zero_pad)", "x.shape =\n (4, 3, 3, 2)\nx_pad.shape =\n (4, 9, 9, 2)\nx[1,1] =\n [[ 0.90085595 -0.68372786]\n [-0.12289023 -0.93576943]\n [-0.26788808 0.53035547]]\nx_pad[1,1] =\n [[0. 0.]\n [0. 0.]\n [0. 0.]\n [0. 0.]\n [0. 0.]\n [0. 0.]\n [0. 0.]\n [0. 0.]\n [0. 0.]]\n[[0. 0. 0. 0. 0. 0. 0. 0. 0.]\n [0. 0. 0. 0. 0. 0. 0. 0. 0.]]\n\u001b[92m All tests passed.\n" ] ], [ [ "<a name='3-2'></a>\n### 3.2 - Single Step of Convolution \n\nIn this part, implement a single step of convolution, in which you apply the filter to a single position of the input. This will be used to build a convolutional unit, which: \n\n- Takes an input volume \n- Applies a filter at every position of the input\n- Outputs another volume (usually of different size)\n\n<img src=\"images/Convolution_schematic.gif\" style=\"width:500px;height:300px;\">\n<caption><center> <u> <font color='purple'> <b>Figure 2</b> </u><font color='purple'> : <b>Convolution operation</b><br> with a filter of 3x3 and a stride of 1 (stride = amount you move the window each time you slide) </center></caption>\n\nIn a computer vision application, each value in the matrix on the left corresponds to a single pixel value. You convolve a 3x3 filter with the image by multiplying its values element-wise with the original matrix, then summing them up and adding a bias. In this first step of the exercise, you will implement a single step of convolution, corresponding to applying a filter to just one of the positions to get a single real-valued output. \n\nLater in this notebook, you'll apply this function to multiple positions of the input to implement the full convolutional operation. \n\n<a name='ex-2'></a>\n### Exercise 2 - conv_single_step\nImplement `conv_single_step()`. \n \n[Hint](https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.sum.html).", "_____no_output_____" ], [ "**Note**: The variable b will be passed in as a numpy array. If you add a scalar (a float or integer) to a numpy array, the result is a numpy array. In the special case of a numpy array containing a single value, you can cast it as a float to convert it to a scalar.", "_____no_output_____" ] ], [ [ "# GRADED FUNCTION: conv_single_step\n\ndef conv_single_step(a_slice_prev, W, b):\n \"\"\"\n Apply one filter defined by parameters W on a single slice (a_slice_prev) of the output activation \n of the previous layer.\n \n Arguments:\n a_slice_prev -- slice of input data of shape (f, f, n_C_prev)\n W -- Weight parameters contained in a window - matrix of shape (f, f, n_C_prev)\n b -- Bias parameters contained in a window - matrix of shape (1, 1, 1)\n \n Returns:\n Z -- a scalar value, the result of convolving the sliding window (W, b) on a slice x of the input data\n \"\"\"\n\n #(≈ 3 lines of code)\n # Element-wise product between a_slice_prev and W. Do not add the bias yet.\n s = a_slice_prev * W\n # Sum over all entries of the volume s.\n Z = s.reshape(1,-1).sum()\n # Add bias b to Z. Cast b to a float() so that Z results in a scalar value.\n Z = Z + float(b)\n # YOUR CODE STARTS HERE\n \n \n # YOUR CODE ENDS HERE\n\n return Z", "_____no_output_____" ], [ "np.random.seed(1)\na_slice_prev = np.random.randn(4, 4, 3)\nW = np.random.randn(4, 4, 3)\nb = np.random.randn(1, 1, 1)\n\nZ = conv_single_step(a_slice_prev, W, b)\nprint(\"Z =\", Z)\nconv_single_step_test(conv_single_step)\n\nassert (type(Z) == np.float64 or type(Z) == np.float32), \"You must cast the output to float\"\nassert np.isclose(Z, -6.999089450680221), \"Wrong value\"", "Z = -6.999089450680221\n\u001b[92m All tests passed.\n" ] ], [ [ "<a name='3-3'></a>\n### 3.3 - Convolutional Neural Networks - Forward Pass\n\nIn the forward pass, you will take many filters and convolve them on the input. Each 'convolution' gives you a 2D matrix output. You will then stack these outputs to get a 3D volume: \n\n<center>\n<video width=\"620\" height=\"440\" src=\"images/conv_kiank.mp4\" type=\"video/mp4\" controls>\n</video>\n</center>\n\n<a name='ex-3'></a>\n### Exercise 3 - conv_forward\nImplement the function below to convolve the filters `W` on an input activation `A_prev`. \nThis function takes the following inputs:\n* `A_prev`, the activations output by the previous layer (for a batch of m inputs); \n* Weights are denoted by `W`. The filter window size is `f` by `f`.\n* The bias vector is `b`, where each filter has its own (single) bias. \n\nYou also have access to the hyperparameters dictionary, which contains the stride and the padding. \n\n**Hint**: \n1. To select a 2x2 slice at the upper left corner of a matrix \"a_prev\" (shape (5,5,3)), you would do:\n```python\na_slice_prev = a_prev[0:2,0:2,:]\n```\nNotice how this gives a 3D slice that has height 2, width 2, and depth 3. Depth is the number of channels. \nThis will be useful when you will define `a_slice_prev` below, using the `start/end` indexes you will define.\n\n2. To define a_slice you will need to first define its corners `vert_start`, `vert_end`, `horiz_start` and `horiz_end`. This figure may be helpful for you to find out how each of the corners can be defined using h, w, f and s in the code below.\n\n<img src=\"images/vert_horiz_kiank.png\" style=\"width:400px;height:300px;\">\n<caption><center> <u> <font color='purple'> <b>Figure 3</b> </u><font color='purple'> : <b>Definition of a slice using vertical and horizontal start/end (with a 2x2 filter)</b> <br> This figure shows only a single channel. </center></caption>\n\n\n**Reminder**:\n \nThe formulas relating the output shape of the convolution to the input shape are:\n \n$$n_H = \\Bigl\\lfloor \\frac{n_{H_{prev}} - f + 2 \\times pad}{stride} \\Bigr\\rfloor +1$$\n$$n_W = \\Bigl\\lfloor \\frac{n_{W_{prev}} - f + 2 \\times pad}{stride} \\Bigr\\rfloor +1$$\n$$n_C = \\text{number of filters used in the convolution}$$\n \n\n\n\nFor this exercise, don't worry about vectorization! Just implement everything with for-loops.", "_____no_output_____" ], [ "#### Additional Hints (if you're stuck):\n\n\n* Use array slicing (e.g.`varname[0:1,:,3:5]`) for the following variables: \n `a_prev_pad` ,`W`, `b` \n - Copy the starter code of the function and run it outside of the defined function, in separate cells. \n - Check that the subset of each array is the size and dimension that you're expecting. \n* To decide how to get the `vert_start`, `vert_end`, `horiz_start`, `horiz_end`, remember that these are indices of the previous layer. \n - Draw an example of a previous padded layer (8 x 8, for instance), and the current (output layer) (2 x 2, for instance). \n - The output layer's indices are denoted by `h` and `w`. \n* Make sure that `a_slice_prev` has a height, width and depth.\n* Remember that `a_prev_pad` is a subset of `A_prev_pad`. \n - Think about which one should be used within the for loops.", "_____no_output_____" ] ], [ [ "# GRADED FUNCTION: conv_forward\n\ndef conv_forward(A_prev, W, b, hparameters):\n '''\n Implements the forward propagation for a convolution function\n \n Arguments:\n A_prev -- output activations of the previous layer, \n numpy array of shape (m, n_H_prev, n_W_prev, n_C_prev)\n W -- Weights, numpy array of shape (f, f, n_C_prev, n_C)\n b -- Biases, numpy array of shape (1, 1, 1, n_C)\n hparameters -- python dictionary containing \"stride\" and \"pad\"\n \n Returns:\n Z -- conv output, numpy array of shape (m, n_H, n_W, n_C)\n cache -- cache of values needed for the conv_backward() function\n '''\n \n # Retrieve dimensions from A_prev's shape (≈1 line) \n (m, n_H_prev, n_W_prev, n_C_prev) = A_prev.shape\n \n # Retrieve dimensions from W's shape (≈1 line)\n (f, f, n_C_prev, n_C) = W.shape\n \n # Retrieve information from \"hparameters\" (≈2 lines)\n stride = hparameters['stride']\n pad = hparameters['pad']\n \n # Compute the dimensions of the CONV output volume using the formula given above. \n # Hint: use int() to apply the 'floor' operation. (≈2 lines)\n n_H = int((n_H_prev+(2*pad) -f )/stride) + 1\n n_W = int((n_W_prev+(2*pad) -f )/stride) + 1\n \n # Initialize the output volume Z with zeros. (≈1 line)\n Z = np.zeros((m, n_H, n_W, n_C))\n \n # Create A_prev_pad by padding A_prev\n A_prev_pad = zero_pad(A_prev, pad)\n \n for i in range(m): # loop over the batch of training examples\n a_prev_pad = A_prev_pad[i] # Select ith training example's padded activation\n for h in range(n_H): # loop over vertical axis of the output volume\n # Find the vertical start and end of the current \"slice\" (≈2 lines)\n vert_start = h * stride\n vert_end = h * stride+ f\n \n for w in range(n_W): # loop over horizontal axis of the output volume\n # Find the horizontal start and end of the current \"slice\" (≈2 lines)\n horiz_start = w * stride\n horiz_end = w * stride + f\n \n for c in range(n_C): # loop over channels (= #filters) of the output volume\n \n # Use the corners to define the (3D) slice of a_prev_pad (See Hint above the cell). (≈1 line)\n a_slice_prev = a_prev_pad[vert_start:vert_end,horiz_start:horiz_end,:]\n \n # Convolve the (3D) slice with the correct filter W and bias b, to get back one output neuron. (≈3 line)\n weights = W[:,:,:,c]\n biases = b[:,:,:,c]\n Z[i, h, w, c] = conv_single_step(a_slice_prev, W[:,:,:,c], b[:,:,:,c])\n # YOUR CODE STARTS HERE\n \n \n # YOUR CODE ENDS HERE\n \n # Save information in \"cache\" for the backprop\n cache = (A_prev, W, b, hparameters)\n \n return Z, cache", "_____no_output_____" ], [ "np.random.seed(1)\nA_prev = np.random.randn(2, 5, 7, 4)\nW = np.random.randn(3, 3, 4, 8)\nb = np.random.randn(1, 1, 1, 8)\nhparameters = {\"pad\" : 1,\n \"stride\": 2}\n\nZ, cache_conv = conv_forward(A_prev, W, b, hparameters)\nprint(\"Z's mean =\\n\", np.mean(Z))\nprint(\"Z[0,2,1] =\\n\", Z[0, 2, 1])\nprint(\"cache_conv[0][1][2][3] =\\n\", cache_conv[0][1][2][3])\n\nconv_forward_test(conv_forward)\n", "Z's mean =\n 0.5511276474566768\nZ[0,2,1] =\n [-2.17796037 8.07171329 -0.5772704 3.36286738 4.48113645 -2.89198428\n 10.99288867 3.03171932]\ncache_conv[0][1][2][3] =\n [-1.1191154 1.9560789 -0.3264995 -1.34267579]\n(2, 13, 15, 8)\n\u001b[92m All tests passed.\n" ] ], [ [ "Finally, a CONV layer should also contain an activation, in which case you would add the following line of code:\n\n```python\n# Convolve the window to get back one output neuron\nZ[i, h, w, c] = ...\n# Apply activation\nA[i, h, w, c] = activation(Z[i, h, w, c])\n```\n\nYou don't need to do it here, however. \n", "_____no_output_____" ], [ "<a name='4'></a>\n## 4 - Pooling Layer \n\nThe pooling (POOL) layer reduces the height and width of the input. It helps reduce computation, as well as helps make feature detectors more invariant to its position in the input. The two types of pooling layers are: \n\n- Max-pooling layer: slides an ($f, f$) window over the input and stores the max value of the window in the output.\n\n- Average-pooling layer: slides an ($f, f$) window over the input and stores the average value of the window in the output.\n\n<table>\n<td>\n<img src=\"images/max_pool1.png\" style=\"width:500px;height:300px;\">\n<td>\n\n<td>\n<img src=\"images/a_pool.png\" style=\"width:500px;height:300px;\">\n<td>\n</table>\n\nThese pooling layers have no parameters for backpropagation to train. However, they have hyperparameters such as the window size $f$. This specifies the height and width of the $f \\times f$ window you would compute a *max* or *average* over. \n\n<a name='4-1'></a>\n### 4.1 - Forward Pooling\nNow, you are going to implement MAX-POOL and AVG-POOL, in the same function. \n\n<a name='ex-4'></a>\n### Exercise 4 - pool_forward\n\nImplement the forward pass of the pooling layer. Follow the hints in the comments below.\n\n**Reminder**:\nAs there's no padding, the formulas binding the output shape of the pooling to the input shape is:\n\n$$n_H = \\Bigl\\lfloor \\frac{n_{H_{prev}} - f}{stride} \\Bigr\\rfloor +1$$\n\n$$n_W = \\Bigl\\lfloor \\frac{n_{W_{prev}} - f}{stride} \\Bigr\\rfloor +1$$\n\n$$n_C = n_{C_{prev}}$$\n\n\n", "_____no_output_____" ] ], [ [ "# GRADED FUNCTION: pool_forward\n\ndef pool_forward(A_prev, hparameters, mode = \"max\"):\n \"\"\"\n Implements the forward pass of the pooling layer\n \n Arguments:\n A_prev -- Input data, numpy array of shape (m, n_H_prev, n_W_prev, n_C_prev)\n hparameters -- python dictionary containing \"f\" and \"stride\"\n mode -- the pooling mode you would like to use, defined as a string (\"max\" or \"average\")\n \n Returns:\n A -- output of the pool layer, a numpy array of shape (m, n_H, n_W, n_C)\n cache -- cache used in the backward pass of the pooling layer, contains the input and hparameters \n \"\"\"\n \n # Retrieve dimensions from the input shape\n (m, n_H_prev, n_W_prev, n_C_prev) = A_prev.shape\n \n # Retrieve hyperparameters from \"hparameters\"\n f = hparameters[\"f\"]\n stride = hparameters[\"stride\"]\n \n # Define the dimensions of the output\n n_H = int(1 + (n_H_prev - f) / stride)\n n_W = int(1 + (n_W_prev - f) / stride)\n n_C = n_C_prev\n \n # Initialize output matrix A\n A = np.zeros((m, n_H, n_W, n_C)) \n \n for i in range(m): # loop over the training examples\n for h in range(n_H): # loop on the vertical axis of the output volume\n # Find the vertical start and end of the current \"slice\" (≈2 lines)\n vert_start = h * stride\n vert_end = h * stride+ f\n \n for w in range(n_W): # loop on the horizontal axis of the output volume\n #Find the vertical start and end of the current \"slice\" (≈2 lines)\n horiz_start = w * stride\n horiz_end = w * stride + f\n \n for c in range (n_C): # loop over the channels of the output volume\n \n # Use the corners to define the current slice on the ith training example of A_prev, channel c. (≈1 line)\n a_prev_slice = A_prev[i, vert_start:vert_end, horiz_start:horiz_end,c]\n \n # Compute the pooling operation on the slice. \n # Use an if statement to differentiate the modes. \n # Use np.max and np.mean.\n if mode == \"max\":\n A[i, h, w, c] = np.max(a_prev_slice)\n elif mode == \"average\":\n A[i, h, w, c] = np.mean(a_prev_slice)\n \n # YOUR CODE STARTS HERE\n \n \n # YOUR CODE ENDS HERE\n \n # Store the input and hparameters in \"cache\" for pool_backward()\n cache = (A_prev, hparameters)\n \n # Making sure your output shape is correct\n assert(A.shape == (m, n_H, n_W, n_C))\n \n return A, cache", "_____no_output_____" ], [ "# Case 1: stride of 1\nnp.random.seed(1)\nA_prev = np.random.randn(2, 5, 5, 3)\nhparameters = {\"stride\" : 1, \"f\": 3}\n\nA, cache = pool_forward(A_prev, hparameters, mode = \"max\")\nprint(\"mode = max\")\nprint(\"A.shape = \" + str(A.shape))\nprint(\"A[1, 1] =\\n\", A[1, 1])\nprint()\nA, cache = pool_forward(A_prev, hparameters, mode = \"average\")\nprint(\"mode = average\")\nprint(\"A.shape = \" + str(A.shape))\nprint(\"A[1, 1] =\\n\", A[1, 1])\n\npool_forward_test(pool_forward)", "mode = max\nA.shape = (2, 3, 3, 3)\nA[1, 1] =\n [[1.96710175 0.84616065 1.27375593]\n [1.96710175 0.84616065 1.23616403]\n [1.62765075 1.12141771 1.2245077 ]]\n\nmode = average\nA.shape = (2, 3, 3, 3)\nA[1, 1] =\n [[ 0.44497696 -0.00261695 -0.31040307]\n [ 0.50811474 -0.23493734 -0.23961183]\n [ 0.11872677 0.17255229 -0.22112197]]\n\u001b[92m All tests passed.\n" ] ], [ [ "**Expected output**\n\n```\nmode = max\nA.shape = (2, 3, 3, 3)\nA[1, 1] =\n [[1.96710175 0.84616065 1.27375593]\n [1.96710175 0.84616065 1.23616403]\n [1.62765075 1.12141771 1.2245077 ]]\n\nmode = average\nA.shape = (2, 3, 3, 3)\nA[1, 1] =\n [[ 0.44497696 -0.00261695 -0.31040307]\n [ 0.50811474 -0.23493734 -0.23961183]\n [ 0.11872677 0.17255229 -0.22112197]]\n```", "_____no_output_____" ] ], [ [ "# Case 2: stride of 2\nnp.random.seed(1)\nA_prev = np.random.randn(2, 5, 5, 3)\nhparameters = {\"stride\" : 2, \"f\": 3}\n\nA, cache = pool_forward(A_prev, hparameters)\nprint(\"mode = max\")\nprint(\"A.shape = \" + str(A.shape))\nprint(\"A[0] =\\n\", A[0])\nprint()\n\nA, cache = pool_forward(A_prev, hparameters, mode = \"average\")\nprint(\"mode = average\")\nprint(\"A.shape = \" + str(A.shape))\nprint(\"A[1] =\\n\", A[1])", "mode = max\nA.shape = (2, 2, 2, 3)\nA[0] =\n [[[1.74481176 0.90159072 1.65980218]\n [1.74481176 1.6924546 1.65980218]]\n\n [[1.13162939 1.51981682 2.18557541]\n [1.13162939 1.6924546 2.18557541]]]\n\nmode = average\nA.shape = (2, 2, 2, 3)\nA[1] =\n [[[-0.17313416 0.32377198 -0.34317572]\n [ 0.02030094 0.14141479 -0.01231585]]\n\n [[ 0.42944926 0.08446996 -0.27290905]\n [ 0.15077452 0.28911175 0.00123239]]]\n" ] ], [ [ "**Expected Output:**\n \n```\nmode = max\nA.shape = (2, 2, 2, 3)\nA[0] =\n [[[1.74481176 0.90159072 1.65980218]\n [1.74481176 1.6924546 1.65980218]]\n\n [[1.13162939 1.51981682 2.18557541]\n [1.13162939 1.6924546 2.18557541]]]\n\nmode = average\nA.shape = (2, 2, 2, 3)\nA[1] =\n [[[-0.17313416 0.32377198 -0.34317572]\n [ 0.02030094 0.14141479 -0.01231585]]\n\n [[ 0.42944926 0.08446996 -0.27290905]\n [ 0.15077452 0.28911175 0.00123239]]]\n```", "_____no_output_____" ], [ "<font color='blue'>\n \n**What you should remember**:\n\n* A convolution extracts features from an input image by taking the dot product between the input data and a 3D array of weights (the filter). \n* The 2D output of the convolution is called the feature map\n* A convolution layer is where the filter slides over the image and computes the dot product \n * This transforms the input volume into an output volume of different size \n* Zero padding helps keep more information at the image borders, and is helpful for building deeper networks, because you can build a CONV layer without shrinking the height and width of the volumes\n* Pooling layers gradually reduce the height and width of the input by sliding a 2D window over each specified region, then summarizing the features in that region", "_____no_output_____" ], [ "**Congratulations**! You have now implemented the forward passes of all the layers of a convolutional network. Great work!\n\nThe remainder of this notebook is optional, and will not be graded. If you carry on, just remember to hit the Submit button to submit your work for grading first. ", "_____no_output_____" ], [ "<a name='5'></a>\n## 5 - Backpropagation in Convolutional Neural Networks (OPTIONAL / UNGRADED)\n\nIn modern deep learning frameworks, you only have to implement the forward pass, and the framework takes care of the backward pass, so most deep learning engineers don't need to bother with the details of the backward pass. The backward pass for convolutional networks is complicated. If you wish, you can work through this optional portion of the notebook to get a sense of what backprop in a convolutional network looks like. \n\nWhen in an earlier course you implemented a simple (fully connected) neural network, you used backpropagation to compute the derivatives with respect to the cost to update the parameters. Similarly, in convolutional neural networks you can calculate the derivatives with respect to the cost in order to update the parameters. The backprop equations are not trivial and were not derived in lecture, but are briefly presented below.\n\n<a name='5-1'></a>\n### 5.1 - Convolutional Layer Backward Pass \n\nLet's start by implementing the backward pass for a CONV layer. \n\n<a name='5-1-1'></a>\n#### 5.1.1 - Computing dA:\nThis is the formula for computing $dA$ with respect to the cost for a certain filter $W_c$ and a given training example:\n\n$$dA \\mathrel{+}= \\sum _{h=0} ^{n_H} \\sum_{w=0} ^{n_W} W_c \\times dZ_{hw} \\tag{1}$$\n\nWhere $W_c$ is a filter and $dZ_{hw}$ is a scalar corresponding to the gradient of the cost with respect to the output of the conv layer Z at the hth row and wth column (corresponding to the dot product taken at the ith stride left and jth stride down). Note that at each time, you multiply the the same filter $W_c$ by a different dZ when updating dA. We do so mainly because when computing the forward propagation, each filter is dotted and summed by a different a_slice. Therefore when computing the backprop for dA, you are just adding the gradients of all the a_slices. \n\nIn code, inside the appropriate for-loops, this formula translates into:\n```python\nda_prev_pad[vert_start:vert_end, horiz_start:horiz_end, :] += W[:,:,:,c] * dZ[i, h, w, c]\n```\n\n<a name='5-1-2'></a>\n#### 5.1.2 - Computing dW:\nThis is the formula for computing $dW_c$ ($dW_c$ is the derivative of one filter) with respect to the loss:\n\n$$dW_c \\mathrel{+}= \\sum _{h=0} ^{n_H} \\sum_{w=0} ^ {n_W} a_{slice} \\times dZ_{hw} \\tag{2}$$\n\nWhere $a_{slice}$ corresponds to the slice which was used to generate the activation $Z_{ij}$. Hence, this ends up giving us the gradient for $W$ with respect to that slice. Since it is the same $W$, we will just add up all such gradients to get $dW$. \n\nIn code, inside the appropriate for-loops, this formula translates into:\n```python\ndW[:,:,:,c] \\mathrel{+}= a_slice * dZ[i, h, w, c]\n```\n\n<a name='5-1-3'></a>\n#### 5.1.3 - Computing db:\n\nThis is the formula for computing $db$ with respect to the cost for a certain filter $W_c$:\n\n$$db = \\sum_h \\sum_w dZ_{hw} \\tag{3}$$\n\nAs you have previously seen in basic neural networks, db is computed by summing $dZ$. In this case, you are just summing over all the gradients of the conv output (Z) with respect to the cost. \n\nIn code, inside the appropriate for-loops, this formula translates into:\n```python\ndb[:,:,:,c] += dZ[i, h, w, c]\n```\n\n<a name='ex-5'></a>\n### Exercise 5 - conv_backward\n\nImplement the `conv_backward` function below. You should sum over all the training examples, filters, heights, and widths. You should then compute the derivatives using formulas 1, 2 and 3 above. ", "_____no_output_____" ] ], [ [ "def conv_backward(dZ, cache):\n \"\"\"\n Implement the backward propagation for a convolution function\n \n Arguments:\n dZ -- gradient of the cost with respect to the output of the conv layer (Z), numpy array of shape (m, n_H, n_W, n_C)\n cache -- cache of values needed for the conv_backward(), output of conv_forward()\n \n Returns:\n dA_prev -- gradient of the cost with respect to the input of the conv layer (A_prev),\n numpy array of shape (m, n_H_prev, n_W_prev, n_C_prev)\n dW -- gradient of the cost with respect to the weights of the conv layer (W)\n numpy array of shape (f, f, n_C_prev, n_C)\n db -- gradient of the cost with respect to the biases of the conv layer (b)\n numpy array of shape (1, 1, 1, n_C)\n \"\"\" \n \n \n # Retrieve information from \"cache\"\n (A_prev, W, b, hparameters) = cache\n # Retrieve dimensions from A_prev's shape\n (m, n_H_prev, n_W_prev, n_C_prev) = A_prev.shape\n # Retrieve dimensions from W's shape\n (f, f, n_C_prev, n_C) = W.shape\n \n # Retrieve information from \"hparameters\"\n stride = hparameters['stride']\n pad = hparameters['pad']\n \n # Retrieve dimensions from dZ's shape\n (m, n_H, n_W, n_C) = dZ.shape\n \n # Initialize dA_prev, dW, db with the correct shapes\n dA_prev = np.zeros((m, n_H_prev, n_W_prev, n_C_prev)) \n dW = np.zeros((f, f, n_C_prev, n_C))\n db = np.zeros((1, 1, 1, n_C))\n \n # Pad A_prev and dA_prev\n A_prev_pad = zero_pad(A_prev, pad)\n dA_prev_pad = zero_pad(dA_prev, pad)\n \n for i in range(m): # loop over the training examples\n \n # select ith training example from A_prev_pad and dA_prev_pad\n a_prev_pad = A_prev_pad[i]\n da_prev_pad = dA_prev_pad[i]\n \n for h in range(n_H): # loop over vertical axis of the output volume\n for w in range(n_W): # loop over horizontal axis of the output volume\n for c in range(n_C): # loop over the channels of the output volume\n \n # Find the corners of the current \"slice\"\n vert_start = h * stride\n vert_end = h * stride + f\n horiz_start = w * stride\n horiz_end = w * stride + f\n\n # Use the corners to define the slice from a_prev_pad\n a_slice = a_prev_pad[vert_start:vert_end,horiz_start:horiz_end,:]\n\n\n # Update gradients for the window and the filter's parameters using the code formulas given above\n da_prev_pad[vert_start:vert_end, horiz_start:horiz_end, :] += W[:,:,:,c] * dZ[i, h, w, c]\n dW[:,:,:,c] += a_slice * dZ[i, h, w, c]\n db[:,:,:,c] += dZ[i, h, w, c]\n \n # Set the ith training example's dA_prev to the unpadded da_prev_pad (Hint: use X[pad:-pad, pad:-pad, :])\n dA_prev[i, :, :, :] = da_prev_pad[pad:-pad, pad:-pad, :]\n # YOUR CODE STARTS HERE\n \n \n # YOUR CODE ENDS HERE\n \n # Making sure your output shape is correct\n assert(dA_prev.shape == (m, n_H_prev, n_W_prev, n_C_prev))\n \n return dA_prev, dW, db", "_____no_output_____" ], [ "# We'll run conv_forward to initialize the 'Z' and 'cache_conv\",\n# which we'll use to test the conv_backward function\nnp.random.seed(1)\nA_prev = np.random.randn(10, 4, 4, 3)\nW = np.random.randn(2, 2, 3, 8)\nb = np.random.randn(1, 1, 1, 8)\nhparameters = {\"pad\" : 2,\n \"stride\": 2}\nZ, cache_conv = conv_forward(A_prev, W, b, hparameters)\n\n# Test conv_backward\ndA, dW, db = conv_backward(Z, cache_conv)\n\nprint(\"dA_mean =\", np.mean(dA))\nprint(\"dW_mean =\", np.mean(dW))\nprint(\"db_mean =\", np.mean(db))\n\nassert type(dA) == np.ndarray, \"Output must be a np.ndarray\"\nassert type(dW) == np.ndarray, \"Output must be a np.ndarray\"\nassert type(db) == np.ndarray, \"Output must be a np.ndarray\"\nassert dA.shape == (10, 4, 4, 3), f\"Wrong shape for dA {dA.shape} != (10, 4, 4, 3)\"\nassert dW.shape == (2, 2, 3, 8), f\"Wrong shape for dW {dW.shape} != (2, 2, 3, 8)\"\nassert db.shape == (1, 1, 1, 8), f\"Wrong shape for db {db.shape} != (1, 1, 1, 8)\"\nassert np.isclose(np.mean(dA), 1.4524377), \"Wrong values for dA\"\nassert np.isclose(np.mean(dW), 1.7269914), \"Wrong values for dW\"\nassert np.isclose(np.mean(db), 7.8392325), \"Wrong values for db\"\n\nprint(\"\\033[92m All tests passed.\")", "dA_mean = 1.4524377775388075\ndW_mean = 1.7269914583139097\ndb_mean = 7.839232564616838\n\u001b[92m All tests passed.\n" ] ], [ [ "**Expected Output**:\n<table>\n <tr>\n <td>\n dA_mean\n </td>\n <td>\n 1.45243777754\n </td>\n </tr>\n <tr>\n <td>\n dW_mean\n </td>\n <td>\n 1.72699145831\n </td>\n </tr>\n <tr>\n <td>\n db_mean\n </td>\n <td>\n 7.83923256462\n </td>\n </tr>\n\n</table>\n", "_____no_output_____" ], [ "<a name='5-2'></a>\n## 5.2 Pooling Layer - Backward Pass\n\nNext, let's implement the backward pass for the pooling layer, starting with the MAX-POOL layer. Even though a pooling layer has no parameters for backprop to update, you still need to backpropagate the gradient through the pooling layer in order to compute gradients for layers that came before the pooling layer. \n\n<a name='5-2-1'></a>\n### 5.2.1 Max Pooling - Backward Pass \n\nBefore jumping into the backpropagation of the pooling layer, you are going to build a helper function called `create_mask_from_window()` which does the following: \n\n$$ X = \\begin{bmatrix}\n1 && 3 \\\\\n4 && 2\n\\end{bmatrix} \\quad \\rightarrow \\quad M =\\begin{bmatrix}\n0 && 0 \\\\\n1 && 0\n\\end{bmatrix}\\tag{4}$$\n\nAs you can see, this function creates a \"mask\" matrix which keeps track of where the maximum of the matrix is. True (1) indicates the position of the maximum in X, the other entries are False (0). You'll see later that the backward pass for average pooling is similar to this, but uses a different mask. \n\n<a name='ex-6'></a>\n### Exercise 6 - create_mask_from_window\n\nImplement `create_mask_from_window()`. This function will be helpful for pooling backward. \nHints:\n- [np.max()]() may be helpful. It computes the maximum of an array.\n- If you have a matrix X and a scalar x: `A = (X == x)` will return a matrix A of the same size as X such that:\n```\nA[i,j] = True if X[i,j] = x\nA[i,j] = False if X[i,j] != x\n```\n- Here, you don't need to consider cases where there are several maxima in a matrix.", "_____no_output_____" ] ], [ [ "def create_mask_from_window(x):\n \"\"\"\n Creates a mask from an input matrix x, to identify the max entry of x.\n \n Arguments:\n x -- Array of shape (f, f)\n \n Returns:\n mask -- Array of the same shape as window, contains a True at the position corresponding to the max entry of x.\n \"\"\" \n # (≈1 line)\n mask = np.max(x) == x\n # YOUR CODE STARTS HERE\n \n \n # YOUR CODE ENDS HERE\n return mask", "_____no_output_____" ], [ "np.random.seed(1)\nx = np.random.randn(2, 3)\nmask = create_mask_from_window(x)\nprint('x = ', x)\nprint(\"mask = \", mask)\n\nx = np.array([[-1, 2, 3],\n [2, -3, 2],\n [1, 5, -2]])\n\ny = np.array([[False, False, False],\n [False, False, False],\n [False, True, False]])\nmask = create_mask_from_window(x)\n\nassert type(mask) == np.ndarray, \"Output must be a np.ndarray\"\nassert mask.shape == x.shape, \"Input and output shapes must match\"\nassert np.allclose(mask, y), \"Wrong output. The True value must be at position (2, 1)\"\n\nprint(\"\\033[92m All tests passed.\")", "x = [[ 1.62434536 -0.61175641 -0.52817175]\n [-1.07296862 0.86540763 -2.3015387 ]]\nmask = [[ True False False]\n [False False False]]\n\u001b[92m All tests passed.\n" ] ], [ [ "**Expected Output:** \n\n<table> \n<tr> \n<td>\n\n**x =**\n</td>\n\n<td>\n\n[[ 1.62434536 -0.61175641 -0.52817175] <br>\n [-1.07296862 0.86540763 -2.3015387 ]]\n\n </td>\n</tr>\n\n<tr> \n<td>\nmask =\n</td>\n<td>\n[[ True False False] <br>\n [False False False]]\n</td>\n</tr>\n\n\n</table>", "_____no_output_____" ], [ "Why keep track of the position of the max? It's because this is the input value that ultimately influenced the output, and therefore the cost. Backprop is computing gradients with respect to the cost, so anything that influences the ultimate cost should have a non-zero gradient. So, backprop will \"propagate\" the gradient back to this particular input value that had influenced the cost. ", "_____no_output_____" ], [ "<a name='5-2-2'></a>\n### 5.2.2 - Average Pooling - Backward Pass \n\nIn max pooling, for each input window, all the \"influence\" on the output came from a single input value--the max. In average pooling, every element of the input window has equal influence on the output. So to implement backprop, you will now implement a helper function that reflects this.\n\nFor example if we did average pooling in the forward pass using a 2x2 filter, then the mask you'll use for the backward pass will look like: \n$$ dZ = 1 \\quad \\rightarrow \\quad dZ =\\begin{bmatrix}\n1/4 && 1/4 \\\\\n1/4 && 1/4\n\\end{bmatrix}\\tag{5}$$\n\nThis implies that each position in the $dZ$ matrix contributes equally to output because in the forward pass, we took an average. \n\n<a name='ex-7'></a>\n### Exercise 7 - distribute_value\n\nImplement the function below to equally distribute a value dz through a matrix of dimension shape. \n\n[Hint](https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.ones.html)", "_____no_output_____" ] ], [ [ "def distribute_value(dz, shape):\n \"\"\"\n Distributes the input value in the matrix of dimension shape\n \n Arguments:\n dz -- input scalar\n shape -- the shape (n_H, n_W) of the output matrix for which we want to distribute the value of dz\n \n Returns:\n a -- Array of size (n_H, n_W) for which we distributed the value of dz\n \"\"\" \n # Retrieve dimensions from shape (≈1 line)\n (n_H, n_W) = shape\n \n # Compute the value to distribute on the matrix (≈1 line)\n average = dz / (n_H * n_W)\n \n # Create a matrix where every entry is the \"average\" value (≈1 line)\n a = np.ones(shape) * average\n # YOUR CODE STARTS HERE\n \n \n # YOUR CODE ENDS HERE\n return a", "_____no_output_____" ], [ "a = distribute_value(2, (2, 2))\nprint('distributed value =', a)\n\n\nassert type(a) == np.ndarray, \"Output must be a np.ndarray\"\nassert a.shape == (2, 2), f\"Wrong shape {a.shape} != (2, 2)\"\nassert np.sum(a) == 2, \"Values must sum to 2\"\n\na = distribute_value(100, (10, 10))\nassert type(a) == np.ndarray, \"Output must be a np.ndarray\"\nassert a.shape == (10, 10), f\"Wrong shape {a.shape} != (10, 10)\"\nassert np.sum(a) == 100, \"Values must sum to 100\"\n\nprint(\"\\033[92m All tests passed.\")", "distributed value = [[0.5 0.5]\n [0.5 0.5]]\n\u001b[92m All tests passed.\n" ] ], [ [ "**Expected Output**: \n\n<table> \n<tr> \n<td>\ndistributed_value =\n</td>\n<td>\n[[ 0.5 0.5]\n<br\\> \n[ 0.5 0.5]]\n</td>\n</tr>\n</table>", "_____no_output_____" ], [ "<a name='5-2-3'></a>\n### 5.2.3 Putting it Together: Pooling Backward \n\nYou now have everything you need to compute backward propagation on a pooling layer.\n\n<a name='ex-8'></a>\n### Exercise 8 - pool_backward\n\nImplement the `pool_backward` function in both modes (`\"max\"` and `\"average\"`). You will once again use 4 for-loops (iterating over training examples, height, width, and channels). You should use an `if/elif` statement to see if the mode is equal to `'max'` or `'average'`. If it is equal to 'average' you should use the `distribute_value()` function you implemented above to create a matrix of the same shape as `a_slice`. Otherwise, the mode is equal to '`max`', and you will create a mask with `create_mask_from_window()` and multiply it by the corresponding value of dA.", "_____no_output_____" ] ], [ [ "def pool_backward(dA, cache, mode = \"max\"):\n \"\"\"\n Implements the backward pass of the pooling layer\n \n Arguments:\n dA -- gradient of cost with respect to the output of the pooling layer, same shape as A\n cache -- cache output from the forward pass of the pooling layer, contains the layer's input and hparameters \n mode -- the pooling mode you would like to use, defined as a string (\"max\" or \"average\")\n \n Returns:\n dA_prev -- gradient of cost with respect to the input of the pooling layer, same shape as A_prev\n \"\"\"\n # Retrieve information from cache (≈1 line)\n (A_prev, hparameters) = cache\n \n # Retrieve hyperparameters from \"hparameters\" (≈2 lines)\n stride = hparameters['stride']\n f = hparameters['f']\n \n # Retrieve dimensions from A_prev's shape and dA's shape (≈2 lines)\n m, n_H_prev, n_W_prev, n_C_prev = A_prev.shape\n m, n_H, n_W, n_C = dA.shape\n \n # Initialize dA_prev with zeros (≈1 line)\n dA_prev = np.zeros((m, n_H_prev, n_W_prev, n_C_prev))\n \n for i in range(m): # loop over the training examples\n \n # select training example from A_prev (≈1 line)\n a_prev = A_prev[i]\n \n for h in range(n_H): # loop on the vertical axis\n for w in range(n_W): # loop on the horizontal axis\n for c in range(n_C): # loop over the channels (depth)\n \n # Find the corners of the current \"slice\" (≈4 lines)\n vert_start = h * stride\n vert_end = h * stride + f\n horiz_start = w * stride\n horiz_end = w * stride + f\n \n # Compute the backward propagation in both modes.\n if mode == \"max\":\n \n # Use the corners and \"c\" to define the current slice from a_prev (≈1 line)\n a_prev_slice = a_prev_slice = a_prev[vert_start:vert_end, horiz_start:horiz_end, c]\n \n # Create the mask from a_prev_slice (≈1 line)\n mask = create_mask_from_window(a_prev_slice)\n\n # Set dA_prev to be dA_prev + (the mask multiplied by the correct entry of dA) (≈1 line)\n dA_prev[i, vert_start: vert_end, horiz_start: horiz_end, c] += mask * dA[i, h, w, c]\n \n elif mode == \"average\":\n \n # Get the value da from dA (≈1 line)\n da = dA[i, h, w, c]\n \n # Define the shape of the filter as fxf (≈1 line)\n shape = (f, f)\n\n # Distribute it to get the correct slice of dA_prev. i.e. Add the distributed value of da. (≈1 line)\n dA_prev[i, vert_start: vert_end, horiz_start: horiz_end, c] += distribute_value(da, shape)\n # YOUR CODE STARTS HERE\n \n \n # YOUR CODE ENDS HERE\n \n # Making sure your output shape is correct\n assert(dA_prev.shape == A_prev.shape)\n \n return dA_prev", "_____no_output_____" ], [ "np.random.seed(1)\nA_prev = np.random.randn(5, 5, 3, 2)\nhparameters = {\"stride\" : 1, \"f\": 2}\nA, cache = pool_forward(A_prev, hparameters)\nprint(A.shape)\nprint(cache[0].shape)\ndA = np.random.randn(5, 4, 2, 2)\n\ndA_prev1 = pool_backward(dA, cache, mode = \"max\")\nprint(\"mode = max\")\nprint('mean of dA = ', np.mean(dA))\nprint('dA_prev1[1,1] = ', dA_prev1[1, 1]) \nprint()\ndA_prev2 = pool_backward(dA, cache, mode = \"average\")\nprint(\"mode = average\")\nprint('mean of dA = ', np.mean(dA))\nprint('dA_prev2[1,1] = ', dA_prev2[1, 1]) \n\nassert type(dA_prev1) == np.ndarray, \"Wrong type\"\nassert dA_prev1.shape == (5, 5, 3, 2), f\"Wrong shape {dA_prev1.shape} != (5, 5, 3, 2)\"\nassert np.allclose(dA_prev1[1, 1], [[0, 0], \n [ 5.05844394, -1.68282702],\n [ 0, 0]]), \"Wrong values for mode max\"\nassert np.allclose(dA_prev2[1, 1], [[0.08485462, 0.2787552], \n [1.26461098, -0.25749373], \n [1.17975636, -0.53624893]]), \"Wrong values for mode average\"\nprint(\"\\033[92m All tests passed.\")", "(5, 4, 2, 2)\n(5, 5, 3, 2)\nmode = max\nmean of dA = 0.14571390272918056\ndA_prev1[1,1] = [[ 0. 0. ]\n [ 5.05844394 -1.68282702]\n [ 0. 0. ]]\n\nmode = average\nmean of dA = 0.14571390272918056\ndA_prev2[1,1] = [[ 0.08485462 0.2787552 ]\n [ 1.26461098 -0.25749373]\n [ 1.17975636 -0.53624893]]\n\u001b[92m All tests passed.\n" ] ], [ [ "**Expected Output**: \n\nmode = max:\n<table> \n<tr> \n<td>\n\n**mean of dA =**\n</td>\n\n<td>\n\n0.145713902729\n\n </td>\n</tr>\n\n<tr> \n<td>\ndA_prev[1,1] =\n</td>\n<td>\n[[ 0. 0. ] <br>\n [ 5.05844394 -1.68282702] <br>\n [ 0. 0. ]]\n</td>\n</tr>\n</table>\n\nmode = average\n<table> \n<tr> \n<td>\n\nmean of dA =\n</td>\n\n<td>\n\n0.145713902729\n\n </td>\n</tr>\n\n<tr> \n<td>\ndA_prev[1,1] =\n</td>\n<td>\n[[ 0.08485462 0.2787552 ] <br>\n [ 1.26461098 -0.25749373] <br>\n [ 1.17975636 -0.53624893]]\n</td>\n</tr>\n</table>", "_____no_output_____" ], [ "**Congratulations**! You've completed the assignment and its optional portion. You now understand how convolutional neural networks work, and have implemented all the building blocks of a neural network. In the next assignment you will implement a ConvNet using TensorFlow. Nicely done! See you there.", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown" ]

[ [ "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown", "markdown", "markdown" ], [ "code", "code" ], [ "markdown", "markdown" ], [ "code", "code" ], [ "markdown", "markdown" ], [ "code", "code" ], [ "markdown", "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown", "markdown", "markdown" ], [ "code", "code" ], [ "markdown", "markdown" ], [ "code", "code" ], [ "markdown", "markdown", "markdown" ], [ "code", "code" ], [ "markdown", "markdown" ], [ "code", "code" ], [ "markdown", "markdown" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "Beginning Programming with Python\n===========================\n\n\nSession 1: Getting Started\n--------------------------\n\n* The Zen of Python\n* Variables\n* Comments\n* Strings\n* Numbers\n* Booleans\n* String Methods\n* Using Variables in Strings\n* Numerical Operations\n* Working with Numerical Data\n* Using the Math library\n* Setting up Anaconda Python\n * Anaconda\n * Good Youtube Video\n \n**Assignment**\n\n1. Install Anaconda Python\nhttps://www.youtube.com/watch?v=Z1Yd7upQsXY\n2. Create GitHub account\n3. Create Codewars account\n4. Create Reddit account", "_____no_output_____" ] ], [ [ "2 + 5", "_____no_output_____" ], [ "import this", "_____no_output_____" ] ], [ [ "Variables\n---------", "_____no_output_____" ] ], [ [ "phoebe = 8", "_____no_output_____" ], [ "print(phoebe)", "8\n" ], [ "print(id(phoebe))", "4473468240\n" ], [ "maru = 7", "_____no_output_____" ], [ "id(maru)", "_____no_output_____" ], [ "4fun = 3", "_____no_output_____" ], [ "fun4 = 3", "_____no_output_____" ], [ "m = 7", "_____no_output_____" ], [ "phoebe_age_in_years = 8", "_____no_output_____" ], [ "dog_name = \"Phoebe\"", "_____no_output_____" ], [ "print(dog_name)", "Phoebe\n" ], [ "type(dog_name)", "_____no_output_____" ], [ "type(phoebe)", "_____no_output_____" ], [ "dog_name + phoebe", "_____no_output_____" ], [ "dog_name + \" is a dog\"", "_____no_output_____" ] ], [ [ "Comments\n--------", "_____no_output_____" ] ], [ [ "# I'm attempting to calculate dog years for humans\ndog = 1222", "_____no_output_____" ], [ "# Here's something\n# ... And etc.\n# and more...", "_____no_output_____" ], [ "\"\"\"Calculate the age very simply.\nI'm using the standard algorithm...\n\"\"\"\nage = 10", "_____no_output_____" ] ], [ [ "Strings\n-------", "_____no_output_____" ] ], [ [ "4 + 4", "_____no_output_____" ], [ "\"4\" + \"4\"", "_____no_output_____" ], [ "len(\"robb\")", "_____no_output_____" ], [ "dir(\"robb\")", "_____no_output_____" ], [ "\"robb\".endswith('b')", "_____no_output_____" ], [ "\"robb\".endswith('x')", "_____no_output_____" ], [ "\"Phoebe\".lower()\n", "_____no_output_____" ], [ "\"Here's a meanlingless sentence.\".split()", "_____no_output_____" ], [ "\"phoebe\"", "_____no_output_____" ], [ "print(\"hey\")", "hey\n" ], [ "name = \"Ramona\"", "_____no_output_____" ], [ "f\"{name} said thanks!\"", "_____no_output_____" ], [ "name + \" said thanks!\"", "_____no_output_____" ], [ "\"%s said thanks!\", name", "_____no_output_____" ], [ "\"{} said thanks! {}\".format(name, \"hey\")", "_____no_output_____" ] ], [ [ "Numbers\n-------", "_____no_output_____" ] ], [ [ "type(5)", "_____no_output_____" ], [ "type(5.5)", "_____no_output_____" ], [ "account_balance = 100.1111111", "_____no_output_____" ], [ "5.5 + 5\n", "_____no_output_____" ], [ "5.5 + \"5\"", "_____no_output_____" ], [ "5.5 + float(\"5\")", "_____no_output_____" ], [ "float(\"0\")", "_____no_output_____" ], [ "type(0.0)", "_____no_output_____" ], [ "float(\"x\")", "_____no_output_____" ], [ "int(\"0\")", "_____no_output_____" ], [ "int(\"x\")", "_____no_output_____" ], [ "\"x\"", "_____no_output_____" ], [ "x", "_____no_output_____" ], [ "x = 1\n", "_____no_output_____" ], [ "x", "_____no_output_____" ], [ "id(x)", "_____no_output_____" ], [ "\"x\"", "_____no_output_____" ] ], [ [ "Booleans\n--------", "_____no_output_____" ] ], [ [ "1, -1, 0 ", "_____no_output_____" ], [ "type(1)", "_____no_output_____" ], [ "False", "_____no_output_____" ], [ "True", "_____no_output_____" ], [ "1 == 2", "_____no_output_____" ], [ "1 == 1", "_____no_output_____" ], [ "not (1 == 1)", "_____no_output_____" ], [ "not (True and False)", "_____no_output_____" ], [ "(not True) or (not False)", "_____no_output_____" ], [ "if 1 == 2:\n print(\"The world's gone mad\")", "_____no_output_____" ], [ "if 1 == 1:\n print(\"No problem with that\")", "No problem with that\n" ], [ "phoebe_is_old = phoebe > 12", "_____no_output_____" ], [ "phoebe_is_old\n", "_____no_output_____" ], [ "if phoebe_is_old:\n print(\"old\")", "_____no_output_____" ], [ "phoebe", "_____no_output_____" ], [ "o = phoebe > 12", "_____no_output_____" ], [ "O = maru > 12", "_____no_output_____" ], [ "Math.PI", "_____no_output_____" ], [ "import Math", "_____no_output_____" ], [ "import math", "_____no_output_____" ], [ "math.PI", "_____no_output_____" ], [ "dir(math)\n", "_____no_output_____" ], [ "GOOGLE_ID = 12354617234567234", "_____no_output_____" ], [ "GOOGLE_ID", "_____no_output_____" ], [ "GOOGLE_ID = 'fu'", "_____no_output_____" ], [ "GOOGLE_ID", "_____no_output_____" ] ], [ [ "Numerical Data and the Math Library\n-----------------------------------", "_____no_output_____" ] ], [ [ "round(1.23456)", "_____no_output_____" ], [ "round(1.5)", "_____no_output_____" ], [ "round(1.4)", "_____no_output_____" ], [ "help(round)", "Help on built-in function round in module builtins:\n\nround(number, ndigits=None)\n Round a number to a given precision in decimal digits.\n \n The return value is an integer if ndigits is omitted or None. Otherwise\n the return value has the same type as the number. ndigits may be negative.\n\n" ], [ "round(1.23456, 2)", "_____no_output_____" ], [ "1.23 * 1.07", "_____no_output_____" ], [ "round(1.23 * 1.07, 2)", "_____no_output_____" ], [ "round('x' + 'y', 2)", "_____no_output_____" ], [ "abs(3)", "_____no_output_____" ], [ "abs(-3)", "_____no_output_____" ], [ "2 % 3", "_____no_output_____" ], [ "3 % 2", "_____no_output_____" ], [ "3 % 2 == 0", "_____no_output_____" ], [ "10 % 2 == 0", "_____no_output_____" ], [ "10 % 3 == 0", "_____no_output_____" ], [ "15 % 3 == 0", "_____no_output_____" ], [ "not(10 % 2 == 0)", "_____no_output_____" ], [ "10 % 7 != 0", "_____no_output_____" ] ], [ [ "Using the Math Library\n----------------------", "_____no_output_____" ] ], [ [ "import math", "_____no_output_____" ], [ "x", "_____no_output_____" ], [ "math", "_____no_output_____" ], [ "dir(math)", "_____no_output_____" ], [ "math", "_____no_output_____" ], [ "pi", "_____no_output_____" ], [ "math.pi", "_____no_output_____" ], [ "math.factorial(4)", "_____no_output_____" ], [ "math.factorial(10)", "_____no_output_____" ], [ "help(math.ceil)", "Help on built-in function ceil in module math:\n\nceil(x, /)\n Return the ceiling of x as an Integral.\n \n This is the smallest integer >= x.\n\n" ], [ "math.ceil(1.1)", "_____no_output_____" ], [ "math.ceil", "_____no_output_____" ], [ "type(math.ceil)", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "#backtest \nimport numpy as np\nimport pandas as pd\nfrom sklearn.model_selection import train_test_split\nfrom sklearn.ensemble import RandomForestClassifier\nfrom sklearn.metrics import precision_score, mean_squared_error, explained_variance_score, r2_score\nimport matplotlib.pyplot as plt", "_____no_output_____" ], [ "def status_calc(stock, sp500, outperformance=10):\n #A simple function to classify whether a stock outperformed the S&P500\n #:param stock: stock price\n #:param sp500: S&P500 price\n #:param outperformance: stock is classified 1 if stock price > S&P500 price + outperformance\n #:return: true/false\n \n if outperformance < 0:\n raise ValueError(\"outperformance must be positive\")\n return stock - sp500 >= outperformance", "_____no_output_____" ], [ "# Build the dataset, and drop any rows with missing values\nbacktest_df = pd.read_csv(\"ketstats_to_train.csv\", index_col=\"calendardate\")\nbacktest_df.dropna(axis=0, how=\"any\", inplace=True)\n\nfeatures = backtest_df.columns[1:-4]\n#X = backtest_df[features].values\nX = pd.DataFrame(backtest_df[features])\n\n# The labels are generated by applying the status_calc to the dataframe.\n# '1' if a stock beats the S&P500 by more than x%, else '0'. Here x is the\n# outperformance parameter, which is set to 10 by default but can be redefined.\ny = pd.DataFrame(list(\n status_calc(\n backtest_df[\"stock_p_change\"], backtest_df[\"sp500_p_change\"], outperformance=10\n )\n ))\n\n# z is required for us to track returns\nz = np.array(backtest_df[[\"stock_p_change\", \"sp500_p_change\"]])\n\n# Generate the train set and test set by randomly splitting the dataset\nX_train, X_test, y_train, y_test, z_train, z_test = train_test_split(\n X, y, z, test_size=0.1\n)\n\n# Instantiate a RandomForestClassifier with 100 trees, then fit it to the training data\nclf = RandomForestClassifier(n_estimators=100, random_state=0)\nclf.fit(X_train, y_train)\n\n# Generate the predictions, then print test set accuracy and precision\ny_pred = clf.predict(X_test)\nprint(\"Classifier performance\\n\", \"=\" * 20)\nprint(f\"Accuracy score: {clf.score(X_test, y_test): .2f}\")\nprint(f\"Precision score: {precision_score(y_test, y_pred): .2f}\")\n\n# Because y_pred is an array of 1s and 0s, the number of positive predictions\n# is equal to the sum of the array\nnum_positive_predictions = sum(y_pred)\nif num_positive_predictions < 0:\n print(\"No stocks predicted!\")\n\n# Recall that z_test stores the change in stock price in column 0, and the\n# change in S&P500 price in column 1.\n# Whenever a stock is predicted to outperform (y_pred = 1), we 'buy' that stock\n# and simultaneously `buy` the index for comparison.\nstock_returns = 1 + z_test[y_pred, 0] / 100\nmarket_returns = 1 + z_test[y_pred, 1] / 100\n\n# Calculate the average growth for each stock we predicted 'buy'\n# and the corresponding index growth\navg_predicted_stock_growth = sum(stock_returns) / num_positive_predictions\nindex_growth = sum(market_returns) / num_positive_predictions\npercentage_stock_returns = 100 * (avg_predicted_stock_growth - 1)\npercentage_market_returns = 100 * (index_growth - 1)\ntotal_outperformance = percentage_stock_returns - percentage_market_returns\n\nprint(\"\\n Stock prediction performance report \\n\", \"=\" * 40)\nprint(f\"Total Trades:\", num_positive_predictions)\nprint(f\"Average return for stock predictions: {percentage_stock_returns: .1f} %\")\nprint(\n f\"Average market return in the same period: {percentage_market_returns: .1f}% \"\n)\nprint(\n f\"Compared to the index, our strategy earns {total_outperformance: .1f} percentage points more\"\n)", "<ipython-input-6-330fb1fcabdc>:28: DataConversionWarning: A column-vector y was passed when a 1d array was expected. Please change the shape of y to (n_samples,), for example using ravel().\n clf.fit(X_train, y_train)\n" ], [ "#Feature Importance\nfeature_importances = pd.Series(clf.feature_importances_, index=X.columns)\nfeature_importances.sort_values(inplace=True)\nax = feature_importances.plot.barh()\nax.set(ylabel='Importance (Gini Coefficient)', title='Feature importances');\nax.set_title('Feature Importance (MDI)')\n\n#RMSE\nprint('Root Mean Squared Error:', \n np.sqrt(mean_squared_error(y_test, y_pred)))", "Root Mean Squared Error: 0.12698699833659\n" ], [ "# The percentage by which a stock has to beat the S&P500 to be considered a 'buy'\nOUTPERFORMANCE = 10\n\ndef build_data_set():\n \"\"\"\n Reads the keystats.csv file and prepares it for scikit-learn\n :return: X_train and y_train numpy arrays\n \"\"\"\n training_data = pd.read_csv(\"ketstats_to_train.csv\", index_col=\"calendardate\")\n training_data.dropna(axis=0, how=\"any\", inplace=True)\n features = training_data.columns[1:-4]\n\n X_train = training_data[features].values\n # Generate the labels: '1' if a stock beats the S&P500 by more than 10%, else '0'.\n y_train = list(\n status_calc(\n training_data[\"stock_p_change\"],\n training_data[\"sp500_p_change\"],\n OUTPERFORMANCE,\n )\n )\n\n return X_train, y_train\n\ndef predict_stocks():\n X_train, y_train = build_data_set()\n # Remove the random_state parameter to generate actual predictions\n clf = RandomForestClassifier(n_estimators=100, random_state=0)\n clf.fit(X_train, y_train)\n\n # Now we get the actual data from which we want to generate predictions.\n data = pd.read_csv(\"forward_sample.csv\", index_col=\"calendardate\")\n data.dropna(axis=0, how=\"any\", inplace=True)\n features = data.columns[1:-4]\n X_test = data[features].values\n z = data[\"ticker\"].values\n\n # Get the predicted tickers\n y_pred = clf.predict(X_test)\n if sum(y_pred) == 0:\n print(\"No stocks predicted!\")\n else:\n invest_list = z[y_pred].tolist()\n print(\n f\"{len(invest_list)} stocks predicted to outperform the S&P500 by more than {OUTPERFORMANCE}%:\"\n )\n print(\" \".join(invest_list))\n return invest_list", "_____no_output_____" ], [ "invest_list = predict_stocks()\ninvest_list", "7 stocks predicted to outperform the S&P500 by more than 10%:\nBE FMI STRP TWLO ZM SLP DKNG\n" ], [ "metadata = pd.read_csv(\"Tickers and Metadata.csv\")\ninvest_list_metadata = metadata[metadata.ticker.isin(invest_list)]\ninvest_list_metadata[[\"ticker\",\"name\",\"category\",\n \"sicindustry\",\"scalemarketcap\",\n \"scalerevenue\",\"location\"]]", "_____no_output_____" ], [ "pd.DataFrame(invest_list).to_csv(\"invest_list.csv\")", "_____no_output_____" ], [ "data = pd.read_csv(\"stock_prices.csv\")\ndata[['BE', 'FMI', 'STRP', 'TWLO', 'ZM', 'SLP', 'DKNG']].tail(10)", "_____no_output_____" ] ] ]

[ "code" ]

[ [ "code", "code", "code", "code", "code", "code", "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "import queue\nimport multiprocessing as mp\nimport concurrent.futures as cf\n\nfrom queue import Queue, SimpleQueue\nfrom concurrent.futures import ThreadPoolExecutor, ProcessPoolExecutor\n\nfrom datetime import datetime, timedelta\n\nimport requests", "_____no_output_____" ] ], [ [ "# `concurrent.futures`\n\nThis lesson has a strange name. `concurrent.futures` is the name of a (relative) modern package in the Python standard library. It's a package with a beautiful and Pythonic API that abstracts us from the low level mechanisms of concurrency.\n\n**`concurrent.futures` should be your default choice for concurrent programming as much as possible**\n\nIn this tutorial, we started from the low levels `threading` and `multiprocessing` because we wanted to explain the concepts behind concurrency, but `concurrent.futures` offers a much safer and intuitive API. Let's start with it.\n\n## Executors and futures\n\n#### Executors\nExecutors are the entry points of `cf`. They are similar to `multiprocessing.Pool`s. Once an executor has been instantiated, we can `submit` jobs, or even `map` tasks, similar to `multiprocessin.Pool.map`. `concurrent.futures.Executor` is an abstract class. `cf` includes two concrete classes: `ThreadPoolExecutor` and `ProcessPoolExecutor`. This means that we can keep the same interface, but use completely different mechanisms just by changing the executor type we're using:", "_____no_output_____" ] ], [ [ "def check_price(exchange, symbol, date):\n base_url = \"http://localhost:5000\"\n resp = requests.get(f\"{base_url}/price/{exchange}/{symbol}/{date}\")\n return resp.json()", "_____no_output_____" ], [ "with ThreadPoolExecutor(max_workers=10) as ex:\n future = ex.submit(check_price, 'bitstamp', 'btc', '2020-04-01')\n print(f\"Price: ${future.result()['close']}\")", "Price: $6421.14\n" ], [ "with ProcessPoolExecutor(max_workers=10, mp_context=mp.get_context('fork')) as ex:\n future = ex.submit(check_price, 'bitstamp', 'btc', '2020-04-01')\n print(f\"Price: ${future.result()['close']}\")", "Price: $6421.14\n" ] ], [ [ "This is the beauty of `cf`: we're using the same logic with two completely different executors; the API is the same.\n\n#### Futures\n\nAs you can see from the the examples above, the `submit` method returns immediately a `Future` object. These objects are an abstraction of a task that is being processed. They have multiple useful methods that we can use (as seen in the following example). The most important one, `result(timeout=None)` will block for `timeout` seconds until a result was produced:", "_____no_output_____" ] ], [ [ "with ThreadPoolExecutor(max_workers=10) as ex:\n future = ex.submit(check_price, 'bitstamp', 'btc', '2020-04-01')\n print(future.done())\n print(f\"Price: ${future.result()['close']}\")\n print(future.done())", "False\nPrice: $6421.14\nTrue\n" ] ], [ [ "#### The `map` method\n\nExecutors have a `map` method that is similar to `mp.Pool.map`, it's convenient as there are no futures to work with, but it's limited as only one parameter can be passed:", "_____no_output_____" ] ], [ [ "EXCHANGES = ['bitfinex', 'bitstamp', 'kraken']", "_____no_output_____" ], [ "def check_price_tuple(arg):\n exchange, symbol, date = arg\n base_url = \"http://localhost:5000\"\n resp = requests.get(f\"{base_url}/price/{exchange}/{symbol}/{date}\")\n return resp.json()", "_____no_output_____" ], [ "with ThreadPoolExecutor(max_workers=10) as ex:\n results = ex.map(check_price_tuple, [\n (exchange, 'btc', '2020-04-01')\n for exchange in EXCHANGES\n ])\n print([price['close'] for price in results])", "[6409.8, 6421.14, 6401.9]\n" ], [ "('bitstamp', 'btc', '2020-04-01')", "_____no_output_____" ] ], [ [ "As you can see, we had to define a new special function that works by receiving a tuple instead of the individual elements.\n\n#### `submit` & `as_completed` pattern\n\nTo overcome the limitation of `Executor.map`, we can use a common pattern of creating multiple futures with `Executor.submit` and waiting for them to complete with the module-level function `concurrent.futures.as_completed`:", "_____no_output_____" ] ], [ [ "with ThreadPoolExecutor(max_workers=10) as ex:\n futures = {\n ex.submit(check_price, exchange, 'btc', '2020-04-01'): exchange\n for exchange in EXCHANGES\n }\n for future in cf.as_completed(futures):\n exchange = futures[future]\n print(f\"{exchange.title()}: ${future.result()['close']}\")", "Kraken: $6401.9\nBitfinex: $6409.8\nBitstamp: $6421.14\n" ] ], [ [ "## Producer/Consumer with `concurrent.futures`\n\nI'll show you an example of the producer/consumer pattern using the `cf` module. There are multiple ways to create this pattern, I'll stick to the basics.", "_____no_output_____" ] ], [ [ "BASE_URL = \"http://localhost:5000\"", "_____no_output_____" ], [ "resp = requests.get(f\"{BASE_URL}/exchanges\")", "_____no_output_____" ], [ "EXCHANGES = resp.json()\nEXCHANGES[:3]", "_____no_output_____" ], [ "START_DATE = datetime(2020, 3, 1)", "_____no_output_____" ], [ "DATES = [(START_DATE + timedelta(days=i)).strftime('%Y-%m-%d') for i in range(31)]", "_____no_output_____" ], [ "DATES[:3]", "_____no_output_____" ], [ "resp = requests.get(f\"{BASE_URL}/symbols\")", "_____no_output_____" ], [ "SYMBOLS = resp.json()\nSYMBOLS", "_____no_output_____" ] ], [ [ "Queues:", "_____no_output_____" ] ], [ [ "work_to_do = Queue()\nwork_done = SimpleQueue()", "_____no_output_____" ], [ "for exchange in EXCHANGES:\n for date in DATES:\n for symbol in SYMBOLS:\n task = {\n 'exchange': exchange,\n 'symbol': symbol,\n 'date': date,\n }\n work_to_do.put(task)", "_____no_output_____" ], [ "work_to_do.qsize()", "_____no_output_____" ], [ "def worker(task_queue, results_queue):\n while True:\n try:\n task = task_queue.get(block=False)\n except queue.Empty:\n print('Queue is empty! My work here is done. Exiting.')\n return\n exchange, symbol, date = task['exchange'], task['symbol'], task['date']\n price = check_price(exchange, symbol, date)\n results_queue.put((price, exchange, symbol, date))\n task_queue.task_done()", "_____no_output_____" ], [ "with ThreadPoolExecutor(max_workers=32) as ex:\n futures = [\n ex.submit(worker, work_to_do, work_done) for _ in range(32)\n ]\n work_to_do.join()", "Queue is empty! My work here is done. Exiting.\nQueue is empty! My work here is done. Exiting.\nQueue is empty! My work here is done. Exiting.Queue is empty! My work here is done. Exiting.\n\nQueue is empty! My work here is done. Exiting.\nQueue is empty! My work here is done. Exiting.\nQueue is empty! My work here is done. Exiting.Queue is empty! My work here is done. Exiting.\n\nQueue is empty! My work here is done. Exiting.\nQueue is empty! My work here is done. Exiting.\nQueue is empty! My work here is done. Exiting.\nQueue is empty! My work here is done. Exiting.\nQueue is empty! My work here is done. Exiting.\nQueue is empty! My work here is done. Exiting.\nQueue is empty! My work here is done. Exiting.\nQueue is empty! My work here is done. Exiting.\nQueue is empty! My work here is done. Exiting.Queue is empty! My work here is done. Exiting.\n\nQueue is empty! My work here is done. Exiting.\nQueue is empty! My work here is done. Exiting.\nQueue is empty! My work here is done. Exiting.\nQueue is empty! My work here is done. Exiting.\nQueue is empty! My work here is done. Exiting.\nQueue is empty! My work here is done. Exiting.\nQueue is empty! My work here is done. Exiting.\nQueue is empty! My work here is done. Exiting.\nQueue is empty! My work here is done. Exiting.Queue is empty! My work here is done. Exiting.\n\nQueue is empty! My work here is done. Exiting.\nQueue is empty! My work here is done. Exiting.\nQueue is empty! My work here is done. Exiting.\nQueue is empty! My work here is done. Exiting.\n" ], [ "all([f.done() for f in futures])", "_____no_output_____" ], [ "work_done.qsize()", "_____no_output_____" ], [ "results = {}", "_____no_output_____" ], [ "while True:\n try:\n price, exchange, symbol, date = work_done.get(block=None)\n results.setdefault(exchange, {})\n results[exchange].setdefault(date, {})\n results[exchange][date][symbol] = price['close'] if price else None\n except queue.Empty:\n break", "_____no_output_____" ], [ "results['bitfinex']['2020-03-10']['btc']", "_____no_output_____" ], [ "results['bitstamp']['2020-03-10']['btc']", "_____no_output_____" ], [ "results['coinbase-pro']['2020-03-10']['btc']", "_____no_output_____" ] ], [ [ "## Summary\n\nThe `concurrent.futures` module is the most abstract, highest level concurrency module in the Python standard library and **it SHOULD be your default option** when writing concurrent code. Only if you need more advanced capabilities, you should use the `threading` or `multiprocessing` modules directly.", "_____no_output_____" ] ] ]

[ "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown" ]

[ [ "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ], [ "markdown" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "## Basic Ensemble Learinng : Hard/Soft Voting/Bagging - OOB/bootstraps/bootstrap_features", "_____no_output_____" ] ], [ [ "##-----------------\n## Copyright in private\n## Modify History :\n## 2018 - 9 - 24\n## Purpose:\n## 1. 集成学习分类器 构建 - 集成多个子模型来投票，提供准确率,理论上子模型越多，整体模型的准确率将很高！\n## 2. Hard（少数服从多数） and Soft voting classifier \n## \n## 3. 为提高每个子模型的差异性。希望每个子模型只看一部分的数据样本。 在看样本的形式上可以分为：放回取样（bagging）和不放回取样方法(pasting)\n## 放回取样的方法，整体有30% 的数据是取不到， 参数bootstraps \n## 4. n_estimators=500 个不同的子模型构成集成学习，而且模型之见存在差异，就构成了随机森林\n## \n## Parameters:\n## \n## ", "_____no_output_____" ], [ "from sklearn import datasets\nimport numpy as np\nimport matplotlib.pyplot as plt\n# help(make_moons)", "_____no_output_____" ], [ "## data sets \nX,y = datasets.make_moons(n_samples = 1200, noise = 0.25, random_state = 100)", "_____no_output_____" ], [ "X.shape", "_____no_output_____" ], [ "y.shape", "_____no_output_____" ], [ "#plot the show \nplt.scatter(X[y == 0,0],X [y == 0,1])\nplt.scatter(X[y == 1,0],X [y == 1,1])\nplt.show()", "_____no_output_____" ], [ "# try to split data into test and train data sets \nfrom sklearn.model_selection import train_test_split\nX_train,X_test,y_train,y_test =train_test_split(X,y)", "_____no_output_____" ] ], [ [ "## 1. EnsembleLearning Classifier ", "_____no_output_____" ] ], [ [ "from sklearn.linear_model import LogisticRegression\n\n# Logistic Regression \nlog_clf = LogisticRegression()\nlog_clf.fit(X_train,y_train)\n#log_clf.score(X_test,y_test) # 0.83\n\n# SVM\nfrom sklearn.svm import SVC\nsvc_clf = SVC()\nsvc_clf.fit(X_train,y_train)\n#svc_clf.score(X_test,y_test) #0.98\n\n# Decision Tree\nfrom sklearn.tree import DecisionTreeClassifier \ndt_clf = DecisionTreeClassifier(random_state = 100) # max_depth = 2,criterion = 'gini'\ndt_clf.fit(X_train,y_train)\n#dt_clf.score(X_test,y_test) # 1.0", "_____no_output_____" ], [ "## test on data sets \nlog_clf.score(X_test,y_test) # 0.88\n", "_____no_output_____" ], [ "svc_clf.score(X_test,y_test) #0.94\n", "_____no_output_____" ], [ "dt_clf.score(X_test,y_test) # 0.92", "_____no_output_____" ] ], [ [ "### 1.1 Ensemble Leaning ", "_____no_output_____" ], [ "### 1.1.1集成学习的效果- 人多力量大，明显比单个分类器的分类结果准确率要高", "_____no_output_____" ] ], [ [ "predict_1 = log_clf.predict(X_test)\npredict_2 = svc_clf.predict(X_test)\npredict_3 = dt_clf.predict(X_test)\n\n# \npredict_y = np.array((predict_1 + predict_2 + predict_3) >=2 ,dtype = 'int')\n", "_____no_output_____" ], [ "predict_y[:10]", "_____no_output_____" ], [ "from sklearn.metrics import accuracy_score\naccuracy_score(y_test,predict_y)", "_____no_output_____" ] ], [ [ "## 2. Voting Classifier - Hard Voting - 少数服从多数准则", "_____no_output_____" ] ], [ [ "# Harding voting - 少数服从多数\nfrom sklearn.ensemble import VotingClassifier \n\nvoting_clf_hard = VotingClassifier(estimators = [\n ('log_clf',LogisticRegression()),\n ('svm_clf',SVC()),\n ('dt_clf',DecisionTreeClassifier(random_state = 200))], voting = 'hard') ", "_____no_output_____" ], [ "# test score by hard voting classifier \nvoting_clf_hard.fit(X_train,y_train)\nvoting_clf_hard.score(X_test,y_test)", "c:\\users\\h155809\\appdata\\local\\programs\\python\\python36\\lib\\site-packages\\sklearn\\preprocessing\\label.py:151: DeprecationWarning: The truth value of an empty array is ambiguous. Returning False, but in future this will result in an error. Use `array.size > 0` to check that an array is not empty.\n if diff:\n" ] ], [ [ "## 3. Voting Classifier - Soft Voting -考虑以分类的概率作为权值 来分类", "_____no_output_____" ] ], [ [ "from sklearn.ensemble import VotingClassifier \n\nvoting_clf_soft = VotingClassifier(estimators = [\n ('log_clf', LogisticRegression()),\n ('svc_clf',SVC(probability=True)),\n ('dt_clf', DecisionTreeClassifier(random_state = 200))],voting = 'soft')", "_____no_output_____" ], [ "voting_clf_soft.fit(X_train,y_train)\nvoting_clf_soft.score(X_test,y_test)", "c:\\users\\h155809\\appdata\\local\\programs\\python\\python36\\lib\\site-packages\\sklearn\\preprocessing\\label.py:151: DeprecationWarning: The truth value of an empty array is ambiguous. Returning False, but in future this will result in an error. Use `array.size > 0` to check that an array is not empty.\n if diff:\n" ] ], [ [ "## 4. 放回的取样方法 - Bagging", "_____no_output_____" ] ], [ [ "# 提高子模型的差异可以提高整体模型的准确率，并且可以让模型看不同数量的样本数量。\n# 在取样本时，分为放回取样方法- Bagging and 不放回取样 pasting \nfrom sklearn.tree import DecisionTreeClassifier\nfrom sklearn.ensemble import BaggingClassifier", "_____no_output_____" ], [ "# 以DecisionTree 为例\n# n_estimators : 集成多少个分类器模型\n# max_sample: 分类器一次看多少数据\n# bootstrap = True： 可放回取样\n\nbagging_clf = BaggingClassifier(\n DecisionTreeClassifier(),\n n_estimators = 100,max_samples = 100, bootstrap = True\n)\n\nbagging_clf.fit(X_train,y_train)\nbagging_clf.score(X_test,y_test)", "_____no_output_____" ] ], [ [ "### 4.1 放回取样方法Bagging - Out of Bag(OOB) 有大约30%的数据取不到", "_____no_output_____" ] ], [ [ "# 有放回的取样本，并通过 oob_score = True 来确定标记没有被取到的样本数量。并将这些样本用于测试样本准确度\nbagging_clf = BaggingClassifier(\n DecisionTreeClassifier(),\n n_estimators = 100,max_samples = 100, bootstrap = True,oob_score = True\n)\n\nbagging_clf.fit(X_train,y_train)\nbagging_clf.score(X_test,y_test)", "_____no_output_____" ], [ "# test model on out of bag data\nbagging_clf.oob_score_", "_____no_output_____" ] ], [ [ "### 4.2 放回取样方法Bagging - Out of Bag(OOB) - 基于样本特征的取样方法 - bootstrap_features", "_____no_output_____" ] ], [ [ "## 有放回的取样本 , 基于样本特征的取样方法，并标记最大取样特征数量max_features\n##", "_____no_output_____" ], [ "from sklearn.tree import DecisionTreeClassifier\nfrom sklearn.ensemble import BaggingClassifier\n\n# max_features - 每次取样特征数量\n# bootstrap_features = True 基于特征的取样方法\n\nbagging_clf_features = BaggingClassifier(\n DecisionTreeClassifier(),\n n_estimators = 100,max_samples = 100, bootstrap = True,oob_score = True,max_features = 2, bootstrap_features = True\n)\n\nbagging_clf_features.fit(X_train,y_train)\nbagging_clf_features.score(X_test,y_test)", "_____no_output_____" ], [ "# test model on out of bag data \nbagging_clf_features.oob_score_", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code" ], [ "markdown", "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "%matplotlib inline\nimport numpy as np\nimport matplotlib.pyplot as plt\nfrom IPython.html.widgets import interact\n\nfrom sklearn.datasets import load_digits\ndigits = load_digits()", "_____no_output_____" ], [ "def sigmoid(x):\n return 1/(1 + np.exp(-x))\n\nsigmoid_v = np.vectorize(sigmoid)\n\ndef sigmoidprime(x):\n return sigmoid(x) * (1 - sigmoid(x))\n\nsigmoidprime_v = np.vectorize(sigmoidprime)", "_____no_output_____" ], [ "size = [64, 20, 10]\n\nweights = []\nfor n in range(1, len(size)):\n weights.append(np.random.rand(size[n-1], size[n]) * 2 - 1)\n\nbiases = []\nfor n in range(1, len(size)):\n biases.append(np.random.rand(size[n]) * 2 - 1)\n\ntrainingdata = digits.data[0:1200]\ntraininganswers = digits.target[0:1200]\nlc = 0.02\n\n#convert the integer answers into a 10-dimension array\ntraininganswervectors = np.zeros((1796,10))\nfor n in range(1796):\n traininganswervectors[n][digits.target[n]] = 1", "_____no_output_____" ], [ "def feedforward(weights, biases, a):\n b = []\n #first element is inputs \"a\"\n b.append(a)\n for n in range(1, len(size)):\n #all other elements depend on the number of neurons\n b.append(np.zeros(size[n]))\n for n2 in range(0, size[n]):\n b[n][n2] = sigmoid_v(np.dot(weights[n-1][0:,n2], b[n-1]) + biases[n-1][n2])\n \n return b", "_____no_output_____" ], [ "feedforward(weights, biases, trainingdata[0])", "_____no_output_____" ], [ "def gradient_descent(weights, biases, inputs, answers, batchsize, lc, epochs):\n for n in range(epochs):\n #pick random locations for input/result data\n locations = np.random.randint(0, len(inputs), batchsize)\n minibatch = []\n #create tuples (inputs, result) based on random locations\n for n2 in range(batchsize):\n minibatch.append((inputs[locations[n2]], answers[locations[n2]]))\n for n3 in range(batchsize):\n weights, biases = train(weights, biases, minibatch, lc)\n \n \n results = []\n for n4 in range(len(trainingdata)):\n results.append(feedforward(weights, biases, inputs[n4])[-1])\n \n accresult = accuracy(inputs, results, answers)\n print(\"Epoch \", n, \" : \", accresult)\n \n return weights, biases", "_____no_output_____" ], [ "def train(weights, biases, minibatch, lc):\n #set the nabla functions to be the functions themselves initially, same size\n nb = [np.zeros(b.shape) for b in biases]\n nw = [np.zeros(w.shape) for w in weights]\n #largely taken from Michael Nielsen's implementation\n for i, r in minibatch:\n dnb, dnw = backprop(weights, biases, i, r)\n nb = [a+b for a, b in zip(nb, dnb)]\n nw = [a+b for a, b in zip(nw, dnw)]\n \n weights = [w-(lc/len(minibatch))*n_w for w, n_w in zip(weights, nw)]\n biases = [b-(lc/len(minibatch))*n_b for b, n_b in zip(biases, nb)]\n return weights, biases", "_____no_output_____" ], [ "def backprop(weights, biases, inputs, answers):\n #set the nabla functions to be the same size as functions\n nb = [np.zeros(b.shape) for b in biases]\n nw = [np.zeros(w.shape) for w in weights]\n a = inputs\n alist = [inputs]\n zlist = []\n #from feedforward\n for n in range(1, len(size)):\n #all other elements depend on the number of neurons\n zlist.append(np.zeros(size[n]))\n alist.append(np.zeros(size[n]))\n for n2 in range(1, size[n]):\n zlist[n-1][n2] = np.dot(weights[n-1][0:,n2], alist[n-1]) + biases[n-1][n2]\n alist[n][n2] = sigmoid_v(alist[n-1][n2])\n \n delta = costderivative(alist[-1], answers) * sigmoidprime_v(zlist[-1])\n nb[-1] = delta\n #different from MN, alist[-2] not same size as delta?\n nw[-1] = np.dot(delta, alist[-1].transpose())\n \n for n in range(2, len(size)):\n delta = np.dot(weights[-n+1], delta) * sigmoidprime_v(zlist[-n])\n nb[-n] = delta\n #same here\n nw[-n] = np.dot(delta, alist[-n].transpose())\n \n return nb, nw", "_____no_output_____" ], [ "def costderivative(output, answers):\n return (output - answers)", "_____no_output_____" ], [ "def accuracy(inputs, results, answers):\n correct = 0\n binresults = results\n for n in range(0, len(results)):\n #converts the output into a binary y/n for each digit\n for n2 in range(len(results[n])):\n if results[n][n2] == np.amax(results[n]):\n binresults[n][n2] = 1\n else:\n binresults[n][n2] = 0\n \n if np.array_equal(answers[n], binresults[n]):\n correct += 1\n return correct / len(results)", "_____no_output_____" ], [ "size = [64, 20, 10]\n\nweights = []\nfor n in range(1, len(size)):\n weights.append(np.random.rand(size[n-1], size[n]) * 2 - 1)\n\nbiases = []\nfor n in range(1, len(size)):\n biases.append(np.random.rand(size[n]) * 2 - 1)\n\ntrainingdata = digits.data[0:1000]\ntraininganswers = digits.target[0:1000]\n\ntraininganswervectors = np.zeros((1000,10))\nfor n in range(1000):\n traininganswervectors[n][digits.target[n]] = 1", "_____no_output_____" ], [ "final_weights, final_biases = gradient_descent(weights, biases, trainingdata,\n traininganswervectors, 5, 1, 30)\n\nprint(final_weights)", "Epoch 0 : 0.048\nEpoch 1 : 0.093\nEpoch 2 : 0.095\nEpoch 3 : 0.101\nEpoch 4 : 0.097\nEpoch 5 : 0.095\nEpoch 6 : 0.101\nEpoch 7 : 0.101\nEpoch 8 : 0.105\nEpoch 9 : 0.091\nEpoch 10 : 0.09\nEpoch 11 : 0.091\nEpoch 12 : 0.091\nEpoch 13 : 0.09\nEpoch 14 : 0.092\nEpoch 15 : 0.091\nEpoch 16 : 0.091\nEpoch 17 : 0.091\nEpoch 18 : 0.09\nEpoch 19 : 0.09\nEpoch 20 : 0.089\nEpoch 21 : 0.089\nEpoch 22 : 0.081\nEpoch 23 : 0.14\nEpoch 24 : 0.139\nEpoch 25 : 0.125\nEpoch 26 : 0.117\nEpoch 27 : 0.113\nEpoch 28 : 0.114\nEpoch 29 : 0.114\n[array([[ 0.76450383, 0.59251208, -0.33663917, ..., 0.03105159,\n -0.50664191, 0.55243318],\n [-0.32381734, -0.19473504, 0.47964496, ..., -0.74232456,\n 0.63391058, 0.11945287],\n [ 0.6231575 , 0.86058574, 0.88342131, ..., -0.22242994,\n 0.25655237, -0.23629923],\n ..., \n [-0.36932122, -0.3959529 , 0.81687002, ..., -0.97764035,\n -0.32230678, 0.12894721],\n [ 0.3298412 , 0.6209686 , 0.84756657, ..., -0.12449487,\n 0.39450935, -0.18668442],\n [-0.95583099, 0.08569311, 0.71531 , ..., 0.87901573,\n -0.25818984, -0.67319341]]), array([[ 0.14131349, -1.45527079, -0.54529945, -1.0817927 , -1.63675429,\n -0.82509237, -0.50577114, -1.25099751, -1.57167739, -1.58229838],\n [-0.26268802, -0.46828519, -0.27403453, -0.27378482, -0.11569697,\n -1.14896415, -1.0756586 , -1.00494417, -1.1274975 , -1.18015684],\n [-0.02772253, 0.15362551, -0.73992948, -0.38885413, 0.10669895,\n -1.04529539, -0.51262275, 0.04903211, 0.07113167, -0.95034216],\n [-1.10066271, -0.4192437 , -0.83153843, -0.08715551, 0.0762273 ,\n -1.61585745, -0.07269733, 0.22141072, -1.47321102, -1.06635828],\n [-1.53605171, -0.20236728, -1.59564818, -0.7736486 , -0.85743672,\n -1.65075395, 0.22338501, -0.41452082, -1.19489058, -0.28412976],\n [-0.97130133, -1.26394292, -0.86485605, 0.21719278, 0.04154922,\n -0.56971456, -1.64255665, -1.71024067, -1.65795937, -1.58724435],\n [-1.17607687, -1.66427757, -1.26226142, 0.03540448, -0.09032715,\n -0.57956052, -1.30498373, 0.19528167, -0.61780775, -1.71499853],\n [-1.32944467, -0.58069153, -0.7444898 , 0.13856652, -1.36328519,\n -0.13302136, -1.30499062, -0.1462735 , 0.09600149, -1.51221257],\n [-1.2194706 , -0.23253221, -0.66908131, -0.61349542, -0.69434693,\n -0.5319174 , -0.53994301, -1.40761353, -0.77824655, -0.47336865],\n [-0.57874113, -1.45537269, -0.52541559, 0.00194201, -1.72241672,\n -0.15125062, -1.50899977, -1.31169813, -0.38594925, 0.17597085],\n [-1.15019981, 0.15592889, -0.31838663, -0.85503774, -1.15652188,\n -0.19242718, -0.88554113, -0.27003906, -0.59227189, -0.38658356],\n [-0.19594764, -0.27451577, 0.12450115, -1.16536117, 0.22683144,\n -1.45717936, 0.05982142, -0.48704248, 0.16055401, 0.07586088],\n [-1.31938199, -0.15037955, -0.54360006, -0.16815748, -0.53387818,\n -1.06238962, -0.71740935, -1.72122009, -0.43972491, -0.02027598],\n [-1.5115723 , -0.84107486, -1.14661439, -1.11010468, -0.39087975,\n 0.10574395, -1.70313065, -1.37999991, -1.2981514 , -0.4824089 ],\n [ 0.17303988, -0.90082792, -0.36351988, -1.04551961, -0.80590839,\n 0.18561133, 0.1314899 , -0.74618971, -0.60030412, -1.45901202],\n [ 0.25517938, -1.462488 , -1.29089867, -0.52794909, 0.10719908,\n -0.02585236, -0.37550124, -0.59261764, -1.13334063, -0.21033767],\n [-0.43664434, -1.39479159, -1.69268381, -0.94948049, -1.27133524,\n -0.70130297, -1.33301927, -0.81116035, -1.07527586, -1.66143407],\n [-0.45148065, -0.91159968, -1.03189045, -1.10330699, -0.22954374,\n -1.23667512, -1.15779083, -1.13007106, 0.11576962, -1.59387454],\n [-1.2342925 , 0.22825142, -0.52002247, -0.94077492, -1.31173937,\n -0.04128284, -0.60117536, -0.52280868, -0.14590897, -0.16754535],\n [-1.38041727, 0.0245013 , -0.88616725, -1.41358959, -0.71826232,\n -0.9898886 , -1.22985433, -0.52484158, -1.2708501 , -1.64708579]])]\n" ] ] ]

[ "code" ]

[ [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "import gym", "_____no_output_____" ], [ "env = gym.make('CartPole-v0')", "/usr/local/lib/python3.6/dist-packages/gym/envs/registration.py:14: PkgResourcesDeprecationWarning: Parameters to load are deprecated. Call .resolve and .require separately.\n result = entry_point.load(False)\n" ], [ "env", "_____no_output_____" ], [ "env.reset()", "_____no_output_____" ], [ "env.action_space", "_____no_output_____" ], [ "env.step(0)", "_____no_output_____" ], [ "env.step(0)", "_____no_output_____" ], [ "env.step(0)", "_____no_output_____" ], [ "\nenv.step(0)", "_____no_output_____" ], [ "env.step(0)", "_____no_output_____" ], [ "env.step(0)", "_____no_output_____" ], [ "env.step(0)", "_____no_output_____" ], [ "env.step(0)", "_____no_output_____" ], [ "env.step(0)", "_____no_output_____" ], [ "env.step(0)", "\u001b[33mWARN: You are calling 'step()' even though this environment has already returned done = True. You should always call 'reset()' once you receive 'done = True' -- any further steps are undefined behavior.\u001b[0m\n" ] ], [ [ "here true says is is stopped and we can say drone is hit something and stopped we can say ", "_____no_output_____" ] ], [ [ "env.step(0)", "_____no_output_____" ] ], [ [ "multiarmbandit", "_____no_output_____" ] ], [ [ "env = MultiArmedBandit()", "_____no_output_____" ], [ "", "_____no_output_____" ] ] ]

[ "code", "markdown", "code", "markdown", "code" ]

[ [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "from IPython.core.display import display, HTML\ndisplay(HTML(\"<style>.container { width:100% !important; }</style>\"))", "_____no_output_____" ], [ "# This will reload imports before executing code, allowing you to easily change contents of custom scripts\n%load_ext autoreload\n%autoreload 2", "_____no_output_____" ] ], [ [ "# Stay classification: MBC evaluation\n\n**31.08.2020**", "_____no_output_____" ] ], [ [ "import numpy as np\nimport pandas as pd", "_____no_output_____" ], [ "import os, sys\nsys.path.append('/home/sandm/Notebooks/stay_classification/src/')", "_____no_output_____" ], [ "%matplotlib inline\nimport matplotlib.pyplot as plt", "_____no_output_____" ], [ "from synthetic_data.trajectory_class import get_pickle_trajectory\nfrom synthetic_data.trajectory import get_stay_segs, get_adjusted_stays", "_____no_output_____" ], [ "from stay_classification.metric_box_classifier.metric_box_classifier import stay_classifier_testing", "_____no_output_____" ] ], [ [ "# Batch data evaluation", "_____no_output_____" ] ], [ [ "dsec = 1/3600.0\nt_total = np.arange(0,24,dsec)", "_____no_output_____" ], [ "time_thresh = 1/6\ndist_thresh=0.25", "_____no_output_____" ], [ "nr_stays = 3", "_____no_output_____" ], [ "data_dir = os.path.abspath('../../')+f\"/classifiers_playground/metric_box_classifier/testdata_training_set__canonical_{nr_stays}stays/\"\nos.path.isdir(data_dir)\n\nimport glob\n#data_dir = os.path.abspath('../../')+f\"/testdata/testdata_training_set__general/\"\nos.path.isdir(data_dir)\npkls = glob.glob(data_dir + \"*.pkl\")", "_____no_output_____" ], [ "from stay_classification.metrics_etc import eval_synth_data\nfrom stay_classification.metrics import get_segments_scores, get_segments_errs\nfrom stay_classification.metrics_cluster_tools import get_pred_labels, get_labels_from_clusters", "_____no_output_____" ], [ "from synthetic_data.trajectory import get_stay_indices, get_adjusted_stays\n\nget_err = lambda trues, preds: np.sum(abs(trues-preds))/trues.size", "_____no_output_____" ] ], [ [ "## Load, Classify, Measure", "_____no_output_____" ] ], [ [ "# For the correct nr of stays\nlens3 = []\nprecs3, a_precs3, w_precs3 = [], [], []\nrecs3, a_recs3, w_recs3 = [], [], []\nerrs3, a_errs3, w_errs3 = [], [], []\n\n# For the incorrect nr of stays\nlens = []\nprecs, a_precs, w_precs = [], [], []\nrecs, a_recs, w_recs = [], [], []\nerrs, a_errs, w_errs = [], [], []\n\n\nbad_list = []\n\nprecrec_limit = 0.80\n\nii = 0\n\nlength_criterion_break = False\niqr_trim = False\nverbose = False\n\ntotal = 1000\n\nfor ii in range(0, total):\n \n # Load the data\n trajectory_tag = f\"trajectory{ii}_{nr_stays}stays\" \n path_to_file = pkls[ii]#data_dir + trajectory_tag\n t_arr, r_arr, x_arr, segments = get_pickle_trajectory(path_to_file)\n t_segs, x_segs = get_stay_segs(get_adjusted_stays(segments, t_arr))\n\n # Get the true event indices (needed for the total error)\n true_indices = get_stay_indices(get_adjusted_stays(segments, t_arr), t_arr)\n true_labels = np.zeros(t_arr.shape)\n for pair in true_indices:\n true_labels[pair[0]:pair[1]+1] = 1\n \n # Get the stay clusters\n #clusters = quick_box_method(t_arr, x_arr, dist_thresh, time_thresh, 1, False) \n all_clusters = stay_classifier_testing(t_arr, x_arr, dist_thresh, time_thresh, verbose) \n clusters = all_clusters[-1].copy()\n # Make some measurements\n final_len=len(clusters)\n # total scores\n prec, rec, conmat = eval_synth_data(t_arr, segments, clusters)\n # seg. scores\n _, a_prec, w_prec, _, a_rec, w_rec = get_segments_scores(t_arr, segments, clusters) \n \n # Total error\n pred_labels = get_pred_labels(clusters, t_arr.shape)\n err = get_err(true_labels,pred_labels)\n # Segment errors\n _, a_err, w_err = get_segments_errs(t_arr, segments, clusters)\n \n # Get the expected number of stays (in general)\n stays_tag = int((x_segs.size)/3)\n \n len_all_clusts = len(clusters)\n if final_len != stays_tag:\n lens.append(final_len)\n precs.append(prec)\n a_precs.append(a_prec)\n w_precs.append(w_prec)\n recs.append(rec) \n a_recs.append(a_rec) \n w_recs.append(w_rec)\n errs.append(err) \n a_errs.append(a_err) \n w_errs.append(w_err) \n else:\n lens3.append(final_len)\n precs3.append(prec)\n a_precs3.append(a_prec)\n w_precs3.append(w_prec)\n recs3.append(rec) \n a_recs3.append(a_rec) \n w_recs3.append(w_rec) \n errs3.append(err) \n a_errs3.append(a_err)\n w_errs3.append(w_err)\n \n # progress output\n if ii % int(0.1*total) == 0:\n print(f\"{ii:4d} of {total:5d}\")", " 0 of 1000\n 100 of 1000\n 200 of 1000\n 300 of 1000\n 400 of 1000\n 500 of 1000\n 600 of 1000\n 700 of 1000\n 800 of 1000\n 900 of 1000\n" ], [ "correct_frac = (len(lens3)/total)\nincorrect_frac = (len(lens)/total)", "_____no_output_____" ], [ "print(f\"\\n * correct number of stays, {correct_frac:6.3f} \", \n f\"\\n * prec.: {sum(w_precs3)/len(w_precs3):6.3}\",\n f\"\\n * rec.: {sum(w_recs3)/len(w_recs3):6.3}\",\n f\"\\n * incorrect number of stays, {incorrect_frac:6.3f}\",\n f\"\\n * prec.: {sum(w_precs)/len(w_precs):6.3}\",\n f\"\\n * rec.: {sum(w_recs)/len(w_recs):6.3}\")", "\n * correct number of stays, 0.796 \n * prec.: 0.956 \n * rec.: 0.994 \n * incorrect number of stays, 0.204 \n * prec.: 0.856 \n * rec.: 0.95\n" ] ], [ [ "## Visualizations", "_____no_output_____" ] ], [ [ "from stay_classification.metrics_plotting import plot_scores_stats, plot_errs_stats, plot_scores_stats_cominbed", "_____no_output_____" ], [ "os.mkdir('./visualizations/metrics_new/')", "_____no_output_____" ] ], [ [ "### Prec/rec score distributions\n\n#### Total scores", "_____no_output_____" ] ], [ [ "title = f\"Tot. scores: correct stays, {correct_frac:6.3f}; incorrect stays, {incorrect_frac:6.3f}\"\n\nfig, axs = plot_scores_stats(precs3, recs3, precs, recs, title)\n\nfig.savefig(\"./visualizations/metrics_new/\" + f\"metrics__{nr_stays}stays.png\")", "_____no_output_____" ], [ "title = f\"Total scores: correct stays, {correct_frac:6.3f}; incorrect stays, {incorrect_frac:6.3f}\"\n\nfig, axs = plot_scores_stats_cominbed(precs3, recs3, precs, recs, title)\n\nfig.savefig(\"./visualizations/metrics_new/\" + f\"metrics__{nr_stays}stays_seg_scores_combined_tot.png\")", "_____no_output_____" ] ], [ [ "#### Segment-averaged score distributions", "_____no_output_____" ] ], [ [ "title = f\"Avg. scores: correct stays, {correct_frac:6.3f}; incorrect stays, {incorrect_frac:6.3f}\"\n\nfig, axs = plot_scores_stats(a_precs3, a_recs3, a_precs, a_recs, title)\n\nfig.savefig(\"./visualizations/metrics_new/\" + f\"metrics__{nr_stays}stays_seg_scores_avg.png\")", "_____no_output_____" ], [ "title = f\"Avg. Scores: correct stays, {correct_frac:6.3f}; incorrect stays, {incorrect_frac:6.3f}\"\n\nfig, axs = plot_scores_stats_cominbed(a_precs3, a_recs3, a_precs, a_recs, title)\n\nfig.savefig(\"./visualizations/metrics_new/\" + f\"metrics__{nr_stays}stays_seg_scores_combined_avg.png\")", "_____no_output_____" ] ], [ [ "#### Segment, weighted-averaged score distributions", "_____no_output_____" ] ], [ [ "title = f\"W-avg. scores: correct stays, {correct_frac:6.3f}; incorrect stays, {incorrect_frac:6.3f}\"\n\nfig, axs = plot_scores_stats(w_precs3, w_recs3, w_precs, w_recs, title)\n\nfig.savefig(\"./visualizations/metrics_new/\" + f\"metrics__{nr_stays}stays_seg_scores_wavg.png\")", "_____no_output_____" ], [ "title = f\"W-avg. Scores: correct stays, {correct_frac:6.3f}; incorrect stays, {incorrect_frac:6.3f}\"\n\nfig, axs = plot_scores_stats_cominbed(w_precs3, w_recs3, w_precs, w_recs, title)\n\nfig.savefig(\"./visualizations/metrics_new/\" + f\"metrics__{nr_stays}stays_seg_scores_combined_wavg.png\")", "_____no_output_____" ] ], [ [ "### Error stats", "_____no_output_____" ] ], [ [ "title = f\"Avg. Error: correct stays, {correct_frac:6.3f}; incorrect stays, {incorrect_frac:6.3f}\"\n\nfig, ax = plot_errs_stats(w_errs3, w_errs, title)\n\nfig.savefig(\"./visualizations/metrics_new/\" + f\"metrics__{nr_stays}stays_seg_errs_avg.png\")", "_____no_output_____" ], [ "title = f\"W-avg. Error: correct stays, {correct_frac:6.3f}; incorrect stays, {incorrect_frac:6.3f}\"\n\nfig, ax = plot_errs_stats(w_errs3, w_errs, title)\n\nfig.savefig(\"./visualizations/metrics_new/\" + f\"metrics__{nr_stays}stays_seg_errs_wavg.png\")", "_____no_output_____" ] ], [ [ "---", "_____no_output_____" ] ] ]

[ "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown" ]

[ [ "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "<h3>Capturer un texte saisi par l'utilisateur</h3>\n<p>Nous nous proposons d'écrire un programme qui demande et stocke le nom de l'utilisateur, l'âge de l'utilisateur et stocke une valeur qui renseigne si l'utilisateur est majeur ou non. Par la suite nous afficherons un message de salutation à cet utilisateur, son âge ainsi que sa majorité.</p>\n<p>Voici comment se présente notre programme</p>", "_____no_output_____" ] ], [ [ "nom = input(\"Veuillez saisir votre nom: \")\nage = int(input(\"Veuillez saisir votre age: \"))\nestMajeur = age >= 18\n\nprint(f'Bonjour {nom}')\nprint(f'vous avez {age} ans')\nprint(f'Etes-vous majeur? {estMajeur}')", "Veuillez saisir votre nom: Hippo\nVeuillez saisir votre age: 15\n" ] ], [ [ "<p>Noter l'utilisation de la fonction <code>input()</code> avec le texte à afficher à l'utilisateur pour lui demander son nom. Cette fonction affichera le message placé entre parenthèses et guillements, ensuite stockera la valeur tapée par l'utilisateur dans la variable <code>nom</code> pour une utilisateur ultérieure dans le programme.</p>\n<p>Autre détail à noter, l'utilisation de la fonction <code>int()</code> qui enveloppe la fonction <code>input()</code> sur la deuxième ligne du programme; il s'agit d'une conversion de texte saisi par l'utilisateur en nombre, car l'age est un nombre et non un texte. Nous verrons ceci en détail au point suivant, sur la <em>conversion des types des variables</em>.</p>", "_____no_output_____" ], [ "NB: si l'utilisateur saisit un texte à la place de l'age, Python va vous générer une erreur comme quoi le texte saisi n'est pas un nombre; en programmation, on appelle cela une <strong><em>exception</em></strong>; nous apprendrons plus tard comment gérer les exceptions, afin de rester à l'abri des erreurs dans des cas pareils.", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown" ]

[ [ "markdown" ], [ "code" ], [ "markdown", "markdown" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Repaso de ciclos, cadenas, tuplas y listas\n", "_____no_output_____" ], [ "## Cobra Mosmas\n\nAl ver los precios y los anuncios del almacén Cobra Mosmas, un cliente le\npide crear un programa de computador que le permita ingresar el precio\nindividual de tres productos y el precio de la promoción en combo de los\ntres productos anunciada por el almacen y determine si es preferible\ncomprarlos por separado o en el combo promoción. (Pensemos por 3\nminutos en definir claramente el problema)", "_____no_output_____" ] ], [ [ "#piense aquí la solución", "_____no_output_____" ] ], [ [ "### Solución", "_____no_output_____" ] ], [ [ "def comprar(p1,p2,p3,pc):\n if pc <= p1+p2+p3:\n return 'Combo'\n else:\n return 'Por separado'\n\na=float(input('Precio primer producto?'))\nb=float(input('Precio segundo producto?'))\nc=float(input('Precio tercer producto?'))\nd=float(input('Precio combo?'))\nprint(\"Comprar\",comprar(a,b,c,d))", "Precio primer producto?1\nPrecio segundo producto?2\nPrecio tercer producto?3\nPrecio combo?4\nComprar Combo\n" ] ], [ [ "## La cerca\n\nUn campesino de la región le pide crear un programa de computador que\nle permita determinar cual de dos opciones (madera o alambre) es la mejor\nopción (menor costo) para encerrar un terreno rectangular de 𝑛 ∗ 𝑚 metros\ncuadrados, sabiendo el costo de un metro lineal de alambre, el costo de un\nmetro de madera y la cantdad de hilos de alambre o hileras de madera. El\ncampesino solo piensa en usar una de las dos opciones, no las piensa\ncombinar. (Pensemos por 3 minutos en definir claramente el problema)", "_____no_output_____" ] ], [ [ "def en_madera(n,m,w,p):\n return (2*n+2*m)*w*p\n\ndef en_alambre(n,m,h,a):\n return (2*n+2*m)*h*a\n\ndef usar(n,m,h,a,w,p):\n if en_madera(n,m,w,p) <= en_alambre(n,m,h,a):\n return 'Madera'\n else:\n return 'Alambre'\n \nn=float(input('Largo terreno?'))\nm=float(input('Ancho terreno?'))\na=float(input('Costo metro alambre?'))\nh=int(input('Hilos de alambre?'))\np=float(input('Costo metro madera?'))\nw=int(input('Hileras de madera?'))\nprint(\"Usar\",usar(n,m,a,h,p,w))", "Largo terreno?2\nAncho terreno?4\nCosto metro alambre?4\nHilos de alambre?3\nCosto metro madera?2\nHileras de madera?2\nUsar Madera\n" ] ], [ [ "## Lista Escolar\n\nUnos padres de familia desesperados por determinar el dinero que debenpedir prestado para pagar los útiles escolares de su hijo, le han pedido crearun programa de computador que a partir de una lista de los precios de cada útil escolar y de la cantidad de cada útil escolar en la lista, determineel precio total de la lista. (Pensemos por 5 minutos en la solución)", "_____no_output_____" ] ], [ [ "def costo(precio, cantidad):\n costo = 0\n for i in range(0,len(precio)):\n costo = costo + precio[i] * cantidad[i]\n return costo\n\nprecio = []\ncantidad = []\nwhile input('Ingresar otro útil?').upper()=='S':\n precio.append(float(input('Precio útil?')))\n cantidad.append(float(input('Cantidad?')))\nprint(\"La lista cuesta\", costo(precio, cantidad))", "Ingresar otro útil?S\nPrecio útil?3999\nCantidad?1\nIngresar otro útil?N\nLa lista cuesta 3999.0\n" ] ], [ [ "## ADN\n\nEn la última edición de la revista científica ”ADN al día” se indica que laspruebas de relación entre individuos a partir de código genético se definede la siguiente manera: Si las dos cadenas se diferencian en menos de𝑝letras, existe una relación de padre-hijo, si se diferencian en menos de𝑓 > 𝑝letras, existe una relación de formar parte de la misma familia. Deotra manera no existe relación. El laboratorioTein Cul Pan, le pidedesarrollar un programa que a partir de dos cadenas de ADN del mismo tamaño, determine si existe una relación pader-hijo, de la misma familia o ninguna, siguiendo las reglas definidas por la revista científica ”ADN aldía”. (Pensemos por 5 minutos en la solución)", "_____no_output_____" ] ], [ [ "def diferencia(a,b):\n cuenta = 0\n for i in range(0,len(a)):\n if a[i] != b[i]:\n cuenta = cuenta + 1\n return cuenta\n \n def relacion(a,b,p,f):\n d = diferencia(a,b)\n if d <= p:\n return 'Padre-Hijo'\n elif d <= f:\n return 'Familia'\n else:\n return 'Ninguna'\n \nind1=input('Cadena ADN individuo 1?')\nind2=input('Cadena ADN individuo 2?')\np=int(input('Diferencia máxima para ser Padre-Hijo?'))\nf=int(input('Diferencia máxima para ser Familia?'))\nprint(\"Relación\", relacion(ind1, ind2, f, p)", "_____no_output_____" ] ], [ [ "## Extraer nombres de universidades Colombianas\n\nDada una lista de Universidades Colombianas, obtener el nombre del sitio web. Se asume que un nombre de universidad está entre los caracteres www. y edu.co. Por ejemplo de www.unal.edu.co se obtiene unal.\n\n*Entrada:*\nUn numero n indicando la cantidad de nombres de sitios web a procesar\n\n*Salida:*\nListado de posibles nombres de universidades.\n\n*Ejemplo:*\n\n<table>\n <tr>\n <td>\n Input\n </td>\n <td>\n Output\n </td>\n </tr>\n <tr>\n <td>\n 5<br>\n www.unal.edu.co<br> \n www.udistrital.edu.co<br>\n www.univalle.edu.co<br>\n www.javeriana.edu.co<br>\n www.konradlorenz.edu.co<br>\n </td>\n <td> \n unal<br>\n udistrital<br>\n univalle<br>\n javeriana<br>\n konradlorenz<br>\n </td>\n </tr> \n</table>\n\n", "_____no_output_____" ] ], [ [ "#escriba aquí la solución", "_____no_output_____" ] ], [ [ "## Solucion:\n", "_____no_output_____" ] ], [ [ "def process(uni):\n return uni.split(\".\")[1]\n\ndef main():\n n = int(input())\n for i in range(n):\n uni = input()\n print(process(uni))\nmain()", "www.k.edu.co\n" ] ], [ [ "## Leer información de estudiantes y calcular el promedio de notas", "_____no_output_____" ], [ "Se tienen que procesar algunos comandos para realizar el procesamiento de notas de una Universidad. Se tiene una lista de estudiantes \n\n- Comando 1: Agregar estudiante y nota `1&nombre_estudiante&nota`\n- Comando 2: Calcular promedio de los estudiantes en un momento dado `2`\n- Comando 3: Ordenar estudiantes agregados por nombre `3`\n- Comando 4: Consultar la nota de un estudiante `4&nombre_estudiante`\n- Comando 5: Visualizar lista de estudiantes `5`\n- Comando 6: Salir `6`\n\n", "_____no_output_____" ] ], [ [ "#ingrese aquí la solución", "_____no_output_____" ] ], [ [ "Para poder resolver el problema se pueden identificar varias partes a resolver que pueden modelarse como funciones:\n\n- Definir la lista de estudiantes.\n- agregar un estudiante dada la información\n- calcular el promedio de notas de los estudiantes en un momento dado\n- ordenar estudiantes agregados por nombre\n- consultar la nota de un estudiante \n- visualizar lista\n- procesar los comandos\n- mostrar menu\n ", "_____no_output_____" ], [ "## Solución", "_____no_output_____" ] ], [ [ "# definir la lista de estudiantes un estudiante puede ser modelado como una tupla (por ahora)", "_____no_output_____" ], [ "\ndef agregar_estudiante(estudiantes, est): \n estudiantes.append(est)\n\ndef promedio(estudiantes):\n prom = 0\n #print(estudiantes)\n for estudiante in estudiantes:\n prom += float(estudiante[1])\n print(\"El promedio de los estudiantes es: \" + str(prom/len(estudiantes)))\n\ndef ordenar(estudiantes):\n estudiantes.sort()\n\ndef consultar(estudiantes, nombre):\n encontrado = False\n for estudiante in estudiantes:\n if estudiante[0] == nombre:\n encontrado = True\n print(estudiante[1])\n if not encontrado:\n print(\"Estudiante no encontrado\")\n \ndef visualizar(estudiantes):\n print(\"Lista de estudiantes\".center(30, \"#\"))\n if len(estudiantes) == 0:\n print(\"No hay estudiantes registrados.\")\n for e in estudiantes:\n print(\"Nombre: \" + e[0] + \", nota:\" + str(e[1]))\n \ndef procesar_comandos():\n bandera = True\n estudiantes = []\n comando = [0]\n while bandera or comando[0] != \"6\":\n bandera = False\n mostrar_menu()\n comando = input().split(\"&\")\n print(comando[0])\n if comando[0] == \"1\":\n agregar_estudiante(estudiantes, (comando[1], float(comando[2])))\n elif comando[0] == \"2\":\n promedio(estudiantes)\n elif comando[0] == \"3\":\n ordenar(estudiantes)\n elif comando[0] == \"4\":\n consultar(estudiantes, comando[1]) \n elif comando[0] == \"5\":\n visualizar(estudiantes)\n \ndef mostrar_menu():\n print(\"Seleccione una opción:\")\n print(\"Comando 1: Agregar estudiante y nota `1&nombre_estudiante&nota`\")\n print(\"Comando 2: Calcular promedio de los estudiantes en un momento dado.\")\n print(\"Comando 3: Ordenar estudiantes agregados por nombre\")\n print(\"Comando 4: Consultar la nota de un estudiante `4&nombre_estudiante`\")\n print(\"Comando 5: Visualizar\")\n print(\"Comando 6: Salir\")\n\n\nprocesar_comandos()\n\n\"\"\"\n1&Antonia&5.0\n1&Juan&2.4\n1&Pedro&4.3\n\"\"\"", "_____no_output_____" ] ], [ [ "## Entendiendo los sentimientos de Groot\n\nEl lenguaje de Groot es muy complicado para expresar sentimientos. Los sentimientos tienen n capas.\n\nSi n = 1, el sentimiento será \"I hate it\", si n = 2 es \"I hate that I love it\", y si n = 3 es \"I hate that I love that I hate it\" y así sucesivamente.\n\n*Entrada:*\nLa cantidad n de capas donde $n \\geq 1$\n\n*Salida:*\nMuestre la frase que Groot está tratando de decir.\n\n\n*Ejemplo 1:*\n\n<table>\n <tr>\n <td>Input</td><td>Output</td>\n </tr>\n <tr>\n <td>1</td><td>I hate it</td>\n </tr>\n</table>\n\n\nEjemplo 2:\n\n<table>\n <tr>\n <td>Input</td><td>Output</td>\n </tr>\n <tr>\n <td>2</td><td>I hate that I love it</td>\n </tr>\n</table>\n\nEjemplo 3:\n\n<table>\n <tr>\n <td>Input</td><td>Output</td>\n </tr>\n <tr>\n <td>3</td><td>I hate that I love that I hate it</td>\n </tr>\n</table>", "_____no_output_____" ] ], [ [ "#piense aquí la solución en caso de no plantear ninguna abra la siguiente celda", "_____no_output_____" ] ], [ [ "### Una posible opción para groot\n\nUna posible opción puede ser construir una tupla ó lista con dos elementos: \n", "_____no_output_____" ] ], [ [ "emocion = [\"I hate\", \"I love\"]\n", "_____no_output_____" ] ], [ [ "Puede ir alternando entre posiciones de la lista el ciclo así:", "_____no_output_____" ] ], [ [ "salida = []\n\nn = int(input())\n\nfor i in range(n):\n salida.append(emocion[i%2])\n\nprint(salida)", "_____no_output_____" ] ], [ [ "Se puede usar join con that...", "_____no_output_____" ] ], [ [ "res = \" that \".join(salida) ", "_____no_output_____" ] ], [ [ "Agregue it:", "_____no_output_____" ] ], [ [ "res += \" it \"\nprint(res)", "_____no_output_____" ] ], [ [ "Es posible generar esta solución con una lista creada por comprensión?", "_____no_output_____" ] ], [ [ "#intentelo aquí", "_____no_output_____" ] ], [ [ "### Solución", "_____no_output_____" ] ], [ [ "emocion = [\"I hate\", \"I love\"]\nsalida = \" that \".join([emocion[i % 2] for i in range(n)])+\" it\"\nprint(salida)", "_____no_output_____" ] ], [ [ "## Simplificador de Fracciones\n\nUtilizando funciones recursivas, elabore un programa que simplifique una fracción escrita de la forma a/b.\n\nEjemplo: \n<table>\n <tr><td>Entrada</td><td>Salida</td></tr>\n <tr><td>6/4</td><td>3/2</td></tr>\n</table>\n", "_____no_output_____" ] ], [ [ "", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "# King-County-House-Price-Prediction\n", "_____no_output_____" ] ], [ [ "# Importing libraries\nimport numpy as np \nimport pandas as pd \nimport matplotlib.pyplot as plt \nimport seaborn as sns\nimport sklearn\nfrom sklearn.metrics import mean_absolute_error, mean_squared_error,r2_score \nfrom sklearn.preprocessing import MinMaxScaler, StandardScaler \n\n# Ignore warnings\nimport warnings\nwarnings.filterwarnings('ignore')\n\n# we can set numbers for how many rows and columns will be displayed\npd.set_option('display.min_rows', 10)\npd.set_option('display.max_columns', 30)", "_____no_output_____" ] ], [ [ "## 1. Loading Dataset and Explore", "_____no_output_____" ] ], [ [ "df = pd.read_csv('/content/dataset/kc_house_data.csv')", "_____no_output_____" ], [ "df.head()", "_____no_output_____" ], [ "df.shape", "_____no_output_____" ], [ "df.info()", "<class 'pandas.core.frame.DataFrame'>\nRangeIndex: 21613 entries, 0 to 21612\nData columns (total 21 columns):\n # Column Non-Null Count Dtype \n--- ------ -------------- ----- \n 0 id 21613 non-null int64 \n 1 date 21613 non-null object \n 2 price 21613 non-null float64\n 3 bedrooms 21613 non-null int64 \n 4 bathrooms 21613 non-null float64\n 5 sqft_living 21613 non-null int64 \n 6 sqft_lot 21613 non-null int64 \n 7 floors 21613 non-null float64\n 8 waterfront 21613 non-null int64 \n 9 view 21613 non-null int64 \n 10 condition 21613 non-null int64 \n 11 grade 21613 non-null int64 \n 12 sqft_above 21613 non-null int64 \n 13 sqft_basement 21613 non-null int64 \n 14 yr_built 21613 non-null int64 \n 15 yr_renovated 21613 non-null int64 \n 16 zipcode 21613 non-null int64 \n 17 lat 21613 non-null float64\n 18 long 21613 non-null float64\n 19 sqft_living15 21613 non-null int64 \n 20 sqft_lot15 21613 non-null int64 \ndtypes: float64(5), int64(15), object(1)\nmemory usage: 3.5+ MB\n" ], [ "# Delete Unwanted columns\ndf.drop(['id', 'date'], axis=1, inplace=True)", "_____no_output_____" ], [ "# Display missing values information\ndf.isna().sum().sort_values(ascending=False)", "_____no_output_____" ], [ "df.describe().T ", "_____no_output_____" ] ], [ [ "## 2. EDA", "_____no_output_____" ] ], [ [ "# classification cat_col and num_col for plotting\ncat_col = []\nnum_col = []\ncol_name = df.columns\nfor idx, value in enumerate(col_name):\n con = df[value].nunique()\n if(con<20):\n cat_col.append(value)\n else:\n num_col.append(value)\n", "_____no_output_____" ], [ "def bar_plot(data, categorical_features):\n \"\"\"Bar plot for categorical fetures of all columns\"\"\"\n\n print(\"Bar Plot for Categorical features\")\n for col in categorical_features:\n counts = data[col].value_counts().sort_index()\n fig = plt.figure(figsize=(9, 6))\n ax = fig.gca()\n counts.plot.bar(ax = ax, color='steelblue')\n ax.set_title(col + ' counts')\n ax.set_xlabel(col) \n ax.set_ylabel(\"Frequency\")\n\n return plt.show()\n\n\ndef histogram_plot(data, numeric_columns):\n \"\"\"Histogram for numerical fetures of all columns\"\"\"\n\n print(\"Histogram for numeric_columns\")\n for col in numeric_columns:\n fig = plt.figure(figsize=(9, 6))\n ax = fig.gca()\n feature = data[col]\n feature.hist(bins=50, ax = ax)\n ax.axvline(feature.mean(), color='magenta', linestyle='dashed', linewidth=2)\n ax.axvline(feature.median(), color='cyan', linestyle='dashed', linewidth=2)\n ax.set_title(col)\n return plt.show()\n\n\ndef scater_plot(data, numeric_columns, target_col):\n \"\"\"Scatter for numerical fetures of columns for target columns\"\"\"\n\n print(\"Scater plot\")\n for col in numeric_columns:\n\n fig = plt.figure(figsize=(9, 6))\n ax = fig.gca()\n feature = df[col]\n label = data[f'{target_col}']\n correlation = feature.corr(label)\n plt.scatter(x=feature, y=label)\n plt.xlabel(col)\n plt.ylabel('Price')\n ax.set_title('Price vs ' + col + '- correlation: ' + str(correlation))\n\n return plt.show()\n", "_____no_output_____" ], [ "# Bar plot for categorical features\nbar_plot(df, cat_col)", "Bar Plot for Categorical features\n" ], [ "# Histogram for numerical columns\nhistogram_plot(df, num_col)", "Histogram for numeric_columns\n" ], [ "# Scatter plot for price vs numerical columns\nscater_plot(df, num_col, 'price')", "Scater plot\n" ], [ "# Correlation plot\ncorr = df.corr().round(2)\nplt.figure(figsize=(12,8))\nsns.heatmap(corr, annot=True, cmap=\"YlGnBu\");", "_____no_output_____" ] ], [ [ "## 3. Split dataset", "_____no_output_____" ] ], [ [ "# create function for Split data\n\ndef split_data(data, target_col):\n \n # Remove rows with missing target, seprate target from predictors\n data_copy = data.copy()\n data_copy.dropna(axis=0, subset=[target_col], inplace=True)\n y = data_copy[target_col]\n data_copy.drop([target_col], axis=1, inplace=True)\n\n # Break off validation set from training data\n from sklearn.model_selection import train_test_split\n X_train, X_valid, y_train, y_valid = train_test_split(data_copy, y, train_size=0.8, test_size=0.2, random_state=4)\n\n return X_train, X_valid, y_train, y_valid", "_____no_output_____" ], [ "# Split data from main dataset to train, test and target\nX_train, X_valid, y_train, y_valid = split_data(df, 'price')", "_____no_output_____" ] ], [ [ "## 4. Model train", "_____no_output_____" ], [ "Create column classification function for pipeline", "_____no_output_____" ] ], [ [ "def col_classification(data, num=20):\n # Select categorical columns\n cat_cols = [cname for cname in data.columns if\n data[cname].nunique() < num and\n data[cname].dtype =='object']\n\n # Select numerical columns\n num_cols = [cname for cname in data.columns if\n data[cname].dtype in ['int64', 'float64']]\n\n return cat_cols, num_cols\n", "_____no_output_____" ], [ "# Categorical cols and numerical columns classfication\ncategorical_cols, numerical_cols = col_classification(X_train, 15)", "_____no_output_____" ] ], [ [ "Create model evaluation function", "_____no_output_____" ] ], [ [ "def evaluation_model(X_test, y_test, title=\"Target price prediction\"):\n \"\"\"Evaluation Model for regression problem, We need to use model name is clf\"\"\"\n # Evaluate the model using the test data\n preds = reg.predict(X_test)\n\n mse = mean_squared_error(y_test, preds)\n print(\"MSE:\", mse)\n rmse = np.sqrt(mse)\n print('Mae: ', mean_absolute_error(y_valid, preds))\n print(\"RMSE:\", rmse)\n r2 = r2_score(y_valid, preds)\n print(\"R2:\", r2)\n\n # Plot predicted vs actual\n plt.scatter(y_test, preds)\n plt.xlabel('Actual Labels')\n plt.ylabel('Predicted Labels')\n plt.title(title)\n\n # overlay the regression line\n z = np.polyfit(y_test, preds, 1)\n p = np.poly1d(z)\n plt.plot(y_valid,p(y_valid), color='magenta')\n return plt.show() ", "_____no_output_____" ] ], [ [ "Try some of the linear regression model and evaluation their performance in our dataset", "_____no_output_____" ] ], [ [ "from sklearn.linear_model import LinearRegression\nfrom sklearn.linear_model import Ridge\nfrom sklearn.linear_model import Lasso\nfrom sklearn.linear_model import ElasticNet\nfrom sklearn.ensemble import GradientBoostingRegressor\nfrom sklearn.tree import DecisionTreeRegressor\nfrom sklearn.ensemble import RandomForestRegressor\n\n\nmodels={'Linear Regression': LinearRegression(),\n 'Decision Tree Regressior' : DecisionTreeRegressor(random_state=0),\n 'Random Forrest Regressor' : RandomForestRegressor(n_estimators=10, random_state=0),\n 'Ridge': Ridge(),\n 'Lasso': Lasso(),\n 'ElasticN': ElasticNet(),\n 'Gradient Boosting Regressor': GradientBoostingRegressor()\n }\n\nprint('####################################################################### \\n')\nfor name, model in models.items():\n name_model = model\n reg = name_model.fit(X_train, y_train)\n print(f'{name}:')\n evaluation_model(X_valid, y_valid, 'Housing Price Prediction')\n print('####################################################################### \\n')", "####################################################################### \n\nLinear Regression:\nMSE: 39200213360.80082\nMae: 126368.12273608406\nRMSE: 197990.43754888978\nR2: 0.6969193739897344\n" ] ], [ [ "After the evaluation Gradient Boosting Regressor works well in our dataset so we will use it.", "_____no_output_____" ], [ "**Model train with Gradient Boosting Regressor using sklearn pipeline**", "_____no_output_____" ] ], [ [ "from sklearn.compose import ColumnTransformer\nfrom sklearn.pipeline import Pipeline\nfrom sklearn.impute import SimpleImputer\nfrom sklearn.preprocessing import OneHotEncoder\nfrom sklearn.metrics import mean_absolute_error\nfrom sklearn.preprocessing import StandardScaler\nfrom sklearn.ensemble import GradientBoostingRegressor\n\n\n# Preprocessing for numerical data\nnumerical_transformer = SimpleImputer(strategy='median')\n\n# Preprocessing for categorical data\ncategorical_transformer = Pipeline(steps=[\n ('imputer', SimpleImputer(strategy='most_frequent')),\n ('onehot', OneHotEncoder(handle_unknown='ignore'))\n])\n\n\n# Bundle preprocessing for numerical and categorical data\npreprocessor = ColumnTransformer(\n transformers=[\n ('num', numerical_transformer, numerical_cols),\n ('cat', categorical_transformer, categorical_cols)\n ])\n\n\nmodel = GradientBoostingRegressor(random_state=4, n_estimators=1600)\n\n# Bundle Preprocessing and modeling code in pipeline\nreg = Pipeline(steps=[\n ('preprocessor', preprocessor),\n ('scaler', StandardScaler()),\n ('model', model),\n ])\n\n# Preprocessing of training data, fit model\nreg.fit(X_train, y_train)", "_____no_output_____" ], [ "evaluation_model(X_valid, y_valid, 'Housing Price Prediction')", "MSE: 13811738810.108906\nMae: 66634.5069623637\nRMSE: 117523.35431780743\nR2: 0.8932130698797662\n" ] ], [ [ "## 5. Save Model, Load and Prediction", "_____no_output_____" ] ], [ [ "import pickle\n!mkdir 'model_save'\npickle.dump(reg, open(\"model_save/model.pkl\",\"wb\"))", "mkdir: cannot create directory ‘model_save’: File exists\n" ], [ "# Load the model from the file\nloaded_model = pickle.load(open('model_save/model.pkl', 'rb'))", "_____no_output_____" ], [ "X_new = pd.read_csv('dataset/user_input.csv')\nX_new", "_____no_output_____" ], [ "result = loaded_model.predict(X_new)\nprint('Prediction: {:.0f} price'.format(np.round(result[0])))", "Prediction: 562858 price\n" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown", "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Table of Contents\n <p><div class=\"lev1 toc-item\"><a href=\"#Testing-lolviz\" data-toc-modified-id=\"Testing-lolviz-1\"><span class=\"toc-item-num\">1&nbsp;&nbsp;</span>Testing <a href=\"https://github.com/parrt/lolviz\" target=\"_blank\">lolviz</a></a></div><div class=\"lev2 toc-item\"><a href=\"#Testing-naively\" data-toc-modified-id=\"Testing-naively-11\"><span class=\"toc-item-num\">1.1&nbsp;&nbsp;</span>Testing naively</a></div><div class=\"lev2 toc-item\"><a href=\"#Testing-from-within-a-Jupyter-notebook\" data-toc-modified-id=\"Testing-from-within-a-Jupyter-notebook-12\"><span class=\"toc-item-num\">1.2&nbsp;&nbsp;</span>Testing from within a Jupyter notebook</a></div><div class=\"lev3 toc-item\"><a href=\"#List\" data-toc-modified-id=\"List-121\"><span class=\"toc-item-num\">1.2.1&nbsp;&nbsp;</span>List</a></div><div class=\"lev3 toc-item\"><a href=\"#List-of-lists\" data-toc-modified-id=\"List-of-lists-122\"><span class=\"toc-item-num\">1.2.2&nbsp;&nbsp;</span>List of lists</a></div><div class=\"lev3 toc-item\"><a href=\"#List-of-lists-of-lists???\" data-toc-modified-id=\"List-of-lists-of-lists???-123\"><span class=\"toc-item-num\">1.2.3&nbsp;&nbsp;</span>List of lists of lists???</a></div><div class=\"lev3 toc-item\"><a href=\"#Tree\" data-toc-modified-id=\"Tree-124\"><span class=\"toc-item-num\">1.2.4&nbsp;&nbsp;</span>Tree</a></div><div class=\"lev3 toc-item\"><a href=\"#Objects\" data-toc-modified-id=\"Objects-125\"><span class=\"toc-item-num\">1.2.5&nbsp;&nbsp;</span>Objects</a></div><div class=\"lev3 toc-item\"><a href=\"#Calls\" data-toc-modified-id=\"Calls-126\"><span class=\"toc-item-num\">1.2.6&nbsp;&nbsp;</span>Calls</a></div><div class=\"lev3 toc-item\"><a href=\"#String\" data-toc-modified-id=\"String-127\"><span class=\"toc-item-num\">1.2.7&nbsp;&nbsp;</span>String</a></div><div class=\"lev2 toc-item\"><a href=\"#Conclusion\" data-toc-modified-id=\"Conclusion-13\"><span class=\"toc-item-num\">1.3&nbsp;&nbsp;</span>Conclusion</a></div>", "_____no_output_____" ], [ "# Testing [lolviz](https://github.com/parrt/lolviz)\n\nI liked how the [lolviz](https://github.com/parrt/lolviz) module looked like. Let's try it!", "_____no_output_____" ] ], [ [ "%load_ext watermark\n%watermark -v -m -p lolviz", "CPython 3.6.5\nIPython 6.4.0\n\nlolviz n\u0007\n\ncompiler : GCC 7.3.0\nsystem : Linux\nrelease : 4.15.0-23-generic\nmachine : x86_64\nprocessor : x86_64\nCPU cores : 4\ninterpreter: 64bit\n" ], [ "from lolviz import *", "_____no_output_____" ] ], [ [ "## Testing naively", "_____no_output_____" ] ], [ [ "data = ['hi', 'mom', {3, 4}, {\"parrt\": \"user\"}]\ng = listviz(data)\nprint(g.source) # if you want to see the graphviz source\ng.view() # render and show graphviz.files.Source object", "\n digraph G {\n nodesep=.05;\n node [penwidth=\"0.5\", width=.1,height=.1];\n node139621679300936 [shape=\"box\", space=\"0.0\", margin=\"0.01\", fontcolor=\"#444443\", fontname=\"Helvetica\", label=<<table BORDER=\"0\" CELLBORDER=\"0\" CELLSPACING=\"0\">\n<tr>\n<td cellspacing=\"0\" cellpadding=\"0\" bgcolor=\"#fefecd\" border=\"1\" sides=\"br\" valign=\"top\"><font color=\"#444443\" point-size=\"9\">0</font></td>\n<td cellspacing=\"0\" cellpadding=\"0\" bgcolor=\"#fefecd\" border=\"1\" sides=\"br\" valign=\"top\"><font color=\"#444443\" point-size=\"9\">1</font></td>\n<td cellspacing=\"0\" cellpadding=\"0\" bgcolor=\"#fefecd\" border=\"1\" sides=\"br\" valign=\"top\"><font color=\"#444443\" point-size=\"9\">2</font></td>\n<td cellspacing=\"0\" cellpadding=\"0\" bgcolor=\"#fefecd\" border=\"1\" sides=\"b\" valign=\"top\"><font color=\"#444443\" point-size=\"9\">3</font></td>\n</tr>\n<tr>\n<td port=\"0\" bgcolor=\"#fefecd\" border=\"1\" sides=\"r\" align=\"center\"><font point-size=\"11\">'hi'</font></td>\n<td port=\"1\" bgcolor=\"#fefecd\" border=\"1\" sides=\"r\" align=\"center\"><font point-size=\"11\">'mom'</font></td>\n<td port=\"2\" bgcolor=\"#fefecd\" border=\"1\" sides=\"r\" align=\"center\"><font point-size=\"11\">{3, 4}</font></td>\n<td port=\"3\" bgcolor=\"#fefecd\" border=\"0\" align=\"center\"><font point-size=\"11\">{'parrt': 'user'}</font></td>\n</tr></table>\n>];\n}\n\n" ] ], [ [ "It opened a window showing me this image:\n\n<img src=\"data/Testing_the_lolviz_Python_module_1.png\" width=55%>", "_____no_output_____" ], [ "## Testing from within a Jupyter notebook\n\nI test here all [the features of lolviz](https://github.com/parrt/lolviz#functionality) :", "_____no_output_____" ], [ "### List", "_____no_output_____" ] ], [ [ "squares = [ i**2 for i in range(10) ]", "_____no_output_____" ], [ "squares", "_____no_output_____" ], [ "listviz(squares)", "_____no_output_____" ] ], [ [ "### List of lists", "_____no_output_____" ] ], [ [ "n, m = 3, 4\nexample_matrix = [[0 if i != j else 1 for i in range(n)] for j in range(m)]", "_____no_output_____" ], [ "example_matrix", "_____no_output_____" ], [ "lolviz(example_matrix)", "_____no_output_____" ] ], [ [ "### List of lists of lists???", "_____no_output_____" ] ], [ [ "n, m, o = 2, 3, 4\nexample_3D_matrix = [[[\n 1 if i < j < k else 0\n for i in range(n)]\n for j in range(m)]\n for k in range(o)]", "_____no_output_____" ], [ "example_3D_matrix", "_____no_output_____" ], [ "lolviz(example_3D_matrix)", "_____no_output_____" ] ], [ [ "It works, even if it is not as pretty.", "_____no_output_____" ], [ "### Tree\nOnly for binary trees, apparently. Let's try with a dictionary that looks like a binary tree:", "_____no_output_____" ] ], [ [ "anakin = {\n \"name\": \"Anakin Skywalker\",\n \"son\": {\n \"name\": \"Luke Skywalker\",\n },\n \"daughter\": {\n \"name\": \"Leia Skywalker\",\n },\n}", "_____no_output_____" ], [ "from pprint import pprint\npprint(anakin)", "{'daughter': {'name': 'Leia Skywalker'},\n 'name': 'Anakin Skywalker',\n 'son': {'name': 'Luke Skywalker'}}\n" ], [ "treeviz(anakin, leftfield='son', rightfield='daugther')", "_____no_output_____" ] ], [ [ "It doesn't work out of the box for dictionaries, sadly.\n\nLet's check another example:", "_____no_output_____" ] ], [ [ "class Tree:\n def __init__(self, value, left=None, right=None):\n self.value = value\n self.left = left\n self.right = right\n \nroot = Tree('parrt',\n Tree('mary',\n Tree('jim',\n Tree('srinivasan'),\n Tree('april'))),\n Tree('xue',None,Tree('mike')))\n\ntreeviz(root)", "_____no_output_____" ] ], [ [ "### Objects", "_____no_output_____" ] ], [ [ "objviz(anakin)", "_____no_output_____" ], [ "objviz(anakin.values())", "_____no_output_____" ], [ "objviz(anakin.items())", "_____no_output_____" ] ], [ [ "For complex numbers for instance?", "_____no_output_____" ] ], [ [ "z = 1+4j", "_____no_output_____" ], [ "print(z)", "(1+4j)\n" ], [ "objviz(z)", "_____no_output_____" ] ], [ [ "OK, this fails.", "_____no_output_____" ], [ "### Calls", "_____no_output_____" ] ], [ [ "def factorial(n):\n if n < 0: return 0\n elif n == 0: return 1\n else: return n * factorial(n - 1)", "_____no_output_____" ], [ "for n in range(12):\n print(f\"{n}! = {factorial(n)}\")", "0! = 1\n1! = 1\n2! = 2\n3! = 6\n4! = 24\n5! = 120\n6! = 720\n7! = 5040\n8! = 40320\n9! = 362880\n10! = 3628800\n11! = 39916800\n" ] ], [ [ "And now with some visualization:", "_____no_output_____" ] ], [ [ "from IPython.display import display", "_____no_output_____" ], [ "def factorial2(n):\n display(callsviz(varnames=[\"n\"]))\n if n < 0: return 0\n elif n == 0: return 1\n else: return n * factorial2(n - 1)", "_____no_output_____" ], [ "n = 4\nprint(f\"{n}! = {factorial2(n)}\")", "_____no_output_____" ] ], [ [ "We really see the \"call stack\" as the system keeps track of the nested calls. I like that! 👌", "_____no_output_____" ], [ "### String", "_____no_output_____" ] ], [ [ "import string\nstring.hexdigits", "_____no_output_____" ], [ "strviz(string.hexdigits)", "_____no_output_____" ] ], [ [ "## Conclusion\nThat's it. See [this other example](https://github.com/parrt/lolviz/blob/master/examples.ipynb) for more.", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown" ]

[ [ "markdown", "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown", "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown", "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown", "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown", "markdown" ], [ "code", "code" ], [ "markdown" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Masks\n\nWith gdsfactory you can easily go from components, to sweeps, to Masks. \n\nLets start with a resistance sweep, where you change the resistance width to measure sheet resistance\n\n## Pack", "_____no_output_____" ] ], [ [ "import gdsfactory as gf\ngf.clear_cache()\n\nsweep = [gf.components.resistance_sheet(width=width) for width in [1, 10, 100]]\nm = gf.pack(sweep)\nm[0]", "_____no_output_____" ], [ "spiral_te = gf.routing.add_fiber_single(gf.functions.rotate(gf.components.spiral_inner_io_fiber_single, 90))\nspiral_te", "_____no_output_____" ], [ "# which is equivalent to\nspiral_te = gf.compose(gf.routing.add_fiber_single, gf.functions.rotate90, gf.components.spiral_inner_io_fiber_single)\nspiral_te(length=10e3)", "_____no_output_____" ], [ "import gdsfactory as gf\n\nspiral_te = gf.compose(gf.routing.add_fiber_single, gf.functions.rotate90, gf.components.spiral_inner_io_fiber_single)\nsweep = [spiral_te(length=length) for length in [10e3, 20e3, 30e3]]\nm = gf.pack(sweep)\nm[0]", "_____no_output_____" ] ], [ [ "You can also add a `prefix` to each text label. For example `S` for the spirals at the `north-center`\n\n`text_rectangular` is DRC clean and is anchored on `nc` (north-center)", "_____no_output_____" ] ], [ [ "text_metal3 = gf.partial(gf.components.text_rectangular_multi_layer, layers=(gf.LAYER.M3,))\n\nm = gf.pack(sweep, text=text_metal3, text_anchors=('nc',), text_prefix='s')\nm[0]", "_____no_output_____" ], [ "text_metal2 = gf.partial(gf.c.text, layer=gf.LAYER.M2)\n\nm = gf.pack(sweep, text=text_metal2, text_anchors=('nc',), text_prefix='s')\nm[0]", "_____no_output_____" ] ], [ [ "## Grid", "_____no_output_____" ] ], [ [ "g = gf.grid(sweep)\ng", "_____no_output_____" ], [ "gh = gf.grid(sweep, shape=(1, len(sweep)))\ngh", "_____no_output_____" ], [ "ghymin = gf.grid(sweep, shape=(1, len(sweep)), align_y='ymin')\nghymin", "_____no_output_____" ] ], [ [ "You can also add text labels to each element of the sweep", "_____no_output_____" ] ], [ [ "ghymin = gf.grid_with_text(sweep, shape=(1, len(sweep)), align_y='ymin', text=text_metal3)\nghymin", "_____no_output_____" ] ], [ [ "## Mask\n\nYou can easily define a mask using `grid` and `pack`", "_____no_output_____" ] ], [ [ "import gdsfactory as gf\n\ntext_metal3 = gf.partial(gf.c.text_rectangular_multi_layer, layers=(gf.LAYER.M3,))\ngrid = gf.partial(gf.grid_with_text, text=text_metal3)\npack = gf.partial(gf.pack, text=text_metal3)\n\ngratings_sweep = [gf.c.grating_coupler_elliptical(taper_angle=taper_angle) for taper_angle in [20, 30, 40]]\ngratings = grid(gratings_sweep, text=None)\ngratings", "_____no_output_____" ], [ "gratings_sweep = [gf.c.grating_coupler_elliptical(taper_angle=taper_angle) for taper_angle in [20, 30, 40]]\ngratings_loss_sweep = [gf.c.grating_coupler_loss_fiber_single(grating_coupler=grating) for grating in gratings_sweep]\ngratings = grid(gratings_loss_sweep, shape=(1, len(gratings_loss_sweep)), spacing = (40,0))\ngratings", "_____no_output_____" ], [ "sweep_resistance = [gf.components.resistance_sheet(width=width) for width in [1, 10, 100]]\nresistance = gf.pack(sweep_resistance)[0]\nresistance", "_____no_output_____" ], [ "spiral_te = gf.compose(gf.routing.add_fiber_single, gf.functions.rotate90, gf.components.spiral_inner_io_fiber_single)\nsweep_spirals = [spiral_te(length=length) for length in [10e3, 20e3, 30e3]]\nspirals = gf.pack(sweep_spirals)[0]\nspirals", "_____no_output_____" ], [ "mask = gf.pack([spirals, resistance, gratings])[0]\nmask", "_____no_output_____" ] ], [ [ "As you can see you can define your mask in a single line.\n\nFor more complex mask, you can also create a new cell to build up more complexity", "_____no_output_____" ] ], [ [ "@gf.cell\ndef mask():\n c = gf.Component()\n c << gf.pack([spirals, resistance, gratings])[0]\n c << gf.c.seal_ring(c)\n return c\n\nc = mask(cache=False)\nc", "_____no_output_____" ], [ "c.write_gds_with_metadata(gdsdir='extra')", "_____no_output_____" ], [ "gf.mask.write_labels(gdspath='extra/mask_d41d8cd9.gds', label_layer=(201, 0))", "_____no_output_____" ] ], [ [ "```\n\nCSV labels ------|\n |--> merge_test_metadata dict\n |\nYAML metatada ---\n\n```", "_____no_output_____" ] ], [ [ "test_metadata = gf.mask.merge_test_metadata(gdspath='extra/mask_d41d8cd9.gds')", "_____no_output_____" ], [ "test_metadata.spiral_inner_io_6dc6250a.full.length", "_____no_output_____" ], [ "spiral_names = [s for s in test_metadata.keys() if s.startswith('spiral')]\nspiral_names", "_____no_output_____" ], [ "spiral_lengths = [test_metadata[spiral_name].length for spiral_name in spiral_names]\nspiral_lengths", "_____no_output_____" ], [ "gc_names = [s for s in test_metadata.keys() if s.startswith('grating')]\ngc_names", "_____no_output_____" ], [ "gc_taper_angles = [test_metadata[name].full.taper_angle for name in gc_names]\ngc_taper_angles", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown" ], [ "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Introduction to Convolution Neural Nets\n========\n\n#### Version 0.1\n\nBy B Nord 2018 Nov 09\n\nThis notebook was developed within the [Google Collaboratory](https://colab.research.google.com/notebooks/welcome.ipynb#recent=true) framework. The original notebook can be run in a web browser, and is available [via Collaboratory](https://colab.research.google.com/drive/1wKzhJ0cOsJbgM9L0uIVUCYW1f2Zdf3PK#scrollTo=qwubzWGWWD6E). It has been recreated below, though we recommend you run the web-based version.", "_____no_output_____" ], [ "# Install packages on the back end", "_____no_output_____" ] ], [ [ "# install software on the backend, which is located at \n# Google's Super Secret Sky Server in an alternate universe.\n# The backend is called a 'hosted runtime' if it is on their server.\n# A local runtime would start a colab notebook on your machine locally. \n# Think of google colab as a Google Docs version of Jupyter Notebooks\n\n# remove display of install details\n%%capture --no-display \n\n# pip install\n!pip install numpy matplotlib scipy pandas scikit-learn astropy seaborn ipython jupyter #standard install for DSFP\n!pip install keras tensorflow # required for deep learning \n!pip install pycm", "_____no_output_____" ], [ "# standard-ish imports\nimport numpy as np\nimport matplotlib.pyplot as plt\nimport time\nimport itertools\n\n# non-standard, but stable package for confusion matrices\nfrom pycm import ConfusionMatrix\n\n\n# neural network / machine learning packages\nfrom sklearn import metrics\nimport keras\nfrom keras.datasets import mnist\nfrom keras.models import Sequential\nfrom keras.layers import Dense, Dropout, Flatten, Activation\nfrom keras.layers import Conv2D, MaxPooling2D, BatchNormalization\nfrom keras import backend as K", "Using TensorFlow backend.\n" ] ], [ [ "# Convolutional Neural Networks make the future now!\n\n**Learning Objectives**\n\n\n1. Gain familiarity with \n 1. Two standard convolutional neural network (CNN) architectures: \n 1. **Feed-forward CNN**\n 2. **Convolutional Autoencoder (CAE)**\n 2. One standard task performed with CNNs: **Binary Classification**\n 3. One new diagnostic of CNNs: **Feature maps from the first layer**\n2. Experience fundamental considerations, pitfalls, and strategies when training NNs\n 1. Data set preparation (never underestimate the time required for this)\n 2. CNN layer manipulation and architecture design\n 5. Model fitting (the learning process)\n 6. Effects of image quality\n3. Apply diagnostics from previous exercises\n4. Apply new diagnostics: look inside the networks with feature maps of the first layer\n5. Continue connecting NN functionality to data set structure and problem of interest\n\n\nSome of this notebook is very similar to the first one, but we're using a new architecture that has more moving pieces.\n\n\n\n*I'm still taking bets that we can start a paper with deep nets during the Saturday hack.*\n", "_____no_output_____" ], [ "## Activity 1: Classify Handwritten Digits with Convolutional Neural Networks (CNNs)\nIs it a \"zero\" [0] or a \"one\" [1]? (ooooh, the suspense; or maybe the suspense has dissipated by now.)\n\n\n\n\n", "_____no_output_____" ], [ "## Prepare the Data", "_____no_output_____" ], [ "### Download the data \n(ooh look it's all stored on Amazon's AWS!)\n(pssst, we're in the cloooud)", "_____no_output_____" ] ], [ [ "# import MNIST data\n(x_train_temp, y_train_temp), (x_test_temp, y_test_temp) = mnist.load_data()", "_____no_output_____" ] ], [ [ "### **Look** at the data\n(always do this so that you **know** what the structure is.)", "_____no_output_____" ] ], [ [ "# Print the shapes\nprint(\"Train Data Shape:\", x_train_temp.shape)\nprint(\"Test Data Shape:\", x_test_temp.shape)\nprint(\"Train Label Shape:\", y_train_temp.shape)\nprint(\"Test Label Shape:\", y_test_temp.shape)", "_____no_output_____" ] ], [ [ "\n**Do the shapes of 'data' and 'label' (for train and test, respectively) match? If they don't now, Keras/TF will kindly yell at you later.**\n\n", "_____no_output_____" ] ], [ [ "# Print an example\nprint(\"Example:\")\nprint(\"y_train[0] is the label for the 0th image, and it is a\", y_train_temp[0])\nprint(\"x_train[0] is the image data, and you kind of see the pattern in the array of numbers\")\nprint(x_train_temp[0])", "_____no_output_____" ] ], [ [ "**Can you see the pattern of the number in the array?** ", "_____no_output_____" ] ], [ [ "# Plot the data! \nf = plt.figure()\nf.add_subplot(1,2, 1)\nplt.imshow(x_train_temp[0])\nf.add_subplot(1,2, 2)\nplt.imshow(x_train_temp[1])\nplt.show(block=True)", "_____no_output_____" ] ], [ [ "### Prepare the data\n\nData often need to be re-shaped and normalized for ingestion into the neural network.", "_____no_output_____" ], [ "#### Normalize the data\n\nThe images are recast as float and normalized to one for the network.\n\n", "_____no_output_____" ] ], [ [ "print(\"Before:\", np.min(x_train_temp), np.max(x_train_temp))\nx_train = x_train_temp.astype('float32')\nx_test = x_test_temp.astype('float32')\nx_train /= 255\nx_test /= 255\ny_train = y_train_temp\ny_test = y_test_temp\nprint(\"After:\", np.min(x_train), np.max(x_train))", "_____no_output_____" ] ], [ [ "#### Reshape the data arrays: set the input shape to be ready for a convolution [NEW]\n\nWe're going to use a Dense Neural Architecture, not as images, so we need to make the input shape appropriate.", "_____no_output_____" ] ], [ [ "# read the dimensions from one example in the training set\nimg_rows, img_cols = x_train[0].shape[0], x_train[0].shape[1]\n\n# Different NN libraries (e.g., TF) use different ordering of dimensions\n# Here we set the \"input shape\" so that later the NN knows what shape to expect\nif K.image_data_format() == 'channels_first':\n x_train = x_train.reshape(x_train.shape[0], 1, img_rows, img_cols)\n x_test = x_test.reshape(x_test.shape[0], 1, img_rows, img_cols)\n input_shape = (1, img_rows, img_cols)\nelse:\n x_train = x_train.reshape(x_train.shape[0], img_rows, img_cols, 1)\n x_test = x_test.reshape(x_test.shape[0], img_rows, img_cols, 1) \n input_shape = (img_rows, img_cols, 1)", "_____no_output_____" ] ], [ [ "#### Apply *one-hot encoding* to the data\n\n\n1. Current encoding provides a literal label. For example, the label for \"3\" is *3*.\n2. One-hot encoding places a \"1\" in an array at the appropriate location for that datum. For example, the label \"3\" becomes *[0, 0, 0, 1, 0, 0, 0, 0, 0, 0]*\n\nThis increases the efficiency of the matrix algebra during network training and evaluation.\n\n\n", "_____no_output_____" ] ], [ [ "# One-hot encoding\nnum_classes = 10\ny_train = keras.utils.to_categorical(y_train, num_classes)\ny_test = keras.utils.to_categorical(y_test, num_classes)", "_____no_output_____" ] ], [ [ "## Design Neural Network Architecture!", "_____no_output_____" ], [ "### Select model format", "_____no_output_____" ] ], [ [ "model = Sequential()", "_____no_output_____" ] ], [ [ "### Add layers to the model sequentially [NEW]", "_____no_output_____" ] ], [ [ "model.add(Conv2D(32, kernel_size=(3, 3), activation='relu', input_shape=input_shape))\nmodel.add(Dropout(0.25))\nmodel.add(Flatten())\nmodel.add(Dense(32, activation='relu'))\nmodel.add(Dropout(0.5))\nmodel.add(Dense(num_classes, activation='softmax'))\nmodel.summary()", "_____no_output_____" ] ], [ [ "*Things to think about and notice:*\n\n1. How does the \"output shape\" column change as you go through the network? How does this relate to pictures of CNNs you've seen (or might find on google images, for example)?\n2. What happens when you re-compile the [cell where you add layers sequentially](https://colab.research.google.com/drive/1wKzhJ0cOsJbgM9L0uIVUCYW1f2Zdf3PK#scrollTo=qXiW9aIx9_CM&line=3&uniqifier=1), without first compiling model-definition cell. Why does that happen?", "_____no_output_____" ], [ "### Compile the model\n\nSelect three key options\n1. **optimizer**: the method for optimizing the weights. \"Stochastic Gradient Descent (SGD)\" is the canonical method.\n2. **loss** function: the form of the function to encode the difference between the data's true label and the predict label.\n3. **metric**: the function by which the model is evaluated.", "_____no_output_____" ] ], [ [ "model.compile(optimizer=\"sgd\", loss='categorical_crossentropy', metrics=['accuracy'])", "_____no_output_____" ] ], [ [ "### Fit (read: Train) the model", "_____no_output_____" ] ], [ [ "# Training parameters\nbatch_size = 32 # number of images per epoch\nnum_epochs = 5 # number of epochs\nvalidation_split = 0.8 # fraction of the training set that is for validation only", "_____no_output_____" ], [ "# Train the model\nhistory = model.fit(x_train, y_train, \n batch_size=batch_size, \n epochs=num_epochs, \n validation_split=validation_split, \n verbose=True)", "_____no_output_____" ] ], [ [ "---\n*Things to think about and notice:*\n\n1. How fast is this training compared to the Dense/Fully Connected Networks? What could be a causing a difference between these two networks?\n2. Why is it taking a long time at the end of each epoch?\n\n", "_____no_output_____" ], [ "## Diagnostics!\n", "_____no_output_____" ], [ "#### Evaluate overall model efficacy\n\nEvaluate model on training and test data and compare. This provides summary values that are equivalent to the final value in the accuracy/loss history plots.", "_____no_output_____" ] ], [ [ "loss_train, acc_train = model.evaluate(x_train, y_train, verbose=False)\nloss_test, acc_test = model.evaluate(x_test, y_test, verbose=False)\nprint(f'Train acc/loss: {acc_train:.3}, {loss_train:.3}')\nprint(f'Test acc/loss: {acc_test:.3}, {loss_test:.3}')", "_____no_output_____" ] ], [ [ "#### Predict train and test data", "_____no_output_____" ] ], [ [ "y_pred_train = model.predict(x_train, verbose=True)\ny_pred_test = model.predict(x_test,verbose=True)", "_____no_output_____" ] ], [ [ "#### Plot accuracy and loss as a function of epochs (equivalently training time)\n", "_____no_output_____" ] ], [ [ "# set up figure\nf = plt.figure(figsize=(12,5))\nf.add_subplot(1,2, 1)\n\n# plot accuracy as a function of epoch\nplt.plot(history.history['acc'])\nplt.plot(history.history['val_acc'])\nplt.title('model accuracy')\nplt.ylabel('accuracy')\nplt.xlabel('epoch')\nplt.legend(['training', 'validation'], loc='best')\n\n# plot loss as a function of epoch\nf.add_subplot(1,2, 2)\nplt.plot(history.history['loss'])\nplt.plot(history.history['val_loss'])\nplt.title('model loss')\nplt.ylabel('loss')\nplt.xlabel('epoch')\nplt.legend(['training', 'validation'], loc='best')\nplt.show(block=True)", "_____no_output_____" ] ], [ [ "---\n*Things to think about and notice:*\n\n1. How do these curve shapes compare to the initial dense network results?\n\n\n", "_____no_output_____" ], [ "#### Confusion Matrix", "_____no_output_____" ] ], [ [ "# Function: Convert from categorical back to numerical value\ndef convert_to_index(array_categorical):\n array_index = [np.argmax(array_temp) for array_temp in array_categorical]\n return array_index\n\ndef plot_confusion_matrix(cm,\n normalize=False,\n title='Confusion matrix',\n cmap=plt.cm.Blues):\n \"\"\"\n This function modified to plots the ConfusionMatrix object.\n Normalization can be applied by setting `normalize=True`.\n \n Code Reference : \n http://scikit-learn.org/stable/auto_examples/model_selection/plot_confusion_matrix.html\n \n This script is derived from PyCM repository: https://github.com/sepandhaghighi/pycm\n \n \"\"\"\n\n plt_cm = []\n for i in cm.classes :\n row=[]\n for j in cm.classes:\n row.append(cm.table[i][j])\n plt_cm.append(row)\n plt_cm = np.array(plt_cm)\n if normalize:\n plt_cm = plt_cm.astype('float') / plt_cm.sum(axis=1)[:, np.newaxis] \n plt.imshow(plt_cm, interpolation='nearest', cmap=cmap)\n plt.title(title)\n plt.colorbar()\n tick_marks = np.arange(len(cm.classes))\n plt.xticks(tick_marks, cm.classes, rotation=45)\n plt.yticks(tick_marks, cm.classes)\n\n fmt = '.2f' if normalize else 'd'\n thresh = plt_cm.max() / 2.\n for i, j in itertools.product(range(plt_cm.shape[0]), range(plt_cm.shape[1])):\n plt.text(j, i, format(plt_cm[i, j], fmt),\n horizontalalignment=\"center\",\n color=\"white\" if plt_cm[i, j] > thresh else \"black\")\n\n plt.tight_layout()\n plt.ylabel('Actual')\n plt.xlabel('Predict')", "_____no_output_____" ], [ "# apply conversion function to data\ny_test_ind = convert_to_index(y_test)\ny_pred_test_ind = convert_to_index(y_pred_test)\n\n# compute confusion matrix\ncm_test = ConfusionMatrix(y_test_ind, y_pred_test_ind)\nnp.set_printoptions(precision=2)\n\n# plot confusion matrix result\nplt.figure()\nplot_confusion_matrix(cm_test,title='cm')", "_____no_output_____" ] ], [ [ "---\n*Things to think about and notice:*\n\n1. How does this confusion matrix compare to that from the Dense network?", "_____no_output_____" ], [ "## Problems for the CNNs (I mean ones that Wolf Blitzer can't solve)", "_____no_output_____" ], [ "---\n### Problem 1: There are a lot of moving parts here. A lot of in's and out's\n(bonus points if you know the 2000's movie, from which this is a near-quote.)\n\nSo, let's reduce the data set size at the beginning of the notebook.\n\nFor the rest of the exercises, we'd like to have the flexibility to experiment with larger networks (MOAR PARAMETERS, MOAR), so let's reduce the data set size. \n\n1. Go to the [cell where we download the data](https://colab.research.google.com/drive/1wKzhJ0cOsJbgM9L0uIVUCYW1f2Zdf3PK#scrollTo=qwXuui6_yYBv&line=2&uniqifier=1), and add a cell after it. \n2. Use array indexing and slicing to create a smaller training set. How about 5000? \n3. When train the model then, we'll want to update that validation fraction so that we get about 3000 in our training set.", "_____no_output_____" ], [ "---\n### Problem 2: Keeeep Learning! \n\n\nWhat happens if you run the cell that does the model-fitting again, right after doing it the first time. What do you notice about the loss and accuracy, as compared to when you did the fitting the first time?\n\nWhy do you think this is happening? \n", "_____no_output_____" ], [ "---\n### Problem 3: What happens if you add a maxpooling layer? \n\nDoes this change the training speed? \nWhy might this be? \nCheck the model summary output to see what effect the pooling layer has.", "_____no_output_____" ], [ "---\n### Problem 4: How deep can you make the network? \n\n1. Make a deep network and see how many parameters you can make. Is it trainable in a reasonable amount of time? Try add Conv layers, but not pooling layers.\n2. What if you want it to be efficient? Try adding a Max Pooling Layers after every Conv layer. How many layers can you possibly add now? Compile the model until you have an output shape of ( None, 1, 1 , #PARAMS) before the first dense layer.", "_____no_output_____" ], [ "---\n### Problem 5: Comparing performance and efficiency between CNNs and Dense Networks\n\nExperiment with the neural network above, and reduce the number of parameters to near that of the Dense network in the first exercise. \n\nIs there a CNN architecture that has the same number of parameters as the Dense network, but can perform better?\n\nRemember to think deeply, to pool your resources. When you're nearing the end it may not be as dense as it looks, but nearly so. \n", "_____no_output_____" ], [ "---\n### Problem 6: What happens to the training result when you degrade the images?\n\nIn this part, we will degrade the images by adding noise, and then by blurring the images, and we'll look at how the network training responds. \n\n", "_____no_output_____" ], [ "---\n### Problem 7: Let's see if we can look inside the neural networks\n\nUsing the [FAQ from Keras](https://keras.io/getting-started/faq/#how-can-i-obtain-the-output-of-an-intermediate-layer) or any other online resource, like examples from Github, can we make a plot of the feature maps for any of the layers, so we can see what the neural net sees?\n\n", "_____no_output_____" ], [ "---\n### Problem 8: Let's progress to Regression.\n\nConsider the labels as real values and modify the network to perform regression instead of classification on those values. You may want to consider the following:\n* normalizing the labels.\n* normalizing the image data.\n* modifying the activations that are used.\n* modifying the loss function that is appropriate for real-valued prediction. (see [keras loss](https://keras.io/losses/) )", "_____no_output_____" ], [ "# Activity 2: Compress Handwritten Digits with a Convolutional Autoencoder (CAE)\n", "_____no_output_____" ], [ "#### Add layers to the model sequentially [NEW]", "_____no_output_____" ] ], [ [ "autoencoder = Sequential()\n\n# Encoder Layers\nautoencoder.add(Conv2D(16, (3, 3), activation='relu', padding='same', input_shape=x_train.shape[1:]))\nautoencoder.add(MaxPooling2D((2, 2), padding='same'))\nautoencoder.add(Conv2D(8, (3, 3), activation='relu', padding='same'))\nautoencoder.add(MaxPooling2D((2, 2), padding='same'))\nautoencoder.add(Conv2D(8, (3, 3), strides=(2,2), activation='relu', padding='same'))\nautoencoder.add(MaxPooling2D((2, 2), padding='same'))\nautoencoder.add(Conv2D(8, (3, 3), strides=(2,2), activation='relu', padding='same'))\nautoencoder.add(MaxPooling2D((2, 2), padding='same'))\n\n# Flatten encoding for visualization\nautoencoder.add(Flatten())\nautoencoder.add(Reshape((1, 1, 8)))\n\n# Decoder Layers\nautoencoder.add(UpSampling2D((2, 2)))\nautoencoder.add(Conv2D(8, (3, 3), activation='relu', padding='same'))\nautoencoder.add(UpSampling2D((2, 2)))\nautoencoder.add(Conv2D(8, (3, 3), activation='relu', padding='same'))\nautoencoder.add(UpSampling2D((2, 2)))\nautoencoder.add(Conv2D(8, (3, 3), activation='relu', padding='same'))\nautoencoder.add(UpSampling2D((2, 2)))\nautoencoder.add(Conv2D(16, (3, 3), activation='relu'))\nautoencoder.add(UpSampling2D((2, 2)))\nautoencoder.add(Conv2D(1, (3, 3), activation='sigmoid', padding='same'))\n\nautoencoder.summary()", "_____no_output_____" ] ], [ [ "#### Create a separate model that is just the encoder\n\nThis will allow us to encode the images and look at what the encoding results in. ", "_____no_output_____" ] ], [ [ "encoder = Model(inputs=autoencoder.input, outputs=autoencoder.get_layer('flatten_8').output)\nencoder.summary()", "_____no_output_____" ] ], [ [ "#### Compile the autencoder", "_____no_output_____" ] ], [ [ "autoencoder.compile(optimizer='adam', loss='binary_crossentropy')\nnum_epochs = 10", "_____no_output_____" ] ], [ [ "#### Plot the input, output, and encoded images", "_____no_output_____" ] ], [ [ "# set number of images to visualize\nnum_images = 10\n\n# select random subsect to visualize\nnp.random.seed(42)\nrandom_test_images = np.random.randint(x_test.shape[0], size=num_images)\n\n# encode images\nencoded_imgs = encoder.predict(x_test)\n\n#decode encode AND decode images\ndecoded_imgs = autoencoder.predict(x_test)\n\n\n# plot figure\nplt.figure(figsize=(18, 4))\n\nnum_rows=4\nnum_pixel_x = 2\nnum_pixel_y = 4\n\nfor i, image_idx in enumerate(random_test_images):\n # plot original image\n ax = plt.subplot(4, num_images, i + 1)\n plt.imshow(x_test[image_idx].reshape(28, 28))\n plt.gray()\n ax.get_xaxis().set_visible(False)\n ax.get_yaxis().set_visible(False)\n \n # plot encoded image\n ax = plt.subplot(num_rows, num_images, num_images + i + 1)\n\n plt.imshow(encoded_imgs[image_idx].reshape(num_pixel_x, num_pixel_y), interpolation=None, resample=None)\n plt.gray()\n ax.get_xaxis().set_visible(False)\n ax.get_yaxis().set_visible(False)\n \n # plot reconstructed image\n ax = plt.subplot(num_rows, num_images, 2*num_images + i + 1)\n plt.imshow(decoded_imgs[image_idx].reshape(28, 28))\n plt.gray()\n ax.get_xaxis().set_visible(False)\n ax.get_yaxis().set_visible(False)\nplt.show()", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown", "markdown" ], [ "code", "code" ], [ "markdown", "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code", "code" ], [ "markdown", "markdown", "markdown", "markdown", "markdown", "markdown", "markdown", "markdown", "markdown", "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "## Main Hartree Code\n<br>\nHartree-Fock Computational Chemistry Method implemented in Python as described in <u>Modern Quantum Chemistry Introduction to Advanced Electronic Structure Theory</u>, by Attila Szabo and Neil S. Ostlund. <br> <br>\nThroughout the rest of the modules in this notebook, the entire text of Modern Quantum Chemistry will simply be refered to as \"Szabo\" for the sake of brevity. \n<br> <br>\nProgram is limited to molecules consisting of hydrogen and helium with only even numbers of electrons. Required user input is the atomic number, 3D space location, and electron number for each atom. A basis set specification is also required. <br>\n<br>\nThe program makes use of Hartree Atomic Units such that the distances and lengths are described in Bohr Radius, energy is in Hartrees and mass is in Hartree Atomic Units, such that the mass of a proton is equal to 1836 atomic units. More information on Hartree Atomic Units can be found on page 41 of Szabo. ", "_____no_output_____" ] ], [ [ "#Python Implementation of the Hartree Fock Method\n#Procedures listed in the code follow as described in Modern Quantum Chemistry: \n#Introduction to Advanced Electronic Structure Theory, By Attila Szabo and Neil S. Ostlund\nimport sys\nsys.path.append(\"..\\\\Comp_Chem_Package\")\nimport numpy as np\nfrom molecule import atom\nfrom vector import vector\nfrom molecule import gaussian\nfrom molecule import molecule\nfrom notebookImporter import importNotebook\n\n#import integrals notebook for the hartree method\nintegrals = importNotebook(\"Hartree_Integrals\")\nscf = importNotebook(\"Hartree_SCF\")\n\n#define SCF convergence critera, and max number of iteration cycles\nSCF_CONVERGENCE = pow(10, -15)\nMAX_ITERATIONS = 500\n\n#Step 1\n#Specify Molecules, Nuclear Coordinates, and Charge of the nucli Number of Electrons,\n\n#generate an h2 atom with a distance of 1.4 AU to compare with Szabo pg. 160\n\n#R is in units of Bohr Radius \nR = 1.4\nsystem = molecule()\nsystem.addAtom(atom(vector(1,1,1), 1, 1))\nsystem.addAtom(atom(vector(1,1,1 + R), 1, 1))\n\n#add a basis set\nsystem.addBasis(\"STO-3G\")\n\nsystem.display()", "Molecule\nBasis Set: STO-3G\nTotal Number of Electrons: 2\nAtoms: \n Atomic Number: 1, Electrons: 1, Coordinate: X: 1 Y: 1 Z: 1\n Atomic Number: 1, Electrons: 1, Coordinate: X: 1 Y: 1 Z: 2.4\n" ], [ "#Step 2\n#Calculate Integrals\n#Overlap, KE, Nuclear Attraaction, and Electron Repulsion\nS = integrals.overlap(system)\nprint(\"Overlap Matrix: \")\nprint(np.matrix(S))\nprint()\n\nT = integrals.kineticEnergy(system)\nprint(\"Electron Kinetic Energy Matrix: \")\nprint(np.matrix(T))\nprint()\n\nV = integrals.nuclearAttraction(system)\nfor index, atom in enumerate(V):\n print(\"Nucli \" + str(index) + \"-Electron Attraction Matrix: \")\n print(np.matrix(atom))\nprint()\n\nelectronRepulsion = integrals.electronElectronRepulsion(system)\nprint(\"Electron Repulsion Tensor: \")\nprint(np.array(electronRepulsion))\nprint()\n\n#Form the electronic hamiltonian\nH = np.matrix(T)\n\n#add in all of the nuclear attractions matricies to the hamiltonian\nfor atom in V:\n H += np.matrix(atom)\n \nprint(\"Electronic Hamiltonian :\")\nprint(H)\nprint()", "Overlap Matrix: \n[[1. 0.65931821]\n [0.65931821 1. ]]\n\nElectron Kinetic Energy Matrix: \n[[0.76003188 0.23645466]\n [0.23645466 0.76003188]]\n\nNucli 0-Electron Attraction Matrix: \n[[-1.22661373 -0.59741731]\n [-0.59741731 -0.65382716]]\nNucli 1-Electron Attraction Matrix: \n[[-0.65382716 -0.59741731]\n [-0.59741731 -1.22661373]]\n\nElectron Repulsion Tensor: \n[[[[0.77460594 0.44410766]\n [0.44410766 0.56967593]]\n\n [[0.44410766 0.29702854]\n [0.29702854 0.44410766]]]\n\n\n [[[0.44410766 0.29702854]\n [0.29702854 0.44410766]]\n\n [[0.56967593 0.44410766]\n [0.44410766 0.77460594]]]]\n\nElectronic Hamiltonian :\n[[-1.12040901 -0.95837996]\n [-0.95837996 -1.12040901]]\n\n" ], [ "#Prepare for the SCF procedure\n\n#get size of the basis set \nsize = len(S)\n\n#compute the Transformation Matrix\nX = scf.X(S, size)\n\n#get guess Fock matrix, assume 2-electron term is equal to 0\nF = H\n\nprint(F)", "[[-1.12040901 -0.95837996]\n [-0.95837996 -1.12040901]]\n" ], [ "# SCF Procedure \n\n#init list to store the energy from each iteration\n#as well as a boolean to signify whether the loop has converged\nE = []\nconverged = False\n\nwhile( not converged ):\n \n #diagnolze the Fock matrix and convert it to MO basis \n F = X.transpose() * F * X \n print(\"F**\",F)\n \n #diagnolize the Fock Matrix to obtain the MOs and the their respective energies\n MOEnergy, MO = np.linalg.eigh(F)\n \n #Transform the MO basis MOs to an AO basis\n C = X * MO\n print(\"C\", C)\n #compute the electron density, the two electron term, and then use G to compute the new Fock matrix\n P = scf.densityMatrix(C, system.N, size)\n G = scf.G(electronRepulsion, P, size)\n F = H + G\n \n #compute the new expectation energy\n #Expectation Energy is in units of Hartrees\n E.append(scf.expectationEnergy(H, F, P))\n \n #check if at least two SCF iterations have occured\n #if more than two have occured, then check if the difference betweeen this E, \n #and the previous E is less then the covergence value, if yes, end the SCF loop\n #if energy has not converged, check whether the max number of iterations have occured so far\n sizeE = len(E)\n if(len(E) > 2):\n if(abs(E[sizeE-2] - E[sizeE-1]) < SCF_CONVERGENCE):\n converged = True\n elif(sizeE > MAX_ITERATIONS):\n print(\"SCF Failed to Converge\")\n break\n \n #compute total energy of the system including nuclear-nuclear repulsion\n totalE = E[sizeE-1] + scf.nuclearRepulsion(system)\n \n #display information about current SCF iteration to the user\n print(\"SCF Iteration #\" + str(sizeE) + \", Electronic Energy: \" + str(E[sizeE-1]) + \" Hartrees, Total Energy: \" + str(totalE) + \" Hartrees\")\n print(\"F\", F)\n print()\nprint(\"-\"*50)\nprint()\nprint(\"Final SCF Energy: \" + str(E[sizeE-1]))", "F** [[-4.75602306e-01 -1.08869060e-16]\n [-2.35585295e-16 -1.25279706e+00]]\nC [[ 0.54893404 1.21146407]\n [ 0.54893404 -1.21146407]]\nSCF Iteration #1, Electronic Energy: -1.8310000394614832 Hartrees, Total Energy: -1.116714325175769 Hartrees\nF [[-0.36553735 -0.59388537]\n [-0.59388537 -0.36553735]]\n\nF** [[ 6.70267761e-01 -2.12833504e-16]\n [-2.35379740e-16 -5.78202977e-01]]\nC [[-0.54893404 1.21146407]\n [-0.54893404 -1.21146407]]\nSCF Iteration #2, Electronic Energy: -1.8310000394614834 Hartrees, Total Energy: -1.116714325175769 Hartrees\nF [[-0.36553735 -0.59388537]\n [-0.59388537 -0.36553735]]\n\nF** [[ 6.70267761e-01 -1.51889583e-16]\n [-3.69879271e-16 -5.78202977e-01]]\nC [[-0.54893404 1.21146407]\n [-0.54893404 -1.21146407]]\nSCF Iteration #3, Electronic Energy: -1.8310000394614832 Hartrees, Total Energy: -1.116714325175769 Hartrees\nF [[-0.36553735 -0.59388537]\n [-0.59388537 -0.36553735]]\n\n--------------------------------------------------\n\nFinal SCF Energy: -1.8310000394614832\n" ], [ "X.transpose() * H * X", "_____no_output_____" ], [ "X.transpose() * H", "_____no_output_____" ], [ "X[0,1]", "_____no_output_____" ], [ "np.matmul(X, H)", "_____no_output_____" ], [ "X * H", "_____no_output_____" ] ] ]

[ "markdown", "code" ]

[ [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code", "code", "code" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "import numpy as np\nimport pandas as pd", "_____no_output_____" ], [ "dataset = pd.read_csv('spam.csv')", "_____no_output_____" ], [ "result = list(map(lambda x: 'no_spam' if x == 'ham' else 'spam', dataset['type']))\ndataset['Type'] = result\ndataset = dataset.drop('type', axis = 1)", "_____no_output_____" ], [ "dataset", "_____no_output_____" ], [ "from sklearn.model_selection import train_test_split\n\nX_train, X_test, y_train, y_test = train_test_split(dataset['text'], dataset['Type'], test_size=0.25, random_state=0)", "_____no_output_____" ], [ "# Both vectorizer method can be used\n\nfrom sklearn.feature_extraction.text import TfidfVectorizer\n\nvectorizer = TfidfVectorizer()\nX_train = vectorizer.fit_transform(X_train)\nX_test = vectorizer.transform(X_test)\n\n'''\nfrom sklearn.feature_extraction.text import CountVectorizer\n\nvectorizer = CountVectorizer()\nX_train = vectorizer.fit_transform(X_train)\nX_test = vectorizer.transform(X_test)\n'''", "_____no_output_____" ], [ "vectorizer.get_feature_names()", "_____no_output_____" ], [ "X_train.toarray()", "_____no_output_____" ], [ "from sklearn.naive_bayes import MultinomialNB\n\nmodel = MultinomialNB()\nmodel.fit(X_train, y_train)\nprediction = model.predict(X_test)", "_____no_output_____" ], [ "from sklearn.metrics import confusion_matrix, classification_report\n\nprint(classification_report(y_test, prediction))\nprint(confusion_matrix(y_test, prediction))", " precision recall f1-score support\n\n no_spam 0.96 1.00 0.98 1221\n spam 1.00 0.71 0.83 169\n\n accuracy 0.96 1390\n macro avg 0.98 0.86 0.91 1390\nweighted avg 0.97 0.96 0.96 1390\n\n[[1221 0]\n [ 49 120]]\n" ], [ "train_data = pd.read_csv('train_fake_news.csv', index_col = 'id')", "_____no_output_____" ], [ "train_data = train_data.rename(columns = {'title': 'Title', 'author': 'Author',\n 'text': 'Text', 'label': 'Label'})\nidx = train_data.index\ntrain_data.index = idx.rename('ID')\n\nlenght = [len(str(train_data.iloc[i]['Text'])) for i in range(len(train_data))]\ntrain_data['Lenght'] = lenght\ntrain_data = train_data[train_data['Lenght'] > 50]", "_____no_output_____" ], [ "train_data = train_data.drop(['Title', 'Author'], axis = 1)\ntrain_data = train_data.dropna(axis = 0)", "_____no_output_____" ], [ "from sklearn.model_selection import train_test_split\n\nX_train_data, X_test_data, y_train_data, y_test_data = train_test_split(train_data['Text'][:10000], train_data['Label'][:10000], test_size=0.20, random_state=0)", "_____no_output_____" ], [ "vectorizer = TfidfVectorizer()\nX_train_data = vectorizer.fit_transform(X_train_data)\nX_test_data = vectorizer.transform(X_test_data)", "_____no_output_____" ], [ "len(X_test_data.toarray())", "_____no_output_____" ], [ "from sklearn.naive_bayes import MultinomialNB\n\nmodel = MultinomialNB()\nmodel.fit(X_train_data, y_train_data)\nprediction_data = model.predict(X_test_data)", "_____no_output_____" ], [ "from sklearn.metrics import confusion_matrix, classification_report\n\nprint(classification_report(y_test_data, prediction_data))\nprint(confusion_matrix(y_test_data, prediction_data))", " precision recall f1-score support\n\n 0 0.73 0.99 0.84 1003\n 1 0.99 0.63 0.77 997\n\n accuracy 0.81 2000\n macro avg 0.86 0.81 0.80 2000\nweighted avg 0.86 0.81 0.80 2000\n\n[[996 7]\n [373 624]]\n" ], [ "prediction_data", "_____no_output_____" ] ] ]

[ "code" ]

[ [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Tracker\n\n[](https://github.com/lab-ml/labml)\n[](https://colab.research.google.com/github/lab-ml/labml/blob/master/guides/tracker.ipynb)\n[](https://docs.labml.ai/api/tracker.html)\n\n\nHere you specify indicators and the logger stores them temporarily and write in batches.\nIt can aggregate and write them as means or histograms.", "_____no_output_____" ] ], [ [ "%%capture\n!pip install labml", "_____no_output_____" ], [ "import time\n\nimport numpy as np\n\nfrom labml import tracker\n\n# dummy train function\ndef train():\n return np.random.randint(100)\n\n# Reset global step because we incremented in previous loop\ntracker.set_global_step(0)", "_____no_output_____" ] ], [ [ "This stores all the loss values and writes the logs the mean on every tenth iteration.\nConsole output line is replaced until\n[`labml.tracker.new_line`](https://docs.labml.ai/api/tracker.html#labml.tracker.new_line)\nis called.", "_____no_output_____" ] ], [ [ "for i in range(1, 401):\n tracker.add_global_step()\n loss = train()\n tracker.add(loss=loss)\n if i % 10 == 0:\n tracker.save()\n if i % 100 == 0:\n tracker.new_line()\n time.sleep(0.02)", "_____no_output_____" ] ], [ [ "## Indicator settings", "_____no_output_____" ] ], [ [ "# dummy train function\ndef train2(idx):\n return idx, 10, np.random.randint(100)\n\n# Reset global step because we incremented in previous loop\ntracker.set_global_step(0)", "_____no_output_____" ] ], [ [ "Histogram indicators will log a histogram of data.\nQueue will store data in a `deque` of size `queue_size`, and log histograms.\nBoth of these will log the means too. And if `is_print` is `True` it will print the mean.", "_____no_output_____" ], [ "queue size of `10` and the values are printed to the console", "_____no_output_____" ] ], [ [ "tracker.set_queue('reward', 10, True)", "_____no_output_____" ] ], [ [ "By default values are not printed to console; i.e. `is_print` defaults to `False`.", "_____no_output_____" ] ], [ [ "tracker.set_scalar('policy')", "_____no_output_____" ] ], [ [ "Settings `is_print` to `True` will print the mean value of histogram to console", "_____no_output_____" ] ], [ [ "tracker.set_histogram('value', True)", "_____no_output_____" ], [ "for i in range(1, 400):\n tracker.add_global_step()\n reward, policy, value = train2(i)\n tracker.add(reward=reward, policy=policy, value=value, loss=1.)\n if i % 10 == 0:\n tracker.save()\n if i % 100 == 0:\n tracker.new_line()", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "<a href=\"https://colab.research.google.com/github/DiploDatos/AnalisisyVisualizacion/blob/master/01_Probabilidad.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>", "_____no_output_____" ], [ "\n**Diplomatura en Ciencia de Datos, Aprendizaje Automático y sus Aplicaciones**\n\n**Edición 2021**\n\n---\n\n# Variables Aleatorias y Probabilidad\n\nEn esta notebook, vamos a realizar una primera aproximación al conjunto de datos. \n\n* Variables aleatorias y sus distintos tipos\n* Probabilidad", "_____no_output_____" ] ], [ [ "import io\nimport matplotlib\nimport matplotlib.pyplot as plt\nimport numpy\nimport pandas as pd\nimport seaborn\n\nseaborn.set_context('talk')", "_____no_output_____" ] ], [ [ "## Lectura del dataset\n\nEn la notebook 00 se explican los detalles de la siguiente sección.", "_____no_output_____" ] ], [ [ "url = 'https://cs.famaf.unc.edu.ar/~mteruel/datasets/diplodatos/sysarmy_survey_2020_processed.csv'\ndf = pd.read_csv(url)", "_____no_output_____" ], [ "df[:3]", "_____no_output_____" ] ], [ [ "# Análisis de salarios\n\nLa primera pregunta que se nos ocurre al ver esta encuenta es: **\"¿Y cuánto cobran los programadores en Argentina?\"**.\n\nEste es un punto de partida para el análisis del conjunto de datos. El proceso total constará de varias iteraciones: a medida que se obtengan conclusiones, se descrubrirán otros aspectos relevantes de los datos, lo cual disparará nuevas preguntas.", "_____no_output_____" ], [ "Para conocer más sobre la distribución de los salarios, es necesario elegir una columna de la encuesta para analizar.", "_____no_output_____" ] ], [ [ "salary_col = 'salary_monthly_NETO'", "_____no_output_____" ] ], [ [ "Una buena forma de comenzar una exploración es a través de la visualización. Seaborn nos provee un tipo de gráfico específico para graficar columnas que contienen números, llamado `displot`. (No confundir con `distplot`, que está deprecado). \n\nEl gráfico generado es un **histograma** de frecuencias. En el eje x se grafican los valores que toma la columna, divididos en intervalos o bins. En el eje y se grafica el conteo de ocurrencias de valores en cada intervalo.", "_____no_output_____" ] ], [ [ "seaborn.displot(df[salary_col], aspect=2)\n## para evitar la notación científica en las etiquetas\nplt.ticklabel_format(style='plain', axis='x')", "_____no_output_____" ] ], [ [ "### ¿Qué estamos viendo?\n\nLas visualizaciones simples son prácticas para conocer la forma de los datos rápidamente, porque condensan mucha información. Por ejemplo:\n* El rango de valores tomados por la columna va desde 0 hasta aproximadamente 2M.\n* La mayoría de ls valores se condensa por debajo de los 250K, y pocos superan los 500K.\n* Los valores más frencuentes aparentan estar cerca de los 100K.\n* Hay un pico de ocurrencias en el valor 0.\n y brindan poco detalle.\n\n## Ejercicio: ¿Qué otro tipo de preguntas nos podemos hacer en este punto que no se responden con un histograma?\n", "_____no_output_____" ], [ "### Análisis, ¡fundamentado!\n\nPara continuar el análisis, es necesario aplicar herramientas teóricas que nos brinda la Estadística y la Probabilidad.", "_____no_output_____" ], [ "## Variables aleatorias y sus tipos\n\nEn base a la definición de variable aleatoria que discutimos, se puede hablar de que cada columna de nuestro dataset es un **variable aleatoria**, y que su valor en cada respuesta es una **realización** de dicha variable. Pero, ¿qué tipo tienen esas variables?", "_____no_output_____" ], [ "### V.A. numéricas\n\nEl salario, la edad, los años de experiencia, son variables aleatorias cuyo rango es un conjunto numérico. Podemos clasificarlas en **continuas** o **discretas**, aunque esa distinción se vuelve difusa cuando trabajamos con datos computacionalmente. ¿Por qué? \n\n* Datos que en teoría son continuos, se miden de manera discreta. Por ejemplo, los *años* de experiencia, la altura de una persona en *centímetros*.\n* Datos que en teoría son continuos, se discretizan a fines prácticos. Por ejemplo, la edad, el salario en pesos argentinos.", "_____no_output_____" ], [ "Para analizar datos continuos se usan frecuentemente los histogramas, como en el caso anterior de los sueldos.\n\n**¡Tip!** Antes de graficar, controlar el rango (ya que seaborn intentará crear miles de segmentos si el rango es muy grande) y remover los valores nulos.", "_____no_output_____" ] ], [ [ "# Obtenemos el rango de valores observados de la variable\ndf.profile_age.min(), df.profile_age.max()", "_____no_output_____" ], [ "seaborn.displot(df.profile_age[df.profile_age < 100].dropna(),\n stat='count', aspect=4)", "_____no_output_____" ] ], [ [ "Sin embargo, los histogramas pueden ocultar información. ¿Por qué? Porque agrupan rangos de valores en intervalos inferidos automáticamente. Como resultado, la visualización varía de con distintas longitudes de segmentos. Comparemos los siguientes histogramas.", "_____no_output_____" ] ], [ [ "# Un ejemplo más avanzado\nfig, ax = plt.subplots(nrows=2, ncols=3, figsize=(15,10), sharey='row')\nseaborn.histplot(df.profile_age[df.profile_age < 100].dropna(), ax=ax[0,0],\n stat='count')\nseaborn.histplot(df.profile_age[df.profile_age < 100].dropna(), ax=ax[0,1],\n bins=20, stat='count')\nseaborn.histplot(df.profile_age[df.profile_age < 100].dropna(), ax=ax[0,2],\n bins=5, stat='count')\n\nseaborn.histplot(df.profile_age[df.profile_age < 100].dropna(), ax=ax[1,0],\n stat='frequency')\nseaborn.histplot(df.profile_age[df.profile_age < 100].dropna(), ax=ax[1,1],\n bins=20, stat='frequency')\nseaborn.histplot(df.profile_age[df.profile_age < 100].dropna(), ax=ax[1,2],\n bins=5, stat='frequency')\n\nfig.show()", "_____no_output_____" ] ], [ [ "Para variables discretas puede usarse un gráfico de línea, que permite visualizar el conteo de cada uno de los puntos en el rango observado.\n\n**¿Se puede usar un gráfico de líneas para la variable `salary_montly_NETO`? ¿Tiene sentido?** ", "_____no_output_____" ] ], [ [ "fig = plt.figure(figsize=(16,4))\nage_counts = df[df.profile_age < 100].profile_age.value_counts()\nseaborn.lineplot(age_counts.index, age_counts.values, color='steelblue')\nplt.xticks(fontsize=14) # Achicamos la letra para que se vea mejor\nseaborn.despine()", "/usr/local/lib/python3.7/dist-packages/seaborn/_decorators.py:43: FutureWarning: Pass the following variables as keyword args: x, y. From version 0.12, the only valid positional argument will be `data`, and passing other arguments without an explicit keyword will result in an error or misinterpretation.\n FutureWarning\n" ] ], [ [ "### V.A. categóricas\n\nLas variables categóricas toman valores de un conjunto pre-definido, usualmente pero no necesariamente finito. Para visualizarlas, puede usarse un gráfico de barras, que representa cada valor observado con una columna, y el conteo de ese valor con la altura de la columna.\n\nLas variables numéricas discretas, ¿son categóricas?", "_____no_output_____" ] ], [ [ "df.profile_gender.unique()", "_____no_output_____" ], [ "fig = plt.figure(figsize=(8,6))\nseaborn.countplot(df.profile_gender, color='steelblue')", "/usr/local/lib/python3.7/dist-packages/seaborn/_decorators.py:43: FutureWarning: Pass the following variable as a keyword arg: x. From version 0.12, the only valid positional argument will be `data`, and passing other arguments without an explicit keyword will result in an error or misinterpretation.\n FutureWarning\n" ] ], [ [ "Las variables categóricas pueden ser *ordinales*, si existe un orden lógico entre sus valores. Esto es independiente de que sean numéricas. En caso de que un orden exista, es adecuado incluirlo en el gráfico.", "_____no_output_____" ] ], [ [ "sorted_studies_levels = ['Primario', 'Secundario', 'Terciario', 'Universitario',\n 'Posgrado', 'Doctorado', 'Posdoctorado']\nfig, axes = plt.subplots(ncols=2, figsize=(15,6))\ng = seaborn.countplot(df.profile_studies_level, color='steelblue', ax=axes[0])\ng = seaborn.countplot(df.profile_studies_level, color='steelblue', ax=axes[1],\n order=sorted_studies_levels)\nfor ax in axes:\n ax.tick_params(labelrotation=30)", "/usr/local/lib/python3.7/dist-packages/seaborn/_decorators.py:43: FutureWarning: Pass the following variable as a keyword arg: x. From version 0.12, the only valid positional argument will be `data`, and passing other arguments without an explicit keyword will result in an error or misinterpretation.\n FutureWarning\n/usr/local/lib/python3.7/dist-packages/seaborn/_decorators.py:43: FutureWarning: Pass the following variable as a keyword arg: x. From version 0.12, the only valid positional argument will be `data`, and passing other arguments without an explicit keyword will result in an error or misinterpretation.\n FutureWarning\n" ] ], [ [ "### Tipos de variables vs tipos de datos\n\nTenemos que distinguir dos conceptos con el mismo nombre y significado similar, pero que no son iguales:\n - **tipo de la variable aleatoria** es el tipo de valores con los que decidimos *intepretar* las realizaciones\n - **tipo de datos** es un concepto de programación que indica en qué formato se representa la información. Cuando asignamos a una variable `age` *del programa de Python* una realización de una variable aleatoria conceptual `profile_age`, esa variable `age` también tiene un *tipo de Python*, por ejemplo `int` o `float`.\n", "_____no_output_____" ] ], [ [ "age = df.profile_age.iloc[0]\ntype(age)", "_____no_output_____" ] ], [ [ "*¡Importante!* Hay que tener en cuenta también los límites de la capacidad computacional al momento de representar entidades matemáticas.\n* Los números reales siempre son \"redondeados\" a una representación racional.\n* Los tipos básicos como `Int` sólo pueden representar números en un rango, por ejemplo `(-2^31, 2^31)`. Exceder el rango puede tener consecuencias inesperadas, como `integer overflow`.\n\n¿Por qué es importante saberlo? Porque se pueden producir errores de redondeo u obtener resultados aproximados.", "_____no_output_____" ] ], [ [ "print(type(3), type(3.44), type(1/3)) # 1/3 es un numero irracional\nimport numpy\nprint(numpy.iinfo('int64').min, numpy.iinfo('int64').max)\nnumpy.int64(numpy.iinfo('int64').max) + 1\n# Traten de hacer numpy.int64(numpy.iinfo('int64').max + 1)", "<class 'int'> <class 'float'> <class 'float'>\n-9223372036854775808 9223372036854775807\n" ] ], [ [ "Se puede acceder a los tipos de datos del DataFrame. El tipo `object` se utiliza para representar cualquier variable que no sea numérica, como por ejemplo los `str`.", "_____no_output_____" ] ], [ [ "df.dtypes[:10]", "_____no_output_____" ] ], [ [ "Hay que tener en cuenta que las librerías de gráficos nos permitirán crear las visualizaciones que querramos, mientras los tipos de datos sean los adecuados.\n\nPor ejemplo, podemos hacer un histograma con la variable `profile_open_source_contributions` si la transformamos a tipo `bool` (que se representa internamente como un tipo entero). Sin embargo, esto no tiene ningún sentido.", "_____no_output_____" ] ], [ [ "df.loc[:,'salary_in_usd_bool'] = \\\n df.salary_in_usd.replace({'Mi sueldo está dolarizado': True}).fillna(False)\nprint(df.salary_in_usd.unique(), df.salary_in_usd_bool.unique())", "[nan 'Mi sueldo está dolarizado'] [False True]\n" ], [ "seaborn.histplot(df.salary_in_usd_bool, bins=5)", "<string>:6: RuntimeWarning: Converting input from bool to <class 'numpy.uint8'> for compatibility.\n<string>:6: RuntimeWarning: Converting input from bool to <class 'numpy.uint8'> for compatibility.\n" ] ], [ [ "También podemos graficar la frecuencia de una variable categórica utilizando un gráfico de líneas. **¿Por qué esta visualización no es correcta?**", "_____no_output_____" ] ], [ [ "count_by_province = df.work_province.value_counts()\nfig = plt.figure(figsize=(16, 4))\nseaborn.lineplot(x=count_by_province.index, y=count_by_province.values)\nplt.xticks(rotation=45)\nseaborn.despine()", "_____no_output_____" ] ], [ [ "# Análisis del impacto de los años de experiencia\n\nAhora que ya sabemos aproximadamente la forma de nuestros datos, podemos pasar a realizar otra pregunta (otra iteración del proceso de análisis): \n\n**¿Tener más años de experiencia significa que se cobra más?**\n\nPara responder a esta pregunta, analizamos la probabilidad de que un programador tenga un salario mensual mayor que el promedio, cuando tiene una experiencia mayor que 5 años.\n\n", "_____no_output_____" ] ], [ [ "avg_salary = df[salary_col].mean()\navg_salary", "_____no_output_____" ] ], [ [ "## Medida de probabilidad\n\nEn el teórico vimos que si cada una de nuestros eventos es independiente e idénticamente distribuido, es decir, que $P(\\{\\omega_i\\})=1/k$, entonces la probabilidad de un conjunto $A \\subset \\Omega$ es la proporción de $A$, donde .\n\n\n$$P(\\{\\omega_i\\})=1/k \\implies P(A)=|A|/|\\Omega|=|A|/k$$\n\n\nEn este problema en particular, $\\Omega$ son todas las respuestas del dataset, cada $a_i$ es una variable que representa una respuesta, y el conjunto $A$ son las respuestas (filas) en la que la columna `salary_col` tiene un valor mayor que el promedio\n", "_____no_output_____" ] ], [ [ "p_above_avg = len(df[df[salary_col] >= avg_salary]) / len(df)\np_above_avg", "_____no_output_____" ] ], [ [ "* ¿Por qué podemos usar la teoría de la probabilidad?\n* ¿Por qué calculamos una probabilidad con esta fórmula?\n* ¿Cómo podemos interpretar esta probabilidad?", "_____no_output_____" ], [ "## Probabilidad condicional\n\nAhora podemos pasar a hablar de la probabilidad condicional entre los dos eventos. La definimos como\n\n$$P(A|B) = \\frac{P(A \\cap B)}{P(B)}$$\n\nEsto es equivalente a:\n\n$$P(A|B) = \\frac{|A \\cap B|}{|B|}$$\n\n## Ejercicio\n\nReponder: **¿Si uno tiene más de 5 años de experiencia, la probabilidad de cobrar más que el promedio aumenta? ¿Estos eventos, son independientes?**\n", "_____no_output_____" ] ], [ [ "is_above_avg = df[salary_col] > avg_salary\nexperience_greater_5 = df.profile_years_experience > 5\nintersection_count = len(df[is_above_avg & experience_greater_5])", "_____no_output_____" ], [ "p_above_avg_given_experience = 0\np_above_avg_given_experience", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown", "markdown", "markdown", "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Umbrella Sampling", "_____no_output_____" ] ] ]

[ "markdown" ]

[ [ "markdown" ] ]

[ "BSD-3-Clause" ]

[ "BSD-3-Clause" ]

[ "BSD-3-Clause" ]

[ [ [ "# Execution Plugins\n\nAs you learned in the [Workflow](basic_workflow.ipynb) tutorial, a workflow is executed with the ``run`` method. For example:\n\n workflow.run()\n\nWhenever you execute a workflow like this, it will be executed in serial order. This means that no node will be executed in parallel, even if they are completely independent of each other. Now, while this might be preferable under certain circumstances, we usually want to executed workflows in parallel. For this, Nipype provides many different plugins.", "_____no_output_____" ], [ "## Local execution\n\n### ``Linear`` Plugin\n\nIf you want to run your workflow in a linear fashion, just use the following code:\n\n workflow.run(plugin='Linear')", "_____no_output_____" ], [ "### ``MultiProc`` Plugin\n\nThe easiest way to executed a workflow locally in parallel is the ``MultiProc`` plugin:\n\n workflow.run(plugin='MultiProc', plugin_args={'n_procs': 4})\n\nThe additional plugin argument ``n_procs``, specifies how many cores should be used for the parallel execution. In this case, it's 4.\n\nThe `MultiProc` plugin uses the [multiprocessing](http://docs.python.org/library/multiprocessing.html) package in the standard library, and is the only parallel plugin that is guaranteed to work right out of the box.", "_____no_output_____" ], [ "## Cluster execution\n\nThere are many different plugins to run Nipype on a cluster, such as: ``PBS``, ``SGE``, ``LSF``, ``Condor`` and ``IPython``. Implementing them is as easy as ``'MultiProc'``.\n\n workflow.run('PBS', plugin_args={'qsub_args': '-q many'})\n workflow.run('SGE', plugin_args={'qsub_args': '-q many'})\n workflow.run('LSF', plugin_args={'qsub_args': '-q many'})\n workflow.run('Condor')\n workflow.run('IPython')\n \n workflow.run('PBSGraph', plugin_args={'qsub_args': '-q many'})\n workflow.run('SGEGraph', plugin_args={'qsub_args': '-q many'})\n workflow.run('CondorDAGMan')\n\nFor a complete list and explanation of all supported plugins, see: http://nipype.readthedocs.io/en/latest/users/plugins.html", "_____no_output_____" ] ] ]

[ "markdown" ]

[ [ "markdown", "markdown", "markdown", "markdown" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "#default_exp maml\n#export\nimport logging\nimport warnings \n\nimport higher\nimport kornia as K\nimport wandb\nimport pytorch_lightning as pl\nimport torch\nimport torchvision\nimport torch.nn as nn\nimport torch.nn.functional as F\nimport torchmetrics\nimport numpy as np\nimport matplotlib.pyplot as plt\n\nfrom copy import deepcopy\nfrom pytorch_lightning import Trainer\nfrom pytorch_lightning.callbacks import LearningRateMonitor, ModelCheckpoint\nfrom pytorch_lightning.loggers import WandbLogger, TensorBoardLogger\nfrom pytorch_lightning.metrics.functional import accuracy\nfrom torchmeta.datasets.helpers import omniglot\nfrom torchmeta.utils.data import BatchMetaDataLoader\nfrom unsupervised_meta_learning.pl_dataloaders import OmniglotDataModule", "_____no_output_____" ], [ "%matplotlib inline", "_____no_output_____" ], [ "#export\nlogger = logging.getLogger(__name__)", "_____no_output_____" ], [ "#export\nclass ConvolutionalNeuralNetwork(nn.Module):\n def __init__(self, in_channels, out_features, hidden_size=64):\n super(ConvolutionalNeuralNetwork, self).__init__()\n self.in_channels = in_channels\n self.out_features = out_features\n self.hidden_size = hidden_size\n\n self.features = nn.Sequential(\n self.conv3x3(in_channels, hidden_size),\n self.conv3x3(hidden_size, hidden_size),\n self.conv3x3(hidden_size, hidden_size),\n self.conv3x3(hidden_size, hidden_size),\n )\n\n self.classifier = nn.Linear(hidden_size, out_features)\n\n def forward(self, inputs, params=None):\n features = self.features(inputs)\n features = features.view((features.size(0), -1))\n logits = self.classifier(features)\n return logits\n\n def conv3x3(self, in_channels, out_channels, **kwargs):\n return nn.Sequential(\n nn.Conv2d(in_channels, out_channels, kernel_size=3, padding=1, **kwargs),\n nn.BatchNorm2d(out_channels, momentum=1.0, track_running_stats=False),\n nn.ReLU(),\n nn.MaxPool2d(2),\n )", "_____no_output_____" ], [ "#export\ndef get_accuracy(logits, targets):\n \"\"\"Compute the accuracy (after adaptation) of MAML on the test/query points\n Parameters\n ----------\n logits : `torch.FloatTensor` instance\n Outputs/logits of the model on the query points. This tensor has shape\n `(num_examples, num_classes)`.\n targets : `torch.LongTensor` instance\n A tensor containing the targets of the query points. This tensor has \n shape `(num_examples,)`.\n Returns\n -------\n accuracy : `torch.FloatTensor` instance\n Mean accuracy on the query points\n \"\"\"\n _, predictions = torch.max(logits, dim=-1)\n return torch.mean(predictions.eq(targets).float())", "_____no_output_____" ], [ "#export\nclass MAML(pl.LightningModule):\n def __init__(self, model, outer_lr, inner_lr, inner_steps=1):\n super().__init__()\n self.model = model\n self.accuracy = get_accuracy\n self.automatic_optimization = False\n self.inner_steps = inner_steps\n self.outer_lr = outer_lr\n self.inner_lr = inner_lr\n \n def forward(self, x):\n return self.model(x)\n \n def inner_loop(self, fmodel, diffopt, train_input, train_target):\n train_logit = fmodel(train_input)\n inner_loss = F.cross_entropy(train_logit, train_target)\n diffopt.step(inner_loss)\n \n return inner_loss.item()\n \n @torch.enable_grad()\n def meta_learn(self, batch, batch_idx, optimizer_idx=None):\n meta_optimizer, inner_optimizer = self.optimizers()\n meta_optimizer = meta_optimizer.optimizer\n inner_optimizer = inner_optimizer.optimizer\n \n train_inputs, train_targets = batch['train']\n test_inputs, test_targets = batch['test']\n \n batch_size = train_inputs.shape[0]\n outer_loss = torch.tensor(0., device=self.device)\n acc = torch.tensor(0., device=self.device)\n self.model.zero_grad()\n \n for task_idx, (train_input, train_target, test_input, test_target) in enumerate(\n zip(train_inputs, train_targets, test_inputs, test_targets)\n ):\n# inner_optimizer.zero_grad()\n with higher.innerloop_ctx(self.model, inner_optimizer, copy_initial_weights=False) as (fmodel, diffopt):\n# train_logit = fmodel(train_input)\n# inner_loss = F.cross_entropy(train_logit, train_target)\n\n# diffopt.step(inner_loss)\n for step in range(self.inner_steps):\n self.inner_loop(fmodel, diffopt, train_input, train_target)\n \n test_logit = fmodel(test_input)\n outer_loss += F.cross_entropy(test_logit, test_target)\n \n with torch.no_grad():\n preds = test_logit.softmax(dim=-1)\n acc += self.accuracy(test_logit, test_target)\n \n\n# self.print(self.accuracy(test_logit, test_target))\n \n outer_loss.div_(batch_size)\n acc.div_(batch_size)\n self.log_dict({\n 'outer_loss': outer_loss,\n 'accuracy': acc\n }, prog_bar=True)\n \n meta_optimizer.zero_grad()\n# outer_loss.backward()\n self.manual_backward(outer_loss, meta_optimizer)\n meta_optimizer.step()\n return outer_loss, acc\n \n \n def training_step(self, batch, batch_idx, optimizer_idx):\n train_loss, acc = self.meta_learn(batch, batch_idx, optimizer_idx)\n \n self.log_dict({\n 'train_loss': train_loss.item(),\n 'train_accuracy': acc.item()\n }, prog_bar=True)\n \n return train_loss.item()\n \n def validation_step(self, batch, batch_idx):\n val_loss, val_acc = self.meta_learn(batch, batch_idx)\n \n self.log_dict({\n 'val_loss': val_loss.item(),\n 'val_accuracy': val_acc.item()\n })\n return val_loss.item()\n \n def test_step(self, batch, batch_idx):\n test_loss, test_acc = self.meta_learn(batch, batch_idx)\n self.log_dict({\n 'test_loss': test_loss.item(),\n 'test_accuracy': test_acc.item()\n })\n return test_loss.item()\n \n \n def configure_optimizers(self):\n meta_optimizer = torch.optim.Adam(self.parameters(), lr=self.outer_lr)\n inner_optimizer = torch.optim.SGD(self.parameters(), lr=self.inner_lr)\n \n return [meta_optimizer, inner_optimizer]", "_____no_output_____" ], [ "#export\nclass UMTRA(pl.LightningModule):\n def __init__(self, model, augmentation, inner_steps, inner_lr, outer_lr):\n super().__init__()\n self.model = model\n self.accuracy = get_accuracy\n self.augmentation = augmentation\n self.inner_steps = inner_steps\n self.inner_lr = inner_lr\n self.outer_lr = outer_lr\n self.automatic_optimization = False\n \n def forward(self, x):\n return self.model(x)\n\n def inner_loop(self, fmodel, diffopt, train_input, train_target):\n train_logit = fmodel(train_input)\n inner_loss = F.cross_entropy(train_logit, train_target)\n diffopt.step(inner_loss)\n \n return inner_loss.item()\n \n @torch.enable_grad()\n def meta_learn(self, batch, batch_idx, inner_copied_optimizer=None, optimizer_idx=None):\n meta_optimizer, inner_optimizer = self.optimizers(use_pl_optimizer=False)\n inner_optimizer = inner_optimizer if inner_copied_optimizer is None else inner_copied_optimizer\n \n train_inputs, train_targets = batch['train']\n test_inputs, test_targets = batch['test']\n \n batch_size = train_inputs.shape[0]\n outer_loss = torch.tensor(0., device=self.device)\n acc = torch.tensor(0., device=self.device)\n self.model.zero_grad()\n \n for task_idx, (train_input, train_target, test_input, test_target) in enumerate(\n zip(train_inputs, train_targets, test_inputs, test_targets)\n ):\n val_input = self.augmentation(train_input).to(self.device)\n val_target = deepcopy(train_target).to(self.device)\n with higher.innerloop_ctx(self.model, inner_optimizer, copy_initial_weights=False) as (fmodel, diffopt):\n for step in range(self.inner_steps):\n self.inner_loop(fmodel, diffopt, train_input, train_target)\n \n val_logits = fmodel(val_input)\n outer_loss += F.cross_entropy(val_logits, val_target)\n\n with torch.no_grad():\n test_logits = fmodel(test_input)\n acc += self.accuracy(test_logits, test_target)\n \n outer_loss.div_(batch_size)\n acc.div_(batch_size)\n \n meta_optimizer.zero_grad()\n# outer_loss.backward()\n\n self.manual_backward(outer_loss, meta_optimizer)\n meta_optimizer.step()\n \n return outer_loss, acc\n \n def training_step(self, batch, batch_idx, optimizer_idx):\n train_loss, acc = self.meta_learn(batch, batch_idx, optimizer_idx=optimizer_idx)\n \n self.log_dict({\n 'train_loss': train_loss.item(),\n 'train_accuracy': acc.item()\n }, prog_bar=True)\n \n return train_loss.item()\n \n def validation_step(self, batch, batch_idx):\n val_loss, val_acc = self.meta_learn(batch, batch_idx)\n self.log_dict({\n 'val_loss': val_loss.item(),\n 'val_accuracy': val_acc.item()\n })\n return val_loss.item()\n \n def test_step(self, batch, batch_idx):\n self.model.train()\n test_loss, test_acc = self.meta_learn(batch, batch_idx)\n \n self.log_dict({\n 'test_loss': test_loss.item(),\n 'test_accuracy': test_acc.item()\n })\n return test_loss.item()\n \n \n def configure_optimizers(self):\n meta_optimizer = torch.optim.Adam(self.parameters(), lr=self.outer_lr)\n inner_optimizer = torch.optim.SGD(self.parameters(), lr=self.inner_lr)\n \n return [meta_optimizer, inner_optimizer]", "_____no_output_____" ], [ "dm = OmniglotDataModule(\n \"data\",\n shots=1,\n ways=5,\n shuffle_ds=True,\n test_shots=15,\n meta_train=True,\n download=True,\n batch_size=16,\n shuffle=True,\n num_workers=8,\n)", "_____no_output_____" ], [ "model = MAML(model=ConvolutionalNeuralNetwork(1, 5, hidden_size=64), outer_lr=3e-3, inner_lr=5e-1, inner_steps=1)", "_____no_output_____" ], [ "logger = WandbLogger(\n project='maml',\n config={\n 'batch_size': 16,\n 'steps': 100,\n 'dataset': \"omniglot\",\n 'inner_steps': 1,\n 'val/test': 'enabled'\n }\n)\ntrainer = Trainer(\n profiler='simple',\n max_epochs=100,\n max_steps=100,\n limit_train_batches=100,\n limit_val_batches=0,\n limit_test_batches=2,\n fast_dev_run=False,\n gpus=1,\n log_every_n_steps=1,\n flush_logs_every_n_steps=1,\n num_sanity_val_steps=2,\n logger=logger\n )", "GPU available: True, used: True\nTPU available: False, using: 0 TPU cores\n" ], [ "with warnings.catch_warnings():\n warnings.simplefilter(\"ignore\")\n trainer.fit(model, datamodule=dm)", "LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]\n\u001b[34m\u001b[1mwandb\u001b[0m: wandb version 0.10.32 is available! To upgrade, please run:\n\u001b[34m\u001b[1mwandb\u001b[0m: $ pip install wandb --upgrade\n" ], [ "wandb.finish()", "_____no_output_____" ], [ "aug = nn.Sequential(\n K.augmentation.RandomAffine(degrees=0, translate=(0.4, 0.4), padding_mode='border'),\n K.augmentation.RandomGaussianNoise(mean=0., std=.1, p=.3)\n)\nmodel = UMTRA(model=ConvolutionalNeuralNetwork(1, 5, hidden_size=64), augmentation=aug, inner_steps=1, outer_lr=3e-3, inner_lr=5e-1)", "_____no_output_____" ], [ "logger = WandbLogger(\n project='umtra',\n config={\n 'batch_size': 16,\n 'steps': 100,\n 'dataset': \"omniglot\",\n 'inner_steps': 5\n }\n)\ntrainer = Trainer(\n profiler='simple',\n max_epochs=100,\n max_steps=100,\n limit_train_batches=50,\n limit_val_batches=0.,\n limit_test_batches=2,\n fast_dev_run=False,\n gpus=1,\n log_every_n_steps=1,\n flush_logs_every_n_steps=1,\n num_sanity_val_steps=2,\n logger=logger\n )", "GPU available: True, used: True\nTPU available: False, using: 0 TPU cores\n" ], [ "with warnings.catch_warnings():\n warnings.simplefilter(\"ignore\")\n trainer.fit(model, datamodule=dm)", "LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]\n\u001b[34m\u001b[1mwandb\u001b[0m: wandb version 0.10.32 is available! To upgrade, please run:\n\u001b[34m\u001b[1mwandb\u001b[0m: $ pip install wandb --upgrade\n" ], [ "wandb.finish()", "_____no_output_____" ], [ "from nbdev.export import notebook2script; notebook2script()", "Converted 01_nn_utils.ipynb.\nConverted 01b_data_loaders_pl.ipynb.\nConverted 01c_grad_utils.ipynb.\nConverted 01d_hessian_free.ipynb.\nConverted 02_maml_pl.ipynb.\nConverted 02b_iMAML.ipynb.\nConverted 03_protonet_pl.ipynb.\nConverted 04_cactus.ipynb.\nConverted index.ipynb.\n" ] ] ]

[ "code" ]

[ [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "# Simulating Clifford randomized benchmarking using implicit models\n\nThis tutorial demonstrates shows how to simulate Clifford RB sequences using $n$-qubit \"implicit\" models which build $n$-qubit process matrices from smaller building blocks. This restricts the noise allowed in the $n$-qubit model; in this tutorial we take $n=3$ and use a `LocalNoiseModel`.", "_____no_output_____" ] ], [ [ "import pygsti\nimport numpy as np", "_____no_output_____" ] ], [ [ "## Get some CRB circuits\n\nFirst, we follow the [Clifford RB](../CliffordRB.ipynb) tutorial to generate a set of sequences. If you want to perform Direct RB instead, just replace this cell with the contents of the [Direct RB](../DirectRB.ipynb) tutorial up until the point where it creates `circuitlist`:", "_____no_output_____" ] ], [ [ "#Specify the device to be benchmarked - in this case 2 qubits\nnQubits = 3\nqubit_labels = list(range(nQubits)) \ngate_names = ['Gxpi2', 'Gypi2','Gcphase'] \navailability = {'Gcphase':[(i,i+1) for i in range(nQubits-1)]}\npspec = pygsti.obj.ProcessorSpec(nQubits, gate_names, availability=availability, \n qubit_labels=qubit_labels)\n\n#Specify RB parameters (k = number of repetitions at each length)\nlengths = [0,1,2,4,8,16]\nk = 10\nsubsetQs = qubit_labels\nrandomizeout = False # ==> all circuits have the *same* ideal outcome (the all-zeros bitstring)\n\n#Generate clifford RB circuits\nexp_design = pygsti.protocols.CliffordRBDesign(pspec, lengths, k, qubit_labels=subsetQs, randomizeout=randomizeout)\n\n#Collect all the circuits into one list:\ncircuitlist = exp_design.all_circuits_needing_data", "_____no_output_____" ] ], [ [ "## Create a model to simulate these circuits\nNow we need to create a model that can simulate circuits like this. The RB circuits use pyGSTi's \"multi-qubit\" conventions, which mean:\n1. RB circuits use our \"multi-qubit\" gate naming, so you have gates like `Gxpi2:0` and `Gcphase:0:1`.\n2. RB circuits do gates in parallel (this only matters for >1 qubits), so you have layers like `[Gypi2:0Gypi2:1]`\n\n\"Implicit\" models in pyGSTi (see the [implicit model tutorial](../../objects/ImplicitModel.ipynb)) are designed to efficiently describe multi-qubit processors. There are numerous ways of constructing implicit models, all of which can simulate the type of circuits described above. Here we'll demonstrate the simplest type: a \"local noise model\" (class `LocalNoiseModel`) where the noise on a gate can only act on that gate's target qubits - so, for instance, 1-qubit gates are still given by 1-qubit operators, not $n$-qubit ones.\n\nThe construction of a local noise model follows the same pattern as building the `ProcessorSpec` above (in fact, `pspec.models['target']` *is* essentially the same model we build below except it was built with the default `parmeterization=\"static\"` argument.", "_____no_output_____" ] ], [ [ "myModel = pygsti.obj.LocalNoiseModel.build_from_parameterization(nQubits, gate_names,\n availability=availability, \n qubit_labels=qubit_labels,\n parameterization=\"full\")", "_____no_output_____" ] ], [ [ "Setting `parameterization=\"full\"` is important, as it lets us assign arbitrary numpy arrays to gates as we'll show below. If you need to use other gates that aren't built into pyGSTi, you can use the `nonstd_gate_unitaries`\nargument of `build_from_parameterization` (see the docstring).\n\nThe `build_from_parameterization` function creates a model with ideal (perfect) gates. We'll now create a 1-qubit depolarization superoperator, and a corresponding 2-qubit one (just the tensor product of two 1-qubit ones) to add some simple noise. ", "_____no_output_____" ] ], [ [ "depol1Q = np.array([[1, 0, 0, 0],\n [0, 0.99, 0, 0],\n [0, 0, 0.99, 0],\n [0, 0, 0, 0.99]], 'd') # 1-qubit depolarizing operator\ndepol2Q = np.kron(depol1Q,depol1Q)", "_____no_output_____" ] ], [ [ "As detailed in the [implicit model tutorial](../../objects/ImplicitModel.ipynb), the gate operations of a `LocalNoiseModel` are held in its `.operation_blks['gates']` dictionary. We'll alter these by assigning new process matrices to each gate. In this case, it will be just a depolarized version of the original gate.", "_____no_output_____" ] ], [ [ "myModel.operation_blks['gates'][\"Gxpi2\"] = np.dot(depol1Q, myModel.operation_blks['gates'][\"Gxpi2\"])\nmyModel.operation_blks['gates'][\"Gypi2\"] = np.dot(depol1Q, myModel.operation_blks['gates'][\"Gypi2\"]) \nmyModel.operation_blks['gates'][\"Gcphase\"] = np.dot(depol2Q, myModel.operation_blks['gates'][\"Gcphase\"])", "_____no_output_____" ] ], [ [ "Here's what the gates look like now:", "_____no_output_____" ] ], [ [ "print(myModel.operation_blks['gates'][\"Gxpi2\"])\nprint(myModel.operation_blks['gates'][\"Gypi2\"])\nprint(myModel.operation_blks['gates'][\"Gcphase\"])", "_____no_output_____" ] ], [ [ "Now that our `Model` object is set to go, generating simulated data is easy:", "_____no_output_____" ] ], [ [ "ds = pygsti.construction.generate_fake_data(myModel, circuitlist, 100, seed=1234)", "_____no_output_____" ] ], [ [ "## Running RB on the simulated `DataSet`\nTo run an RB analysis, we just package up the experiment design and data set into a `ProtocolData` object and give this to a `RB` protocol's `run` method. This returns a `RandomizedBenchmarkingResults` object that can be used to plot the RB decay curve. (See the [RB analysis tutorial](../RBAnalysis.ipynb) for more details.)", "_____no_output_____" ] ], [ [ "data = pygsti.protocols.ProtocolData(exp_design, ds)\nresults = pygsti.protocols.RB().run(data)", "_____no_output_____" ], [ "%matplotlib inline\nresults.plot()", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "<a href=\"https://colab.research.google.com/github/nghoanglong/rat-sql/blob/master/ratsql_colab.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>", "_____no_output_____" ] ], [ [ "gpu_info = !nvidia-smi\ngpu_info = '\\n'.join(gpu_info)\nif gpu_info.find('failed') >= 0:\n print('Not connected to a GPU')\nelse:\n print(gpu_info)", "_____no_output_____" ] ], [ [ "# Set up and Install requirements", "_____no_output_____" ] ], [ [ "from google.colab import drive\ndrive.mount('/content/drive')", "_____no_output_____" ], [ "!git clone https://github.com/nghoanglong/rat-sql.git", "_____no_output_____" ], [ "%cd /content/rat-sql", "_____no_output_____" ], [ "!pip install -r requirements.txt", "_____no_output_____" ], [ "import nltk\nnltk.download('stopwords')\nnltk.download('punkt')", "_____no_output_____" ], [ "from transformers import BertModel\nBertModel.from_pretrained('bert-large-uncased-whole-word-masking')", "_____no_output_____" ], [ "!mkdir -p third_party", "_____no_output_____" ], [ "!git clone https://github.com/salesforce/WikiSQL third_party/wikisql", "_____no_output_____" ], [ "%cd /content/rat-sql", "/content/rat-sql\n" ] ], [ [ "# Run Spider", "_____no_output_____" ], [ "## Spider - Glove", "_____no_output_____" ] ], [ [ "!python run.py preprocess /content/rat-sql/experiments/spider-glove-run.jsonnet", "_____no_output_____" ], [ "!python run.py train /content/rat-sql/experiments/spider-glove-run.jsonnet", "_____no_output_____" ], [ "!python run.py eval /content/rat-sql/experiments/spider-glove-run.jsonnet", "_____no_output_____" ] ], [ [ "## Spider - Bert", "_____no_output_____" ] ], [ [ "!python run.py preprocess /content/rat-sql/experiments/spider-bert-run.jsonnet", "\rDownloading https://huggingface.co/stanfordnlp/CoreNLP/resolve/main/stanford-corenlp-latest.zip: 100% 505M/505M [00:05<00:00, 100MB/s] \n2021-10-28 09:14:19 WARNING: For customized installation location, please set the `CORENLP_HOME` environment variable to the location of the installation. In Unix, this is done with `export CORENLP_HOME=/content/rat-sql/third_party/stanford-corenlp-full-2018-10-05`.\n2021-10-28 09:14:19 INFO: Writing properties to tmp file: corenlp_server-5998d88a11be49f6.props\nclient: <stanza.server.client.CoreNLPClient object at 0x7fce244cce50>\n2021-10-28 09:14:19 INFO: Starting server with command: java -Xmx4G -cp /content/rat-sql/third_party/stanford-corenlp-full-2018-10-05/* edu.stanford.nlp.pipeline.StanfordCoreNLPServer -port 9001 -timeout 60000 -threads 5 -maxCharLength 100000 -quiet True -serverProperties corenlp_server-5998d88a11be49f6.props -annotators tokenize,ssplit,pos,lemma,ner -preload -outputFormat serialized\ntrain section: 100% 7000/7000 [31:16<00:00, 3.73it/s]\nDB connections: 100% 166/166 [00:01<00:00, 129.38it/s]\nval section: 100% 1034/1034 [11:45<00:00, 1.47it/s]\n" ], [ "!python run.py train /content/rat-sql/experiments/spider-bert-run.jsonnet", "[2021-10-28T09:58:21] Logging to /content/drive/MyDrive/Datasets/ratsql/datasets/spider/logdir/bs=3,lr=7.4e-04,bert_lr=3.0e-06,end_lr=0e0,att=1\nSome weights of the model checkpoint at bert-large-uncased-whole-word-masking were not used when initializing BertModel: ['cls.predictions.transform.LayerNorm.bias', 'cls.predictions.transform.dense.weight', 'cls.predictions.decoder.weight', 'cls.predictions.transform.dense.bias', 'cls.predictions.bias', 'cls.predictions.transform.LayerNorm.weight', 'cls.seq_relationship.weight', 'cls.seq_relationship.bias']\n- This IS expected if you are initializing BertModel from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).\n- This IS NOT expected if you are initializing BertModel from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).\n[2021-10-28T09:58:45] Step 0 stats, train: loss = 160.08171844482422\n[2021-10-28T09:58:57] Step 0 stats, val: loss = 190.71556091308594\n[2021-10-28T09:59:02] Step 0: loss=194.38427734375\n[2021-10-28T10:00:35] Step 10: loss=166.95787048339844\n[2021-10-28T10:01:25] Step 20: loss=230.46824645996094\nTraceback (most recent call last):\n File \"/content/rat-sql/ratsql/commands/train.py\", line 208, in train\n loss = self.model.compute_loss(batch)\n File \"/content/rat-sql/ratsql/models/enc_dec.py\", line 76, in _compute_loss_enc_batched\n enc_states = self.encoder([enc_input for enc_input, dec_output in batch])\n File \"/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py\", line 550, in __call__\n result = self.forward(*input, **kwargs)\n File \"/content/rat-sql/ratsql/models/spider/spider_enc.py\", line 1037, in forward\n attention_mask=att_masks_tensor, token_type_ids=tok_type_tensor)[0]\n File \"/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py\", line 550, in __call__\n result = self.forward(*input, **kwargs)\n File \"/usr/local/lib/python3.7/dist-packages/transformers/models/bert/modeling_bert.py\", line 1005, in forward\n return_dict=return_dict,\n File \"/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py\", line 550, in __call__\n result = self.forward(*input, **kwargs)\n File \"/usr/local/lib/python3.7/dist-packages/transformers/models/bert/modeling_bert.py\", line 589, in forward\n output_attentions,\n File \"/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py\", line 550, in __call__\n result = self.forward(*input, **kwargs)\n File \"/usr/local/lib/python3.7/dist-packages/transformers/models/bert/modeling_bert.py\", line 475, in forward\n past_key_value=self_attn_past_key_value,\n File \"/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py\", line 550, in __call__\n result = self.forward(*input, **kwargs)\n File \"/usr/local/lib/python3.7/dist-packages/transformers/models/bert/modeling_bert.py\", line 410, in forward\n attention_output = self.output(self_outputs[0], hidden_states)\n File \"/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py\", line 550, in __call__\n result = self.forward(*input, **kwargs)\n File \"/usr/local/lib/python3.7/dist-packages/transformers/models/bert/modeling_bert.py\", line 361, in forward\n hidden_states = self.dropout(hidden_states)\n File \"/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py\", line 550, in __call__\n result = self.forward(*input, **kwargs)\n File \"/usr/local/lib/python3.7/dist-packages/torch/nn/modules/dropout.py\", line 54, in forward\n return F.dropout(input, self.p, self.training, self.inplace)\n File \"/usr/local/lib/python3.7/dist-packages/torch/nn/functional.py\", line 936, in dropout\n else _VF.dropout(input, p, training))\nKeyboardInterrupt\n\nDuring handling of the above exception, another exception occurred:\n\nTraceback (most recent call last):\n File \"run.py\", line 109, in <module>\n main()\n File \"run.py\", line 77, in main\n train.main(train_config)\n File \"/content/rat-sql/ratsql/commands/train.py\", line 296, in main\n trainer.train(config, modeldir=args.logdir)\n File \"/content/rat-sql/ratsql/commands/train.py\", line 208, in train\n loss = self.model.compute_loss(batch)\nKeyboardInterrupt\n" ], [ "!python run.py eval /content/rat-sql/experiments/spider-bert-run.jsonnet", "_____no_output_____" ] ], [ [ "# Run vitext2sql", "_____no_output_____" ] ], [ [ "!wget -P /content/rat-sql/third_party/phow2v_emb https://public.vinai.io/word2vec_vi_words_300dims.zip", "_____no_output_____" ], [ "cd /content/rat-sql/third_party/phow2v_emb", "/content/rat-sql/third_party/phow2v_emb\n" ], [ "!unzip /content/rat-sql/third_party/phow2v_emb/word2vec_vi_words_300dims.zip", "Archive: /content/rat-sql/third_party/phow2v_emb/word2vec_vi_words_300dims.zip\n inflating: word2vec_vi_words_300dims.txt \n" ], [ "cd /content/rat-sql", "/content/rat-sql\n" ] ], [ [ "## Run Vitext2sql - No PhoBert", "_____no_output_____" ] ], [ [ "!python run.py preprocess /content/rat-sql/experiments/vitext2sql-phow2v-run.jsonnet", "_____no_output_____" ], [ "!python run.py train /content/rat-sql/experiments/vitext2sql-phow2v-run.jsonnet", "_____no_output_____" ], [ "!python run.py eval /content/rat-sql/experiments/vitext2sql-phow2v-run.jsonnet", "_____no_output_____" ] ], [ [ "## Run Vitext2SQL - PhoBert", "_____no_output_____" ] ], [ [ "!python run.py preprocess /content/rat-sql/experiments/vitext2sql-phobert-run.jsonnet", "_____no_output_____" ], [ "!python run.py train /content/rat-sql/experiments/vitext2sql-phobert-run.jsonnet", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code", "code", "code" ], [ "markdown", "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "import os\nimport glob\nimport time\nimport json\nimport numpy as np\nimport theano as th\nimport theano.tensor as tt\nimport theano.tensor.slinalg as sla\nimport bmtools.exact.moments as mom\nimport bmtools.relaxations.gm_relaxations as gmr\nimport bmtools.relaxations.var_mixture as var\nimport bmtools.utils as utils\nimport matplotlib.pyplot as plt\nimport thermomc.continuous_temp as cont_temp\nimport thermomc.discrete_temp as disc_temp\nimport thermomc.control_funcs as ctrl\nimport seaborn as sns\nsns.set_style('whitegrid')\n%matplotlib inline", "_____no_output_____" ] ], [ [ "## Load model parameters and set up", "_____no_output_____" ] ], [ [ "base_dir = os.path.dirname(os.getcwd())\nmodel_dir = os.path.join(base_dir, 'data', 'gaussian-bmr')\nexp_dir = os.path.join(base_dir, 'experiments', 'gaussian-bmr')\nif not os.path.exists(exp_dir):\n os.makedirs(exp_dir)", "_____no_output_____" ], [ "seed = 201702\nrng = np.random.RandomState(seed)", "_____no_output_____" ], [ "class PhiFunc(object):\n \n def __init__(self, Q, b, log_zeta):\n self.Q = Q\n self.b = b\n self.log_zeta = log_zeta\n \n def __call__(self, x):\n return (\n 0.5 * (x**2).sum(-1) - \n tt.log(tt.cosh(x.dot(self.Q.T) + self.b)).sum(-1) + self.log_zeta\n )\n\nclass PsiFunc(object):\n\n def __init__(self, L, m):\n self.L = L\n self.m = m\n \n def __call__(self, x):\n z = x - self.m\n return 0.5 * (\n (z.T * sla.solve_upper_triangular(\n self.L.T, sla.solve_lower_triangular(\n self.L, z.T))).sum(0) +\n self.m.shape[0] * tt.log(2 * np.pi) + \n 2 * tt.log(self.L.diagonal()).sum()\n )", "_____no_output_____" ], [ "def sigmoid(x):\n return 1. / (1. + np.exp(-x))\n\ndef sigmoidal_schedule(num_temp, scale):\n inv_temp_sched = sigmoid(\n scale * (2. * np.arange(num_temp + 1) / num_temp - 1.))\n return (\n (inv_temp_sched - inv_temp_sched[0]) / \n (inv_temp_sched[-1] - inv_temp_sched[0])\n )\n\ndef rmse(x, y):\n return ((x - y)**2).mean()**0.5", "_____no_output_____" ], [ "dtype = 'float64'\nrelaxation_list = []\ntrue_log_norm_list = []\ntrue_mean_list = []\ntrue_covar_list = []\nvar_log_norm_list = []\nvar_mean_list = []\nvar_covar_chol_list = []\nphi_funcs = []\npsi_funcs = []\nfor i, file_path in enumerate(\n sorted(glob.glob(os.path.join(model_dir, 'params_and_moms_*.npz')))):\n loaded = np.load(file_path)\n relaxation_list.append(gmr.IsotropicCovarianceGMRelaxation(\n loaded['weights'], loaded['biases'], True)\n )\n true_log_norm_list.append(loaded['log_norm_const_x'])\n true_mean_list.append(loaded['expc_x'])\n true_covar_list.append(loaded['covar_x'])\n var_mean, var_covar_chol, var_log_norm = (\n var.mixture_of_variational_distributions_moments(\n relaxation_list[-1], rng\n ))\n var_log_norm_list.append(var_log_norm)\n var_mean_list.append(var_mean)\n var_covar_chol_list.append(var_covar_chol)\n np.savez(os.path.join(model_dir, 'var_moms_{0}.npz'.format(i)), \n var_mean=var_mean, var_covar_chol=var_covar_chol, var_log_norm=var_log_norm)\n Q = tt.constant(relaxation_list[-1].Q, 'Q' + str(i), 2, dtype)\n b = tt.constant(relaxation_list[-1].b, 'b' + str(i), 1, dtype)\n L = tt.constant(var_covar_chol, 'L' + str(i), 2, dtype)\n m = tt.constant(var_mean, 'm' + str(i), 1, dtype)\n log_zeta = tt.constant(var_log_norm, 'log_zeta' + str(i), 0, dtype)\n phi_funcs.append(PhiFunc(Q, b, log_zeta))\n psi_funcs.append(PsiFunc(L, m))\n print('Var. log norm RMSE: {0}'.format(rmse(var_log_norm, true_log_norm_list[-1])))", "_____no_output_____" ] ], [ [ "## Annealed Importance Sampling", "_____no_output_____" ] ], [ [ "num_temps = [1000, 5000, 10000, 20000]\ndt = 0.5\ntemp_scale = 4.\nnum_reps = 10\nnum_step = 10\nnum_runs_per_rep = 100\nmom_resample_coeff = 1.\nnum_runs = num_reps * num_runs_per_rep", "_____no_output_____" ], [ "pos = tt.matrix('pos')\ninv_temps = tt.vector('inv_temps')\nhmc_params = {\n 'dt': dt,\n 'n_step': num_step,\n 'mom_resample_coeff': mom_resample_coeff\n}\nais_sampler = disc_temp.AnnealedImportanceSampler(\n tt.shared_randomstreams.RandomStreams(seed), False\n)\nais_run_funcs = []\nfor phi_func, psi_func in zip(phi_funcs, psi_funcs):\n pos_samples, log_weights, accepts, updates = ais_sampler.run(\n pos, None, inv_temps, phi_func, psi_func, hmc_params\n )\n ais_run = th.function(\n [pos, inv_temps],\n [pos_samples, log_weights, accepts],\n updates=updates\n )\n ais_run_funcs.append(ais_run)", "_____no_output_____" ], [ "for i in range(10):\n ais_exp_dir = os.path.join(exp_dir, 'ais', 'params-' + str(i))\n if not os.path.exists(ais_exp_dir):\n os.makedirs(ais_exp_dir)\n for num_temp in num_temps:\n settings = {\n 'dt': dt,\n 'num_temp': num_temp,\n 'temp_scale': temp_scale,\n 'n_step': num_step,\n 'mom_resample_coeff': mom_resample_coeff\n }\n print('Parameters {0} num temps {1}'.format(i, num_temp))\n print('-' * 100)\n print(settings)\n settings_path = os.path.join(ais_exp_dir, 'settings-{0}.json'.format(num_temp))\n results_path = os.path.join(ais_exp_dir, 'results-{0}.npz'.format(num_temp))\n with open(settings_path, 'w') as f:\n json.dump(settings, f, indent=True)\n inv_temp_sched = sigmoidal_schedule(num_temp, temp_scale)\n num_dim = relaxation_list[i].n_dim_r\n pos_init = rng.normal(size=(num_runs, num_dim)).dot(\n var_covar_chol_list[i].T) + var_mean_list[i]\n start_time = time.time()\n pos_samples, log_weights, accepts = ais_run_funcs[i](\n pos_init, inv_temp_sched\n )\n sampling_time = time.time() - start_time\n print('Sampling time: {0:.2f}s'.format(sampling_time))\n log_norm_rmses = []\n mean_rmses = []\n covar_rmses = []\n for lw, ps in zip(\n log_weights.reshape((num_reps, -1)),\n pos_samples.reshape((num_reps, num_runs_per_rep, -1))):\n log_norm_rmses.append(\n rmse(np.log(np.exp(lw).mean(0)) + var_log_norm_list[i], \n true_log_norm_list[i])\n )\n probs = np.exp(lw)\n probs /= probs.sum()\n mean_est = (probs[:, None] * ps).sum(0)\n mean_rmses.append(\n rmse(true_mean_list[i], mean_est)\n )\n ps_zm = ps - mean_est\n covar_est = (ps_zm * probs[:, None]).T.dot(ps_zm)\n covar_rmses.append(\n rmse(true_covar_list[i], covar_est)\n )\n var_log_norm_rmse = rmse(true_log_norm_list[i], var_log_norm_list[i])\n var_mean_rmse = rmse(true_mean_list[i], var_mean_list[i])\n var_covar_rmse = rmse(true_covar_list[i], \n var_covar_chol_list[i].dot(var_covar_chol_list[i].T))\n print('RMSE log_norm={0:.2f} mean={1:.2f} covar={2:.2f}'\n .format(\n np.mean(log_norm_rmses) / var_log_norm_rmse, \n np.mean(mean_rmses) / var_mean_rmse, \n np.mean(covar_rmses) / var_covar_rmse\n )\n )\n np.savez(\n results_path, \n sampling_time=sampling_time, \n pos_samples=pos_samples, \n log_weights=log_weights, \n accepts=accepts,\n log_norm_rmses=np.array(log_norm_rmses),\n mean_rmses=np.array(mean_rmses),\n covar_rmses=np.array(covar_rmses),\n var_log_norm_rmse=var_log_norm_rmse,\n var_mean_rmse=var_mean_rmse,\n var_covar_rmse=var_covar_rmse\n )\n print('Saved to ' + results_path)", "_____no_output_____" ] ], [ [ "## Hamiltonian Annealed Importance Sampling", "_____no_output_____" ] ], [ [ "num_temps = [1000, 5000, 10000, 20000]\ndt = 0.5\ntemp_scale = 4.\nnum_reps = 10\nnum_step = 1\nnum_runs_per_rep = 500\nnum_runs = num_reps * num_runs_per_rep", "_____no_output_____" ], [ "pos = tt.matrix('pos')\ninv_temps = tt.vector('inv_temps')\nhmc_params = {\n 'dt': dt,\n 'n_step': num_step,\n 'mom_resample_coeff': (1. - 0.5**dt)**0.5\n}\nais_sampler = disc_temp.AnnealedImportanceSampler(\n tt.shared_randomstreams.RandomStreams(seed), True\n)\nais_run_funcs = []\nfor phi_func, psi_func in zip(phi_funcs, psi_funcs):\n pos_samples, log_weights, accepts, updates = ais_sampler.run(\n pos, None, inv_temps, phi_func, psi_func, hmc_params\n )\n ais_run = th.function(\n [pos, inv_temps],\n [pos_samples, log_weights, accepts],\n updates=updates\n )\n ais_run_funcs.append(ais_run)", "_____no_output_____" ], [ "for i in range(10):\n ais_exp_dir = os.path.join(exp_dir, 'h-ais', 'params-' + str(i))\n if not os.path.exists(ais_exp_dir):\n os.makedirs(ais_exp_dir)\n for num_temp in num_temps:\n settings = {\n 'dt': dt,\n 'num_temp': num_temp,\n 'temp_scale': temp_scale,\n 'n_step': num_step,\n 'mom_resample_coeff': (1. - 0.5**dt)**0.5\n }\n print('Parameters {0} num temp {1}'.format(i, num_temp))\n print('-' * 100)\n print(settings)\n settings_path = os.path.join(ais_exp_dir, 'settings-{0}.json'.format(num_temp))\n results_path = os.path.join(ais_exp_dir, 'results-{0}.npz'.format(num_temp))\n with open(settings_path, 'w') as f:\n json.dump(settings, f, indent=True)\n inv_temp_sched = sigmoidal_schedule(num_temp, temp_scale)\n num_dim = relaxation_list[i].n_dim_r\n pos_init = rng.normal(size=(num_runs, num_dim)).dot(\n var_covar_chol_list[i].T) + var_mean_list[i]\n start_time = time.time()\n pos_samples, log_weights, accepts = ais_run_funcs[i](\n pos_init, inv_temp_sched\n )\n sampling_time = time.time() - start_time\n print('Sampling time: {0:.2f}s'.format(sampling_time))\n log_norm_rmses = []\n mean_rmses = []\n covar_rmses = []\n for lw, ps in zip(\n log_weights.reshape((num_reps, -1)),\n pos_samples.reshape((num_reps, num_runs_per_rep, -1))):\n log_norm_rmses.append(\n rmse(np.log(np.exp(lw).mean(0)) + var_log_norm_list[i], \n true_log_norm_list[i])\n )\n probs = np.exp(lw)\n probs /= probs.sum()\n mean_est = (probs[:, None] * ps).sum(0)\n mean_rmses.append(\n rmse(true_mean_list[i], mean_est)\n )\n ps_zm = ps - mean_est\n covar_est = (ps_zm * probs[:, None]).T.dot(ps_zm)\n covar_rmses.append(\n rmse(true_covar_list[i], covar_est)\n )\n var_log_norm_rmse = rmse(true_log_norm_list[i], var_log_norm_list[i])\n var_mean_rmse = rmse(true_mean_list[i], var_mean_list[i])\n var_covar_rmse = rmse(true_covar_list[i], \n var_covar_chol_list[i].dot(var_covar_chol_list[i].T))\n print('RMSE log_norm={0:.2f} mean={1:.2f} covar={2:.2f}'\n .format(\n np.mean(log_norm_rmses) / var_log_norm_rmse, \n np.mean(mean_rmses) / var_mean_rmse, \n np.mean(covar_rmses) / var_covar_rmse\n )\n )\n np.savez(\n results_path, \n sampling_time=sampling_time, \n pos_samples=pos_samples, \n log_weights=log_weights, \n accepts=accepts,\n log_norm_rmses=np.array(log_norm_rmses),\n mean_rmses=np.array(mean_rmses),\n covar_rmses=np.array(covar_rmses),\n var_log_norm_rmse=var_log_norm_rmse,\n var_mean_rmse=var_mean_rmse,\n var_covar_rmse=var_covar_rmse\n )\n print('Saved to ' + results_path)\n print('-' * 100)", "_____no_output_____" ] ], [ [ "## Incremental RMSE helper", "_____no_output_____" ] ], [ [ "def rmse(x, y):\n return ((x - y)**2).mean()**0.5\n\ndef calculate_incremental_rmses(x_samples, probs_1, probs_0,\n true_log_norm, true_mean, true_covar):\n n_sample, n_chain, n_dim = x_samples.shape\n sum_probs_1_x = 0\n sum_probs_1_xx = 0\n sum_probs_1 = 0\n sum_probs_0 = 0\n log_norm_rmses = np.empty(n_sample) * np.nan\n mean_rmses = np.empty(n_sample) * np.nan\n covar_rmses = np.empty(n_sample) * np.nan\n for s in range(n_sample):\n p1 = probs_1[s]\n p0 = probs_0[s]\n x = x_samples[s]\n sum_probs_1_x += p1[:, None] * x\n sum_probs_1_xx += p1[:, None, None] * (x[:, :, None] * x[:, None, :])\n sum_probs_1 += p1\n sum_probs_0 += p0\n log_norm_est = np.log(sum_probs_1.sum(0)) - np.log(sum_probs_0.sum(0))\n mean_est = sum_probs_1_x.sum(0) / sum_probs_1.sum(0)\n covar_est = sum_probs_1_xx.sum(0) / sum_probs_1.sum(0) - np.outer(mean_est, mean_est)\n log_norm_rmses[s] = rmse(log_norm_est, true_log_norm)\n mean_rmses[s] = rmse(mean_est, true_mean)\n covar_rmses[s] = rmse(covar_est, true_covar)\n return log_norm_rmses, mean_rmses, covar_rmses", "_____no_output_____" ] ], [ [ "## Simulated Tempering", "_____no_output_____" ] ], [ [ "num_temp = 1000\ndt = 0.5\nnum_step = 20\ntemp_scale = 4.\nnum_reps = 10\nnum_runs_per_rep = 10\nnum_runs = num_reps * num_runs_per_rep\nmom_resample_coeff = 1.", "_____no_output_____" ], [ "pos = tt.matrix('pos')\nidx = tt.lvector('idx')\ninv_temps = tt.vector('inv_temps')\nnum_sample = tt.lscalar('num_sample')\nhmc_params = {\n 'dt': dt,\n 'n_step': num_step,\n 'mom_resample_coeff': mom_resample_coeff\n}\nst_sampler = disc_temp.SimulatedTemperingSampler(\n tt.shared_randomstreams.RandomStreams(seed), False\n)\nst_chain_funcs = []\nfor phi_func, psi_func in zip(phi_funcs, psi_funcs):\n pos_samples, idx_samples, probs_0, probs_1, accepts, updates = st_sampler.chain(\n pos, None, idx, inv_temps, 0, phi_func, psi_func, num_sample, hmc_params\n )\n st_chain = th.function(\n [pos, idx, inv_temps, num_sample],\n [pos_samples, idx_samples, probs_0, probs_1, accepts],\n updates=updates\n )\n st_chain_funcs.append(st_chain)", "_____no_output_____" ], [ "num_sample = 40000\nfor i in range(10):\n st_exp_dir = os.path.join(exp_dir, 'st', 'params-' + str(i))\n if not os.path.exists(st_exp_dir):\n os.makedirs(st_exp_dir)\n settings = {\n 'dt': dt,\n 'num_temp': num_temp,\n 'num_sample': num_sample,\n 'num_step': num_step,\n 'temp_scale': temp_scale,\n 'mom_resample_coeff': mom_resample_coeff\n }\n print('Parameters {0}'.format(i))\n print('-' * 100)\n print(settings)\n settings_path = os.path.join(st_exp_dir, 'settings.json')\n results_path = os.path.join(st_exp_dir, 'results.npz')\n with open(settings_path, 'w') as f:\n json.dump(settings, f, indent=True)\n inv_temp_sched = sigmoidal_schedule(num_temp, temp_scale)\n num_dim = relaxation_list[i].n_dim_r\n pos_init = rng.normal(size=(num_runs, num_dim)).dot(\n var_covar_chol_list[i].T) + var_mean_list[i]\n idx_init = np.zeros(num_runs, 'int64')\n start_time = time.time()\n pos_samples, idx_samples, probs_0, probs_1, accepts = st_chain_funcs[i](\n pos_init, idx_init, inv_temp_sched, num_sample\n )\n sampling_time = time.time() - start_time\n print('Sampling time: {0:.2f}s'.format(sampling_time))\n log_norm_rmses = np.empty((num_reps, num_sample))\n mean_rmses = np.empty((num_reps, num_sample))\n covar_rmses = np.empty((num_reps, num_sample))\n for r in range(num_reps):\n log_norm_rmses[r], mean_rmses[r], covar_rmses[r] = calculate_incremental_rmses(\n pos_samples[:, r:(r+1)*num_runs_per_rep], \n probs_1[:, r:(r+1)*num_runs_per_rep], \n probs_0[:, r:(r+1)*num_runs_per_rep], \n true_log_norm_list[i] - var_log_norm_list[i],\n true_mean_list[i], true_covar_list[i]\n )\n var_log_norm_rmse = rmse(true_log_norm_list[i], var_log_norm_list[i])\n var_mean_rmse = rmse(true_mean_list[i], var_mean_list[i])\n var_covar_rmse = rmse(true_covar_list[i], \n var_covar_chol_list[i].dot(var_covar_chol_list[i].T))\n print('RMSE log_norm={0:.2f} mean={1:.2f} covar={2:.2f}'\n .format(\n np.mean(log_norm_rmses[:, -1]) / var_log_norm_rmse, \n np.mean(mean_rmses[:, -1]) / var_mean_rmse, \n np.mean(covar_rmses[:, -1]) / var_covar_rmse\n )\n )\n fig, axes = plt.subplots(1, 3, figsize=(9, 3))\n axes[0].semilogy(log_norm_rmses.mean(0) / var_log_norm_rmse)\n axes[0].set_title('Log norm RMSE')\n axes[1].semilogy(mean_rmses.mean(0) / var_mean_rmse)\n axes[1].set_title('Mean RMSE')\n axes[2].semilogy(covar_rmses.mean(0) / var_covar_rmse)\n axes[2].set_title('Covariance RMSE')\n plt.show()\n np.savez(\n results_path, \n sampling_time=sampling_time, \n pos_samples=pos_samples,\n idx_samples=idx_samples,\n probs_1=probs_1,\n probs_0=probs_0,\n accepts=accepts,\n log_norm_rmses=log_norm_rmses,\n mean_rmses=mean_rmses,\n covar_rmses=covar_rmses,\n var_log_norm_rmse=var_log_norm_rmse,\n var_mean_rmse=var_mean_rmse,\n var_covar_rmse=var_covar_rmse\n )\n print('Saved to ' + results_path)\n print('-' * 100)", "_____no_output_____" ] ], [ [ "## Continuous tempering", "_____no_output_____" ], [ "### Gibbs", "_____no_output_____" ] ], [ [ "dt = 0.5\nnum_step = 20\nnum_reps = 10\nnum_runs_per_rep = 10\nnum_runs = num_reps * num_runs_per_rep\nmom_resample_coeff = 1.", "_____no_output_____" ], [ "pos = tt.matrix('pos')\nidx = tt.lvector('idx')\ninv_temp = tt.vector('inv_temp')\nnum_sample = tt.lscalar('n_sample')\nhmc_params = {\n 'dt': dt,\n 'n_step': num_step,\n 'mom_resample_coeff': mom_resample_coeff\n}\ngct_sampler = cont_temp.GibbsContinuousTemperingSampler(\n tt.shared_randomstreams.RandomStreams(seed), False\n)\ngct_chain_funcs = []\nfor phi_func, psi_func in zip(phi_funcs, psi_funcs):\n pos_samples, inv_temp_samples, probs_0, probs_1, accepts, updates = gct_sampler.chain(\n pos, None, inv_temp, phi_func, psi_func, num_sample, hmc_params\n )\n gct_chain = th.function(\n [pos, inv_temp, num_sample],\n [pos_samples, inv_temp_samples, probs_0, probs_1, accepts],\n updates=updates\n )\n gct_chain_funcs.append(gct_chain)", "_____no_output_____" ], [ "num_sample = 60000\nfor i in range(10):\n gct_exp_dir = os.path.join(exp_dir, 'gibbs-ct', 'params-' + str(i))\n if not os.path.exists(gct_exp_dir):\n os.makedirs(gct_exp_dir)\n settings = {\n 'dt': dt,\n 'num_sample': num_sample,\n 'num_step': num_step,\n 'mom_resample_coeff': mom_resample_coeff\n }\n print('Parameters {0}'.format(i))\n print('-' * 100)\n print(settings)\n settings_path = os.path.join(gct_exp_dir, 'settings.json')\n results_path = os.path.join(gct_exp_dir, 'results.npz')\n with open(settings_path, 'w') as f:\n json.dump(settings, f, indent=True)\n num_dim = relaxation_list[i].n_dim_r\n pos_init = rng.normal(size=(num_runs, num_dim)).dot(\n var_covar_chol_list[i].T) + var_mean_list[i]\n inv_temp_init = np.zeros(num_runs)\n start_time = time.time()\n pos_samples, inv_temp_samples, probs_0, probs_1, accepts = gct_chain_funcs[i](\n pos_init, inv_temp_init, num_sample\n )\n sampling_time = time.time() - start_time\n print('Sampling time: {0:.2f}s'.format(sampling_time))\n log_norm_rmses = np.empty((num_reps, num_sample))\n mean_rmses = np.empty((num_reps, num_sample))\n covar_rmses = np.empty((num_reps, num_sample))\n for r in range(num_reps):\n log_norm_rmses[r], mean_rmses[r], covar_rmses[r] = calculate_incremental_rmses(\n pos_samples[:, r:(r+1)*num_runs_per_rep], \n probs_1[:, r:(r+1)*num_runs_per_rep], \n probs_0[:, r:(r+1)*num_runs_per_rep], \n true_log_norm_list[i] - var_log_norm_list[i],\n true_mean_list[i], true_covar_list[i]\n )\n var_log_norm_rmse = rmse(true_log_norm_list[i], var_log_norm_list[i])\n var_mean_rmse = rmse(true_mean_list[i], var_mean_list[i])\n var_covar_rmse = rmse(true_covar_list[i], \n var_covar_chol_list[i].dot(var_covar_chol_list[i].T))\n print('RMSE log_norm={0:.2f} mean={1:.2f} covar={2:.2f}'\n .format(\n np.mean(log_norm_rmses[:, -1]) / var_log_norm_rmse, \n np.mean(mean_rmses[:, -1]) / var_mean_rmse, \n np.mean(covar_rmses[:, -1]) / var_covar_rmse\n )\n )\n fig, axes = plt.subplots(1, 3, figsize=(9, 3))\n axes[0].semilogy(log_norm_rmses.mean(0) / var_log_norm_rmse)\n axes[0].set_title('Log norm RMSE')\n axes[1].semilogy(mean_rmses.mean(0) / var_mean_rmse)\n axes[1].set_title('Mean RMSE')\n axes[2].semilogy(covar_rmses.mean(0) / var_covar_rmse)\n axes[2].set_title('Covariance RMSE')\n plt.show()\n np.savez(\n results_path, \n sampling_time=sampling_time, \n pos_samples=pos_samples,\n inv_temp_samples=inv_temp_samples,\n probs_1=probs_1,\n probs_0=probs_0,\n accepts=accepts,\n log_norm_rmses=log_norm_rmses,\n mean_rmses=mean_rmses,\n covar_rmses=covar_rmses,\n var_log_norm_rmse=var_log_norm_rmse,\n var_mean_rmse=var_mean_rmse,\n var_covar_rmse=var_covar_rmse\n )\n print('Saved to ' + results_path)\n print('-' * 100)", "_____no_output_____" ] ], [ [ "### Joint", "_____no_output_____" ] ], [ [ "dt = 0.5\nnum_step = 20\ntemp_scale = 1.\nnum_reps = 10\nnum_runs_per_rep = 10\nnum_runs = num_reps * num_runs_per_rep\nmom_resample_coeff = 1.", "_____no_output_____" ], [ "pos = tt.matrix('pos')\ntmp_ctrl = tt.vector('tmp_ctrl')\nnum_sample = tt.lscalar('n_sample')\nctrl_func = ctrl.SigmoidalControlFunction(temp_scale)\nhmc_params = {\n 'dt': dt,\n 'n_step': num_step,\n 'mom_resample_coeff': mom_resample_coeff\n}\njct_sampler = cont_temp.JointContinuousTemperingSampler(\n tt.shared_randomstreams.RandomStreams(seed), False\n)\njct_chain_funcs = []\nfor phi_func, psi_func in zip(phi_funcs, psi_funcs):\n (pos_samples, tmp_ctrl_sample, inv_temp_samples, \n probs_0, probs_1, accepts, updates) = jct_sampler.chain(\n pos, tmp_ctrl, None, phi_func, psi_func, ctrl_func, num_sample, hmc_params\n )\n jct_chain = th.function(\n [pos, tmp_ctrl, num_sample],\n [pos_samples, inv_temp_samples, probs_0, probs_1, accepts],\n updates=updates\n )\n jct_chain_funcs.append(jct_chain)", "_____no_output_____" ], [ "num_sample = 50000\nfor i in range(10):\n jct_exp_dir = os.path.join(exp_dir, 'joint-ct', 'params-' + str(i))\n if not os.path.exists(jct_exp_dir):\n os.makedirs(jct_exp_dir)\n settings = {\n 'dt': dt,\n 'num_sample': num_sample,\n 'num_step': num_step,\n 'temp_scale': temp_scale,\n 'mom_resample_coeff': mom_resample_coeff\n }\n print('Parameters {0}'.format(i))\n print('-' * 100)\n print(settings)\n settings_path = os.path.join(jct_exp_dir, 'settings.json')\n results_path = os.path.join(jct_exp_dir, 'results.npz')\n with open(settings_path, 'w') as f:\n json.dump(settings, f, indent=True)\n num_dim = relaxation_list[i].n_dim_r\n pos_init = rng.normal(size=(num_runs, num_dim)).dot(\n var_covar_chol_list[i].T) + var_mean_list[i]\n tmp_ctrl_init = np.zeros(num_runs) - 10.\n start_time = time.time()\n pos_samples, inv_temp_samples, probs_0, probs_1, accepts = jct_chain_funcs[i](\n pos_init, tmp_ctrl_init, num_sample\n )\n sampling_time = time.time() - start_time\n print('Sampling time: {0:.2f}s'.format(sampling_time))\n log_norm_rmses = np.empty((num_reps, num_sample))\n mean_rmses = np.empty((num_reps, num_sample))\n covar_rmses = np.empty((num_reps, num_sample))\n for r in range(num_reps):\n log_norm_rmses[r], mean_rmses[r], covar_rmses[r] = calculate_incremental_rmses(\n pos_samples[:, r:(r+1)*num_runs_per_rep], \n probs_1[:, r:(r+1)*num_runs_per_rep], \n probs_0[:, r:(r+1)*num_runs_per_rep], \n true_log_norm_list[i] - var_log_norm_list[i],\n true_mean_list[i], true_covar_list[i]\n )\n var_log_norm_rmse = rmse(true_log_norm_list[i], var_log_norm_list[i])\n var_mean_rmse = rmse(true_mean_list[i], var_mean_list[i])\n var_covar_rmse = rmse(true_covar_list[i], \n var_covar_chol_list[i].dot(var_covar_chol_list[i].T))\n print('RMSE log_norm={0:.2f} mean={1:.2f} covar={2:.2f}'\n .format(\n np.mean(log_norm_rmses[:, -1]) / var_log_norm_rmse, \n np.mean(mean_rmses[:, -1]) / var_mean_rmse, \n np.mean(covar_rmses[:, -1]) / var_covar_rmse\n )\n )\n fig, axes = plt.subplots(1, 3, figsize=(9, 3))\n axes[0].semilogy(log_norm_rmses.mean(0) / var_log_norm_rmse)\n axes[0].set_title('Log norm RMSE')\n axes[1].semilogy(mean_rmses.mean(0) / var_mean_rmse)\n axes[1].set_title('Mean RMSE')\n axes[2].semilogy(covar_rmses.mean(0) / var_covar_rmse)\n axes[2].set_title('Covariance RMSE')\n plt.show()\n np.savez(\n results_path, \n sampling_time=sampling_time, \n pos_samples=pos_samples,\n inv_temp_samples=inv_temp_samples,\n probs_1=probs_1,\n probs_0=probs_0,\n accepts=accepts,\n log_norm_rmses=log_norm_rmses,\n mean_rmses=mean_rmses,\n covar_rmses=covar_rmses,\n var_log_norm_rmse=var_log_norm_rmse,\n var_mean_rmse=var_mean_rmse,\n var_covar_rmse=var_covar_rmse\n )\n print('Saved to ' + results_path)\n print('-' * 100)", "_____no_output_____" ] ] ]

[ "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown", "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Topological Data Analysis with Python and the Gudhi Library \n\n# Introduction to simplex trees ", "_____no_output_____" ], [ "**Authors** : F. Chazal and B. Michel", "_____no_output_____" ], [ "TDA typically aims at extracting topological signatures from a point cloud in $\\mathbb R^d$ or in a general metric space. By studying the topology of the point clouds, we actually mean studying the topology of unions of balls centered at the point cloud (offsets). However, non-discrete sets such as offsets, and also continuous mathematical shapes like curves, surfaces and more generally manifolds, cannot easily be encoded as finite discrete structures. [Simplicial complexes](https://en.wikipedia.org/wiki/Simplicial_complex) are therefore used in computational geometry to approximate such shapes.\n\nA simplicial complex is a set of [simplices](https://en.wikipedia.org/wiki/Simplex). It can be seen as a higher dimensional generalization of a graph. It is a mathematical object that is both topological and combinatorial, which makes it particularly useful for TDA. Here is an exemple of simplicial complex:\n\n\n \nA filtration is a increasing sequence of sub-complexes of a simplicial complex $\\mathcal K$. It can be seen as ordering the simplices included in the complex. Indeed, simpicial complexes often come with a specific order, as for [Vietoris-Rips complexes](https://en.wikipedia.org/wiki/Vietoris%E2%80%93Rips_complex), [Cech complexes](https://en.wikipedia.org/wiki/%C4%8Cech_complex) and [alpha complexes](https://en.wikipedia.org/wiki/Alpha_shape#Alpha_complex). ", "_____no_output_____" ] ], [ [ "from IPython.display import Image\nfrom os import chdir\nimport numpy as np\nimport gudhi as gd\nimport matplotlib.pyplot as plt", "_____no_output_____" ] ], [ [ "In Gudhi, filtered simplicial complexes are encoded through a data structure called simplex tree. \n\n\nThis notebook illustrates the use of simplex tree to represent simplicial complexes from data points.\n\nSee the [Python Gudhi documentation](https://gudhi.inria.fr/python/latest/simplex_tree_ref.html#) for more details on simplex trees.", "_____no_output_____" ], [ "### My first simplex tree", "_____no_output_____" ], [ "Let's create our first simplicial complex, represented by a simplex tree :", "_____no_output_____" ] ], [ [ "st = gd.SimplexTree()", "_____no_output_____" ] ], [ [ "The `st` object has class `SimplexTree`. For now, `st` is an empty simplex tree.\n\nThe `SimplexTree` class has several useful methods for the practice of TDA. For instance, there are methods to define new types of simplicial complexes from existing ones.\n\nThe `insert()` method can be used to insert simplices in the simplex tree. In the simplex tree:\n\n- vertices (0-dimensional simplices) are represented with integers, \n- edges (1-dimensional simplices) are represented with a length-2 list of integers (corresponding to the two vertices involved in the edge),\n- triangles (2-dimensional simplices) by three integers are represented with a length-3 list of integers (corresponding to the three vertices involved in the triangle),\n- etc.\n\nFor example, the following piece of code inserts three edges into the simplex tree:", "_____no_output_____" ] ], [ [ "st.insert([0, 1])\nst.insert([1, 2])\nst.insert([3, 1])", "_____no_output_____" ] ], [ [ "When the simplex is successfully inserted into the simplex tree, the `insert()` method outputs `True` as you can see from the execution of the above code. On the contrary, if the simplex is already in the filtration, the `insert()` method outputs `False`:", "_____no_output_____" ] ], [ [ "st.insert([3, 1])", "_____no_output_____" ] ], [ [ "We obtain the list of all the simplices in the simplex tree with the `get_filtration()` method : ", "_____no_output_____" ] ], [ [ "st_gen = st.get_filtration() ", "_____no_output_____" ] ], [ [ "The output `st_gen` is a generator and we thus we can iterate on its elements. Each element in the list is a tuple that contains a simplex and its **filtration value**.", "_____no_output_____" ] ], [ [ "for splx in st_gen :\n print(splx)", "([0], 0.0)\n([1], 0.0)\n([0, 1], 0.0)\n([2], 0.0)\n([1, 2], 0.0)\n([3], 0.0)\n([1, 3], 0.0)\n" ] ], [ [ "Intuitively, the filtration value of a simplex in a filtered complex acts as a *time stamp* corresponding to \"when\" the simplex appears in the filtration. By default, the `insert()` method assigns a filtration value equal to 0.\n\nNotice that inserting an edge automatically inserts its vertices (if they were not already in the complex) in order to satisfy the **inclusion property** of a filtered complex: any simplex with filtration value $t$ must have all its faces in the filtered complex, with filtration values smaller than or equal to $t$.", "_____no_output_____" ], [ "### Simplex tree description", "_____no_output_____" ], [ "The dimension of a simplical complex is the largest dimension of the simplices in it. It can be retrieved by the simplex tree `dimension()` method:", "_____no_output_____" ] ], [ [ "st.dimension()", "_____no_output_____" ] ], [ [ "It is possible to compute the number of vertices in a simplex tree via the `num_vertices()` method:", "_____no_output_____" ] ], [ [ "st.num_vertices()", "_____no_output_____" ] ], [ [ "The number of simplices in the simplex tree is given by", "_____no_output_____" ] ], [ [ "st.num_simplices()", "_____no_output_____" ] ], [ [ "The [$d$-skeleton](https://en.wikipedia.org/wiki/N-skeleton) -- which is the union of all simplices of dimensions smaller than or equal to $d$ -- can be also computed with the `get_skeleton()` method. This method takes as argument the dimension of the desired skeleton. To retrieve the topological graph from a simplex tree, we can therefore call:", "_____no_output_____" ] ], [ [ "print(st.get_skeleton(1))", "[([0, 1], 0.0), ([0], 0.0), ([1, 2], 0.0), ([1, 3], 0.0), ([1], 0.0), ([2], 0.0), ([3], 0.0)]\n" ] ], [ [ "One can also check whether a simplex is already in the filtration. This is achieved with the `find()` method:", "_____no_output_____" ] ], [ [ "st.find([2, 4])", "_____no_output_____" ] ], [ [ "### Filtration values\n\nWe can insert simplices at a given filtration value. For example, the following piece of code will insert three triangles in the simplex tree at three different filtration values:", "_____no_output_____" ] ], [ [ "st.insert([0, 1, 2], filtration = 0.1)\nst.insert([1, 2, 3], filtration = 0.2)\nst.insert([0, 1, 3], filtration = 0.4)\nst_gen = st.get_filtration() \n\nfor splx in st_gen :\n print(splx)", "([0], 0.0)\n([1], 0.0)\n([0, 1], 0.0)\n([2], 0.0)\n([1, 2], 0.0)\n([3], 0.0)\n([1, 3], 0.0)\n([0, 2], 0.1)\n([0, 1, 2], 0.1)\n([2, 3], 0.2)\n([1, 2, 3], 0.2)\n([0, 3], 0.4)\n([0, 1, 3], 0.4)\n" ] ], [ [ "As you can see, when we add a new simplex with a given filtration value, all its faces that were not already in the complex are added with the same filtration value: here the edge `[0, 3]` was not part of the tree before including the triangle `[0, 1, 3]` and is thus inserted with the filtration value of the inserted triangle. On the other hand, the filtration value of the faces of added simplices that were already part of the tree before is left alone. One can modify the filtration value of any simplex included in the tree with the `assign_filtration()` method:", "_____no_output_____" ] ], [ [ "st.assign_filtration([3], filtration = 0.8)\nst_gen = st.get_filtration()\nfor splx in st_gen:\n print(splx) ", "([0], 0.0)\n([1], 0.0)\n([0, 1], 0.0)\n([2], 0.0)\n([1, 2], 0.0)\n([1, 3], 0.0)\n([0, 2], 0.1)\n([0, 1, 2], 0.1)\n([2, 3], 0.2)\n([1, 2, 3], 0.2)\n([0, 3], 0.4)\n([0, 1, 3], 0.4)\n([3], 0.8)\n" ] ], [ [ "Notice that, the vertex `[3]` has been moved to the end of the filtration because it now has the highest filtration value. However, this simplex tree is not a filtered simplicial complex anymore because the filtration value of the vertex `[3]` is higher than the filtration value of the edge `[2 3]`. We can use the `make_filtration_non_decreasing()` method to solve the problem:", "_____no_output_____" ] ], [ [ "st.make_filtration_non_decreasing()\nst_gen = st.get_filtration()\nfor splx in st_gen:\n print(splx) ", "([0], 0.0)\n([1], 0.0)\n([0, 1], 0.0)\n([2], 0.0)\n([1, 2], 0.0)\n([0, 2], 0.1)\n([0, 1, 2], 0.1)\n([3], 0.8)\n([0, 3], 0.8)\n([1, 3], 0.8)\n([0, 1, 3], 0.8)\n([2, 3], 0.8)\n([1, 2, 3], 0.8)\n" ] ], [ [ "Finally, it is worth mentioning the `filtration()` method, which returns the filtration value of a given simplex in the filtration :", "_____no_output_____" ] ], [ [ "st.filtration([2, 3])", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Non-Linear Classifiers", "_____no_output_____" ] ], [ [ "# Global variables for testing changes to this notebook quickly\nRANDOM_SEED = 0\nNUM_FOLDS = 5", "_____no_output_____" ], [ "import numpy as np\nimport pandas as pd\nimport time\nimport math\nimport os\nimport pyarrow\nimport gc\n\n# scikit-learn optimization\nfrom sklearnex import patch_sklearn\npatch_sklearn()\n\n# Model evaluation\nfrom sklearn.base import clone\nfrom sklearn.model_selection import StratifiedKFold, train_test_split\nfrom sklearn.metrics import accuracy_score\n\n# Plotting\nimport matplotlib\nimport seaborn as sns\nfrom matplotlib import pyplot as plt\nfrom IPython.display import Image\n\n# Hide warnings\nimport warnings\nwarnings.filterwarnings('ignore')", "Intel(R) Extension for Scikit-learn* enabled (https://github.com/intel/scikit-learn-intelex)\n" ] ], [ [ "# Scoring Function", "_____no_output_____" ] ], [ [ "# Scoring/Training Baseline Function\ndef score_model(sklearn_model, preprocessing = None):\n \n # Store the holdout predictions\n oof_preds = np.zeros((train.shape[0],))\n scores = np.zeros(NUM_FOLDS)\n times = np.zeros(NUM_FOLDS)\n print('')\n \n # Stratified k-fold cross-validation\n skf = StratifiedKFold(n_splits = NUM_FOLDS, shuffle = True, random_state = RANDOM_SEED)\n for fold, (train_idx, valid_idx) in enumerate(skf.split(train, target_bins)):\n \n # Training and Validation Sets\n X_train, y_train = train[features].iloc[train_idx], train['target'].iloc[train_idx]\n X_valid, y_valid = train[features].iloc[valid_idx], train['target'].iloc[valid_idx]\n train_weight, valid_weight = train['sample_weight'].iloc[train_idx], train['sample_weight'].iloc[valid_idx]\n \n # Preprocessing\n start = time.time()\n if preprocessing:\n X_train = preprocessing.fit_transform(X_train)\n X_valid = preprocessing.transform(X_valid)\n \n # Create model\n model = clone(sklearn_model)\n try:\n model.fit(X_train, y_train, sample_weight = train_weight)\n except:\n model.fit(X_train, y_train)\n \n # validation\n valid_preds = model.predict(X_valid)\n scores[fold] = accuracy_score(y_valid, valid_preds, sample_weight = valid_weight)\n oof_preds[valid_idx] = valid_preds\n end = time.time()\n print(f'Fold {fold}: {round(scores[fold], 5)} accuracy in {round(end-start,2)}s.')\n times[fold] = end-start\n \n mask1, mask10 = train.gcd == 1, train.gcd == 10 \n mask1000, mask10000 = train.gcd == 1000, train.gcd == 10000\n print(\"\\nAccuracy (1M Reads):\", round(accuracy_score(oof_preds[mask1], train['target'].loc[mask1], sample_weight = train['sample_weight'].loc[mask1]), 5))\n print(\"Accuracy (100k Reads):\", round(accuracy_score(oof_preds[mask10], train['target'].loc[mask10], sample_weight = train['sample_weight'].loc[mask10]), 5))\n print(\"Accuracy (1k Reads):\", round(accuracy_score(oof_preds[mask1000], train['target'].loc[mask1000], sample_weight = train['sample_weight'].loc[mask1000]), 5))\n print(\"Accuracy (100 Reads):\", round(accuracy_score(oof_preds[mask10000], train['target'].loc[mask10000], sample_weight = train['sample_weight'].loc[mask10000]), 5))\n print(\"Out-of-Fold Accuracy:\", round(accuracy_score(oof_preds, train['target'], sample_weight = train['sample_weight']), 5))\n print(f'Training Time: {round(times.sum(), 2)}s')\n \n return oof_preds", "_____no_output_____" ], [ "from sklearn.metrics import ConfusionMatrixDisplay, confusion_matrix, accuracy_score\nimport matplotlib.pyplot as plt\n\n# Plot confusion matrix\ndef plot_confusion_matrix(true_values, pred_values, gcds, plot_title = \"Confusion Matrix\"):\n \n gcd = [[1,10],[1000,10000]]\n \n # Confusion matrix\n fig, ax = plt.subplots(2, 2, figsize = (12,9))\n for row in range(2):\n for col in range(2):\n idx = 2*row + col\n cm = confusion_matrix(true_values[gcds == gcd[row][col]], pred_values[gcds == gcd[row][col]])\n np.fill_diagonal(cm, 0)\n disp = ConfusionMatrixDisplay(confusion_matrix = cm)\n disp.plot(ax = ax[row,col])\n plt.show()", "_____no_output_____" ] ], [ [ "# Load Data", "_____no_output_____" ] ], [ [ "%%time\nfrom sklearn.preprocessing import LabelEncoder\n\ntrain = pd.read_feather('../data/train.feather')\nfeatures = [x for x in train.columns if x not in ['row_id','target','sample_weight','gcd']]\n\nencoder = LabelEncoder()\ntrain['target'] = encoder.fit_transform(train['target'])\ntarget_bins = train['target'].astype(str) + train['gcd'].astype(str)\n\nprint(f'Training Samples: {len(train)}')", "Training Samples: 123993\nCPU times: total: 1.12 s\nWall time: 195 ms\n" ] ], [ [ "# Naive Bayes", "_____no_output_____" ] ], [ [ "from sklearn.naive_bayes import MultinomialNB\nfrom sklearn.preprocessing import PowerTransformer, MinMaxScaler\nfrom math import factorial\n\ndef fix_bias(input_df, add = True):\n df = input_df.copy()\n bias = lambda w, x, y, z: factorial(10) / (factorial(w) * factorial(x) * factorial(y) * factorial(z) * 4**10)\n \n for col in features:\n w = int(col[1:col.index('T')])\n x = int(col[col.index('T')+1:col.index('G')])\n y = int(col[col.index('G')+1:col.index('C')])\n z = int(col[col.index('C')+1:])\n if add:\n df[col] = df[col] + bias(w, x, y, z)\n else:\n df[col] = df[col] - bias(w, x, y, z)\n return df", "_____no_output_____" ], [ "# Naive Bayes\noof_preds = score_model(\n MultinomialNB(),\n MinMaxScaler()\n)\n\nplot_confusion_matrix(train['target'], oof_preds, train['gcd'])", "\nFold 0: 0.55738 accuracy in 0.43s.\nFold 1: 0.56818 accuracy in 0.29s.\nFold 2: 0.54745 accuracy in 0.31s.\nFold 3: 0.56104 accuracy in 0.31s.\nFold 4: 0.55401 accuracy in 0.31s.\n\nAccuracy (1M Reads): 0.61076\nAccuracy (100k Reads): 0.61098\nAccuracy (1k Reads): 0.56488\nAccuracy (100 Reads): 0.4438\nOut-of-Fold Accuracy: 0.55762\nTraining Time: 1.65s\n" ] ], [ [ "# KNN Classifier", "_____no_output_____" ] ], [ [ "from sklearn.neighbors import KNeighborsClassifier\nfrom sklearn.preprocessing import StandardScaler", "_____no_output_____" ], [ "# KNN\noof_preds = score_model(\n KNeighborsClassifier(n_neighbors = 1),\n StandardScaler()\n)\n\nplot_confusion_matrix(train['target'], oof_preds, train['gcd'])", "\nFold 0: 0.91848 accuracy in 3.23s.\nFold 1: 0.91989 accuracy in 3.25s.\nFold 2: 0.91832 accuracy in 3.45s.\nFold 3: 0.91746 accuracy in 3.47s.\nFold 4: 0.91839 accuracy in 3.42s.\n\nAccuracy (1M Reads): 1.0\nAccuracy (100k Reads): 0.99998\nAccuracy (1k Reads): 0.84836\nAccuracy (100 Reads): 0.82578\nOut-of-Fold Accuracy: 0.91851\nTraining Time: 16.82s\n" ] ], [ [ "# Radius Neighbors", "_____no_output_____" ] ], [ [ "from sklearn.neighbors import RadiusNeighborsClassifier\n\n# Radius Neighbors\noof_preds = score_model(\n RadiusNeighborsClassifier(\n n_jobs = -1,\n outlier_label = 'most_frequent',\n ),\n StandardScaler()\n)\n\nplot_confusion_matrix(train['target'], oof_preds, train['gcd'])", "\nFold 0: 0.36938 accuracy in 55.13s.\nFold 1: 0.36505 accuracy in 54.6s.\nFold 2: 0.36727 accuracy in 54.28s.\nFold 3: 0.37093 accuracy in 54.42s.\nFold 4: 0.37039 accuracy in 54.55s.\n\nAccuracy (1M Reads): 0.88689\nAccuracy (100k Reads): 0.38838\nAccuracy (1k Reads): 0.09996\nAccuracy (100 Reads): 0.09964\nOut-of-Fold Accuracy: 0.3686\nTraining Time: 272.97s\n" ] ], [ [ "# Nearest Centroid", "_____no_output_____" ] ], [ [ "from sklearn.neighbors import NearestCentroid\n\n# Nearest Centroid\noof_preds = score_model(\n NearestCentroid(),\n StandardScaler()\n)\n\nplot_confusion_matrix(train['target'], oof_preds, train['gcd'])", "\nFold 0: 0.51745 accuracy in 0.52s.\nFold 1: 0.52542 accuracy in 0.51s.\nFold 2: 0.5295 accuracy in 0.55s.\nFold 3: 0.53634 accuracy in 0.56s.\nFold 4: 0.51784 accuracy in 0.53s.\n\nAccuracy (1M Reads): 0.56773\nAccuracy (100k Reads): 0.57384\nAccuracy (1k Reads): 0.55603\nAccuracy (100 Reads): 0.40347\nOut-of-Fold Accuracy: 0.52529\nTraining Time: 2.66s\n" ] ], [ [ "# Support Vector Machines", "_____no_output_____" ] ], [ [ "from sklearn.svm import SVC\n\n# Polynomial SVM\noof_preds = score_model(\n SVC(kernel = \"poly\", degree = 2, coef0 = 1),\n StandardScaler()\n)\n\nplot_confusion_matrix(train['target'], oof_preds, train['gcd'])", "\nFold 0: 0.90259 accuracy in 17.89s.\nFold 1: 0.90317 accuracy in 18.06s.\nFold 2: 0.90762 accuracy in 17.77s.\nFold 3: 0.90415 accuracy in 17.83s.\nFold 4: 0.90168 accuracy in 17.53s.\n\nAccuracy (1M Reads): 0.95937\nAccuracy (100k Reads): 0.9743\nAccuracy (1k Reads): 0.92233\nAccuracy (100 Reads): 0.75928\nOut-of-Fold Accuracy: 0.90384\nTraining Time: 89.09s\n" ], [ "# Polynomial SVM\noof_preds = score_model(\n SVC(kernel = \"rbf\"),\n StandardScaler()\n)\n\nplot_confusion_matrix(train['target'], oof_preds, train['gcd'])", "\nFold 0: 0.93306 accuracy in 28.33s.\nFold 1: 0.92856 accuracy in 27.51s.\nFold 2: 0.92868 accuracy in 29.49s.\nFold 3: 0.93257 accuracy in 29.49s.\nFold 4: 0.92989 accuracy in 28.31s.\n\nAccuracy (1M Reads): 0.96342\nAccuracy (100k Reads): 0.98104\nAccuracy (1k Reads): 0.9288\nAccuracy (100 Reads): 0.84891\nOut-of-Fold Accuracy: 0.93055\nTraining Time: 143.12s\n" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Hydrogen atom\n\n\\begin{equation}\n\\label{eq1}\n-\\frac{\\hbar^2}{2 \\mu} \\left[ \\frac{1}{r^2} \\frac{\\partial }{\\partial r} \\left( r^2 \\frac{ \\partial \\psi}{\\partial r}\\right) + \\frac{1}{r^2 \\sin \\theta} \\frac{\\partial }{\\partial \\theta} \\left( \\sin \\theta \\frac{\\partial \\psi}{\\partial \\theta}\\right) + \\frac{1}{r^2 \\sin^2 \\theta} \\frac{\\partial^2 \\psi}{\\partial \\phi^2} \\right] - \\frac{e^2}{ 4 \\pi \\epsilon_0 r} \\psi= E \\psi\n\\end{equation}\n\n\n\\begin{equation}\n\\label{eqsol1}\n\\psi_{n\\ell m}(r,\\vartheta,\\varphi) = \\sqrt {{\\left ( \\frac{2}{n a^*_0} \\right )}^3 \\frac{(n-\\ell-1)!}{2n(n+\\ell)!}} e^{- \\rho / 2} \\rho^{\\ell} L_{n-\\ell-1}^{2\\ell+1}(\\rho) Y_{\\ell}^{m}(\\vartheta, \\varphi ) \n\\end{equation}\n", "_____no_output_____" ] ], [ [ "import k3d\nfrom ipywidgets import interact, FloatSlider\n\nimport numpy as np\nimport scipy.special\nimport scipy.misc\n\nr = lambda x,y,z: np.sqrt(x**2+y**2+z**2)\ntheta = lambda x,y,z: np.arccos(z/r(x,y,z))\nphi = lambda x,y,z: np.arctan2(y,x)\n\na0 = 1.\nR = lambda r,n,l: (2*r/(n*a0))**l * np.exp(-r/n/a0) * scipy.special.genlaguerre(n-l-1,2*l+1)(2*r/n/a0)\nWF = lambda r,theta,phi,n,l,m: R(r,n,l) * scipy.special.sph_harm(m,l,phi,theta)\nabsWF = lambda r,theta,phi,n,l,m: abs(WF(r,theta,phi,n,l,m)).astype(np.float32)**2\nN = 50j\na = 30.0\nx,y,z = np.ogrid[-a:a:N,-a:a:N,-a:a:N]\nx = x.astype(np.float32)\ny = y.astype(np.float32)\nz = z.astype(np.float32)", "_____no_output_____" ], [ "orbital = WF(r(x,y,z),theta(x,y,z),phi(x,y,z),4,1,0).real.astype(np.float32) # 4p", "_____no_output_____" ], [ "plt_vol = k3d.volume(orbital)\nplt_label = k3d.text2d(r'n=1\\; l=0\\; m=0',(0.,0.))\nplot = k3d.plot()\nplot += plt_vol\nplot += plt_label\n\nplt_vol.opacity_function = [0. , 0. , 0.21327923, 0.98025 , 0.32439035,\n 0. , 0.5 , 0. , 0.67560965, 0. ,\n 0.74537706, 0.9915 , 1. , 0. ]\n\nplt_vol.color_map = k3d.colormaps.paraview_color_maps.Cool_to_Warm_Extended\nplt_vol.color_range = (-0.5,0.5)\n\n\n\nplot.display()", "_____no_output_____" ] ], [ [ "## animation \n### single wave function is sent at a time", "_____no_output_____" ] ], [ [ "E = 4\nfor l in range(E):\n for m in range(-l,l+1):\n psi2 = WF(r(x,y,z),theta(x,y,z),phi(x,y,z),E,l,m).real.astype(np.float32)\n plt_vol.volume = psi2/np.max(psi2)\n plt_label.text = 'n=%d \\quad l=%d \\quad m=%d'%(E,l,m)\n \n", "_____no_output_____" ] ], [ [ "### using time series \n\n - series of volumetric data are sent to k3d, \n - player interpolates between ", "_____no_output_____" ] ], [ [ "E = 4\npsi_t = {}\nt = 0.0\nfor l in range(E):\n for m in range(-l,l+1):\n \n psi2 = WF(r(x,y,z),theta(x,y,z),phi(x,y,z),E,l,m)\n psi_t[str(t)] = (psi2.real/np.max(np.abs(psi2))).astype(np.float32)\n t += 0.3 \nplt_vol.volume = psi_t \n", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "\n\n# Demo text-mining: Pharma case\n\nIn this demo, I will demonstrate what are the basic steps that you will have to use in most text-mining cases. This are also some of the steps that have been used in the ResuMe app, available here: [ResuMe](https://resume.businessdecision.be/). The case that we will cover here, is a simplified version of a project that has actually been carried out by B&D, where the goal was to identify if a given paper is treating about Pharmacovigilance or not. Pharmacovigilance is a domain of study in healthcare about drug safety. Consequently, we would like to predict, based on the text of the scientific article if the article treats about Pharmacovigilance or not.\n\nFor this we can use any kind of model, but in any case we will have to transform the words in numbers in some way. We'll see different methods and compare their performance.\n\n## Downloading the dataset from PubMed", "_____no_output_____" ] ], [ [ "#import documents from PubMed\nfrom Bio import Entrez\n\n# Function to search for a certain number articles based on a certain keyword\ndef search(keyword,number=20):\n Entrez.email = '[email protected]'\n handle = Entrez.esearch(db='pubmed', \n sort='relevance', \n retmax=str(number),\n retmode='xml', \n term=keyword)\n results = Entrez.read(handle)\n return results\n\n# Function to retrieve the results of previous search query\ndef fetch_details(id_list):\n ids = ','.join(id_list)\n Entrez.email = '[email protected]'\n handle = Entrez.efetch(db='pubmed',\n retmode='xml',\n id=ids)\n results = Entrez.read(handle)\n return results\n", "_____no_output_____" ] ], [ [ "### Retrieving top 200 articles with Pharmacovigilance keyword", "_____no_output_____" ] ], [ [ "results = search('Pharmacovigilance', 200) #querying PubMed\nid_list = results['IdList']\npapers_pharmacov = fetch_details(id_list) #retrieving the info about the articles in nested lists & dictionary format", "_____no_output_____" ], [ "# checking article title for the first 10 retrieved articles\nfor i, paper in enumerate(papers_pharmacov['PubmedArticle'][:10]):\n print(\"%d) %s\" % (i+1,paper['MedlineCitation']['Article']['ArticleTitle']))", "1) FarmaREL: An Italian pharmacovigilance project to monitor and evaluate adverse drug reactions in haematologic patients.\n2) Feasibility and Educational Value of a Student-Run Pharmacovigilance Programme: A Prospective Cohort Study.\n3) Developing a Crowdsourcing Approach and Tool for Pharmacovigilance Education Material Delivery.\n4) Promoting and Protecting Public Health: How the European Union Pharmacovigilance System Works.\n5) Effect of an educational intervention on knowledge and attitude regarding pharmacovigilance and consumer pharmacovigilance among community pharmacists in Lalitpur district, Nepal.\n6) Pharmacovigilance and Biomedical Informatics: A Model for Future Development.\n7) Pharmacovigilance in Europe: Place of the Pharmacovigilance Risk Assessment Committee (PRAC) in organisation and decisional processes.\n8) Tamoxifen Pharmacovigilance: Implications for Safe Use in the Future.\n9) Pharmacovigilance Skills, Knowledge and Attitudes in our Future Doctors - A Nationwide Study in the Netherlands.\n10) Adverse drug reactions reporting in Calabria (Southern Italy) in the four-year period 2011-2014: impact of a regional pharmacovigilance project in light of the new European Legislation.\n" ] ], [ [ "### Retrieving top 1.000 articles with Pharma keyword\nThis will be our base of comparison, we want to separate them from the others", "_____no_output_____" ] ], [ [ "results = search('Pharma', 1000) #querying PubMed\nid_list = results['IdList']\npapers_pharma = fetch_details(id_list)#retrieving the info about the articles in nested lists & dictionary format", "_____no_output_____" ], [ "# checking article title for the first 10 retrieved articles\nfor i, paper in enumerate(papers_pharma['PubmedArticle'][:10]):\n print(\"%d) %s\" % (i+1,paper['MedlineCitation']['Article']['ArticleTitle']))", "1) Recent trends in specialty pharma business model.\n2) The moderating role of absorptive capacity and the differential effects of acquisitions and alliances on Big Pharma firms' innovation performance.\n3) Space-related pharma-motifs for fast search of protein binding motifs and polypharmacological targets.\n4) Pharma Websites and \"Professionals-Only\" Information: The Implications for Patient Trust and Autonomy.\n5) BRIC Health Systems and Big Pharma: A Challenge for Health Policy and Management.\n6) Developing Deep Learning Applications for Life Science and Pharma Industry.\n7) Exzellenz in der Bildung für eine innovative Schweiz: Die Position des Wirtschaftsdachverbandes Chemie Pharma Biotech.\n8) Shaking Up Biotech/Pharma: Can Cues Be Taken from the Tech Industry?\n9) Pharma-Nutritional Properties of Olive Oil Phenols. Transfer of New Findings to Human Nutrition.\n10) Pharma Success in Product Development—Does Biotechnology Change the Paradigm in Product Development and Attrition.\n" ] ], [ [ "### Saving ID's, labels and title + abstracts of the articles\n\nWhen an article was retrieved via the Pharmacovigilance keyword, it will receive the label = 1 and = 0 else. We'll per article put the article title and article abstract together as our text data on the article. ", "_____no_output_____" ] ], [ [ "# Save ids & label 1 = pharmacovigilance , 0 = not pharmacovigilance\n# & Save title + abstract in dico\nids = []\nlabels = []\ndata = []\nfor i, paper in enumerate(papers_pharmacov['PubmedArticle']):\n if 'Abstract' in paper['MedlineCitation']['Article'].keys(): #check that abstract info is available\n ids.append(str(paper['MedlineCitation']['PMID']))\n labels.append(1)\n title = paper['MedlineCitation']['Article']['ArticleTitle'] #Article title\n abstract = paper['MedlineCitation']['Article']['Abstract']['AbstractText'][0] #Abstract\n data.append( title + abstract )\nfor i, paper in enumerate(papers_pharma['PubmedArticle']):\n if 'Abstract' in paper['MedlineCitation']['Article'].keys(): #check that abstract info is available\n ids.append(str(paper['MedlineCitation']['PMID']))\n labels.append(0)\n title = paper['MedlineCitation']['Article']['ArticleTitle'] #Article title\n abstract = paper['MedlineCitation']['Article']['Abstract']['AbstractText'][0] #Abstract\n data.append( title + abstract )\n", "_____no_output_____" ], [ "# Check result for one paper\nids[0] # ID\nlabels[0] # 1 = pharmacovigilance , 0 = not pharmacovigilance\ndata[0] # Title & abstract", "_____no_output_____" ] ], [ [ "### Transform to numeric attributes\nWe will now **transform** the **text into numeric attributes**. For this, we will convert every word to a number, but we first need to **split** the full text into **separate words**. This is done by using a ***Tokenizer***. The tokenizer will split the full text based on a certain pattern you specify. Here we'll take a very basic pattern and take any words that contain only upper- or lowercase letters and we will convert everything to lowercase.", "_____no_output_____" ] ], [ [ "from nltk.tokenize.regexp import RegexpTokenizer #import a tokenizer, to split the full text into separate words\n\ndef Tokenize_text_value(value):\n tokenizer1 = RegexpTokenizer(r\"[A-Za-z]+\") # our self defined tokenizera\n value = value.lower() # convert all words to lowercase\n return tokenizer1.tokenize(value) # tokenize each text", "_____no_output_____" ], [ "# example of our tokenizer\nTokenize_text_value(data[0])", "_____no_output_____" ] ], [ [ "Using the ***bag-of-words*** method we can transform any document to a vector. Using this method you have **one column per word and one row per document** and either a binary value 1 if the word is present in a certain document, 0 if not or a count value of the number of times the word appears in the document. \n\nFor instance, the following three sentences:\n1. Intelligent applications creates intelligent business processes\n2. Bots are intelligent applications\n3. I do business intelligence\n\nCan be represented in the following matrix using the counts of each word as values in the matrix\n", "_____no_output_____" ] ], [ [ "# transform non-processed data to nummeric features:\nfrom sklearn.feature_extraction.text import TfidfVectorizer\nbinary_vectorizer = TfidfVectorizer(input=u'content', analyzer=u'word', binary = True,\n tokenizer=Tokenize_text_value) # initialize the binary vectorizer\ncount_vectorizer = TfidfVectorizer(input=u'content', analyzer=u'word', use_idf=False, \n tokenizer=Tokenize_text_value) # initialize the count vectorizer\n\nbinary_matrix = binary_vectorizer.fit_transform(data) # fit & transform\ncount_matrix = count_vectorizer.fit_transform(data) # fit & transform", "_____no_output_____" ], [ "# Check our output matrix shape: rows = documents, columns = words\nbinary_matrix.shape", "_____no_output_____" ] ], [ [ "### Check performance in a basic model\nWe'll apply now a model on our 2 matrices. For this we will use the ***Naive Bayes model***, which (as the name tells) is based on the probabilistic Bayes theorem. It is used a lot in text-mining as it is really **fast** to train and apply and is able to **handle a lot of features**, which is often the case in text-mining, when you have one column per word. We will use the ***kappa*** measure to evaluate model performance. Kappa is a metric that is robust to class-imbalances in the data and varies from -1 to +1 with 0 being a random performance and +1 a perfect performance.", "_____no_output_____" ] ], [ [ "# apply cross validation Naive Bayes model\nfrom sklearn.model_selection import cross_val_score\nfrom sklearn.naive_bayes import MultinomialNB\nfrom sklearn.metrics import (cohen_kappa_score, make_scorer)\n\nNB = MultinomialNB() # our Naive Bayes Model initialisation\nscorer = make_scorer(cohen_kappa_score) # Our kappa score", "_____no_output_____" ], [ "scores = cross_val_score(NB,binary_matrix,labels,scoring = scorer,cv = 5 )\nprint('Cross validation on Binary matrix with a mean kappa score of %f and variance of %f' % (scores.mean(),scores.var()))", "Cross validation on Binary matrix with a mean kappa score of 0.210382 and variance of 0.003118\n" ], [ "scores = cross_val_score(NB,count_matrix,labels,scoring = scorer,cv = 5 )\nprint('Cross validation on Count matrix with a mean kappa score of %f and variance of %f' % (scores.mean(),scores.var()))", "Cross validation on Count matrix with a mean kappa score of 0.064193 and variance of 0.000682\n" ] ], [ [ "### TF-IDF transformation\n\nAn alternative to the binary and count matrix is the **tf-idf transformation**. It stands for ***Term Frequency - Inverse Document Frequency*** and is a measure that will try to find the words that are unique to each document and that characterizes the document compared to the other documents. How this achieved is by taking the term frequency (which is the same as the count that we have defined before) and multiplying it by the inverse document frequency (which is low when the term appears in all other documents and high when it appears in few other documents):\n\n\n*Copyright © Chris Albon, 2018*", "_____no_output_____" ] ], [ [ "# transform non-processed data to nummeric features:\nfrom sklearn.feature_extraction.text import TfidfVectorizer\ntfidf_vectorizer = TfidfVectorizer(input=u'content', analyzer=u'word', use_idf=True, smooth_idf = True ,\n tokenizer=Tokenize_text_value) # initialize the tf-idf vectorizer\n\ntfidf_matrix = tfidf_vectorizer.fit_transform(data) # fit & transform", "_____no_output_____" ], [ "scores = cross_val_score(NB,tfidf_matrix,labels,scoring = scorer,cv = 5 )\nprint('Cross validation on TF-IDF matrix with a mean kappa score of %f and variance of %f' % (scores.mean(),scores.var()))", "Cross validation on TF-IDF matrix with a mean kappa score of 0.320332 and variance of 0.003960\n" ] ], [ [ "### How to improve this score? \nHow come the TF-IDF works the best, followed closely by the binary matrix and with the count matrix far behind? Let's have a look at the words that occur the most in the different documents:", "_____no_output_____" ] ], [ [ "import numpy as np\n# Find words with maximum occurence for each document in the count_matrix\nmax_counts_per_doc = np.asarray(np.argmax(count_matrix,axis = 1)).ravel()\n# Count how many times every word is the most occuring word across all documents\nunique, counts = np.unique(max_counts_per_doc,return_counts=True)\n# Keep only the words that are the most frequent word of at least 5 different documents\nfrequent = unique[counts > 5]", "_____no_output_____" ], [ "# Retrieve the vocabulary of our count matrix\nvocab = count_vectorizer.get_feature_names()\n# print out the words in frequent\nfor i in frequent:\n print(vocab[i])", "a\nand\nfor\nin\nof\nthe\nto\nwith\n" ] ], [ [ "As you can see those words are all words without any added value as they are mostly used to link certain words together in sentences, but have no standalone value. This is what we call ***Stop words***. So knowing that, we can find an intuition of why the tf-idf and binary transformations worked better than the count one. In the count one, we have seen that words that appear a lot, but have no value as such, get a high weight/value, whereas in binary every word gets the same weight and in tf-idf, the words that appear a lot in the other documents are automatically given a lower weight thanks to the IDF part. To avoid this problem we usually remove stop words\n\n### Removing Stop words", "_____no_output_____" ] ], [ [ "# Remove the stop words\nbinary_vectorizer = TfidfVectorizer(input=u'content', analyzer=u'word', binary = True,\n tokenizer=Tokenize_text_value, stop_words = 'english') # initialize the binary vectorizer\ncount_vectorizer = TfidfVectorizer(input=u'content', analyzer=u'word', use_idf=False, smooth_idf=False,\n tokenizer=Tokenize_text_value, stop_words = 'english') # initialize the count vectorizer\ntfidf_vectorizer = TfidfVectorizer(input=u'content', analyzer=u'word', use_idf=True, smooth_idf=True,\n tokenizer=Tokenize_text_value, stop_words = 'english') # initialize the tf-idf vectorizer\nbinary_matrix = binary_vectorizer.fit_transform(data) # fit & transform\ncount_matrix = count_vectorizer.fit_transform(data) # fit & transform\ntfidf_matrix = tfidf_vectorizer.fit_transform(data) # fit & transform", "_____no_output_____" ], [ "scores = cross_val_score(NB,binary_matrix,labels,scoring = scorer,cv = 5 )\nprint('Cross validation on Binary matrix by removing stop-words with a mean kappa score of %f and variance of %f' % (scores.mean(),scores.var()))", "Cross validation on Binary matrix by removing stop-words with a mean kappa score of 0.472072 and variance of 0.001247\n" ], [ "scores = cross_val_score(NB,count_matrix,labels,scoring = scorer,cv = 5 )\nprint('Cross validation on Count matrix by removing stop-words with a mean kappa score of %f and variance of %f' % (scores.mean(),scores.var()))", "Cross validation on Count matrix by removing stop-words with a mean kappa score of 0.753020 and variance of 0.009759\n" ], [ "scores = cross_val_score(NB,tfidf_matrix,labels,scoring = scorer,cv = 5 )\nprint('Cross validation on TF-IDF matrix by removing stop-words with a mean kappa score of %f and variance of %f' % (scores.mean(),scores.var()))", "Cross validation on TF-IDF matrix by removing stop-words with a mean kappa score of 0.682766 and variance of 0.011562\n" ] ], [ [ "We have a big improvement in our performance when we remove the stop words. How can we go a step further? Now the following steps are mostly domain dependent. You have to think about your problem and what you would need to solve it. In this case, if we are using only the abstracts and the titles, if we had to do it ourselves, we would have a look at the most common keywords you have in the articles about Pharmacovigilance and when we have a new article to classify, we would look if we find those same keywords back. However, here we are analyzing all words (minus the stopwords) and not only the keywords. So we could try to filter out to keep only words that appear at least a certain number of times across all documents.\n### Keeping only key-words", "_____no_output_____" ] ], [ [ "# keep only words that appear at least in 5% of the documents:\nbinary_vectorizer = TfidfVectorizer(input=u'content', analyzer=u'word', binary = True,\n tokenizer=Tokenize_text_value, stop_words = 'english'\n , min_df = 0.05) # initialize the binary vectorizer\ncount_vectorizer = TfidfVectorizer(input=u'content', analyzer=u'word', use_idf=False, smooth_idf=False,\n tokenizer=Tokenize_text_value, stop_words = 'english'\n , min_df = 0.05) # initialize the count vectorizer\ntfidf_vectorizer = TfidfVectorizer(input=u'content', analyzer=u'word', use_idf=True, smooth_idf=True,\n tokenizer=Tokenize_text_value, stop_words = 'english'\n , min_df = 0.05) # initialize the tf-idf vectorizer\nbinary_matrix = binary_vectorizer.fit_transform(data) # fit & transform\ncount_matrix = count_vectorizer.fit_transform(data) # fit & transform\ntfidf_matrix = tfidf_vectorizer.fit_transform(data) # fit & transform", "_____no_output_____" ], [ "scores = cross_val_score(NB,binary_matrix,labels,scoring = scorer,cv = 5 )\nprint('Cross validation on Binary matrix by keeping only keywords appearing in at least 5%% of the documents with a mean kappa score of %f and variance of %f' % (scores.mean(),scores.var()))", "Cross validation on Binary matrix by keeping only keywords appearing in at least 5% of the documents with a mean kappa score of 0.876158 and variance of 0.003158\n" ], [ "scores = cross_val_score(NB,count_matrix,labels,scoring = scorer,cv = 5 )\nprint('Cross validation on Count matrix by keeping only keywords appearing in at least 5%% of the documents with a mean kappa score of %f and variance of %f' % (scores.mean(),scores.var()))", "Cross validation on Count matrix by keeping only keywords appearing in at least 5% of the documents with a mean kappa score of 0.951631 and variance of 0.001133\n" ], [ "scores = cross_val_score(NB,tfidf_matrix,labels,scoring = scorer,cv = 5 )\nprint('Cross validation on TF-IDF matrix by keeping only keywords appearing in at least 5%% of the documents with a mean kappa score of %f and variance of %f' % (scores.mean(),scores.var()))", "Cross validation on TF-IDF matrix by keeping only keywords appearing in at least 5% of the documents with a mean kappa score of 0.916734 and variance of 0.001633\n" ] ], [ [ "### Final improvements\nWe've made a big improvement with this one as well. We can even go further and add some extra fine-tunings. Let's have a look at the final key-words:", "_____no_output_____" ] ], [ [ "tfidf_vectorizer.get_feature_names()", "_____no_output_____" ] ], [ [ "We can see that some words all refer to the same thing: *report, reported, reporting, reports* all refer to one same thing *report* and should therefore be grouped together => this can be done by ***stemming***\n### Stemming\nStemming is a technique where we try to reduce words to a common base form, this is done by chopping off the last part of the word: s's are removed, -ing is removed, -ed is removed, ...", "_____no_output_____" ] ], [ [ "# Define a stemmer that will preprocess the text before transforming it\nfrom nltk.stem.porter import PorterStemmer \ndef preprocess(value): \n stemmer = PorterStemmer() \n #split in tokens\n return ' '.join([stemmer.stem(i) for i in Tokenize_text_value(value) ])", "_____no_output_____" ], [ "# Have a look at what it gives on the first article\nprint(' '.join([i for i in Tokenize_text_value(data[0]) ])) # original\nprint('\\n')\nprint(preprocess(data[0])) #stemmed", "farmarel an italian pharmacovigilance project to monitor and evaluate adverse drug reactions in haematologic patients adverse drug reactions adrs reduce patients quality of life increase mortality and morbidity and have a negative economic impact on healthcare systems nevertheless the importance of adr reporting is often underestimated the project farmarel has been developed to monitor and evaluate adrs in haematological patients and to increase pharmacovigilance culture among haematology specialists in haematology units based in lombardy italy a dedicated specialist with the task of encouraging adrs reporting and sensitizing healthcare professionals to pharmacovigilance has been assigned the adrs occurring in haematological patients were collected electronically and then analysed with multiple logistic regression between january and december reports were collected the number of adrs was higher in older adults in male and in non hodgkin lymphoma patients most reactions were severe required or prolonged hospitalization but in most cases they were fully resolved at the time of reporting according to schumock and thornton criteria a percentage of adrs as high as was found to be preventable versus according to reporter opinion patients haematological diagnosis not age or gender resulted to be the variable that most influenced adr in particular severity and outcome the employment of personnel specifically dedicated to pharmacovigilance is a successful strategy to improve the number and quality of adr reports farmarel the first programme of active pharmacovigilance in oncohaematologic patients significantly contributed to reach the who gold standard for pharmacovigilance in lombardy italy\n\n\nfarmarel an italian pharmacovigil project to monitor and evalu advers drug reaction in haematolog patient advers drug reaction adr reduc patient qualiti of life increas mortal and morbid and have a neg econom impact on healthcar system nevertheless the import of adr report is often underestim the project farmarel ha been develop to monitor and evalu adr in haematolog patient and to increas pharmacovigil cultur among haematolog specialist in haematolog unit base in lombardi itali a dedic specialist with the task of encourag adr report and sensit healthcar profession to pharmacovigil ha been assign the adr occur in haematolog patient were collect electron and then analys with multipl logist regress between januari and decemb report were collect the number of adr wa higher in older adult in male and in non hodgkin lymphoma patient most reaction were sever requir or prolong hospit but in most case they were fulli resolv at the time of report accord to schumock and thornton criteria a percentag of adr as high as wa found to be prevent versu accord to report opinion patient haematolog diagnosi not age or gender result to be the variabl that most influenc adr in particular sever and outcom the employ of personnel specif dedic to pharmacovigil is a success strategi to improv the number and qualiti of adr report farmarel the first programm of activ pharmacovigil in oncohaematolog patient significantli contribut to reach the who gold standard for pharmacovigil in lombardi itali\n" ], [ "# Preprocess the documents by stemming the words\nbinary_vectorizer = TfidfVectorizer(input=u'content', analyzer=u'word', binary = True,\n tokenizer=Tokenize_text_value, stop_words = 'english'\n , min_df = 0.05, preprocessor = preprocess) # initialize the binary vectorizer\ncount_vectorizer = TfidfVectorizer(input=u'content', analyzer=u'word', use_idf=False, smooth_idf=False,\n tokenizer=Tokenize_text_value, stop_words = 'english'\n , min_df = 0.05, preprocessor = preprocess) # initialize the count vectorizer\ntfidf_vectorizer = TfidfVectorizer(input=u'content', analyzer=u'word', use_idf=True, smooth_idf=True,\n tokenizer=Tokenize_text_value, stop_words = 'english'\n , min_df = 0.05, preprocessor = preprocess) # initialize the tf-idf vectorizer\nbinary_matrix = binary_vectorizer.fit_transform(data) # fit & transform\ncount_matrix = count_vectorizer.fit_transform(data) # fit & transform\ntfidf_matrix = tfidf_vectorizer.fit_transform(data) # fit & transform", "_____no_output_____" ], [ "scores = cross_val_score(NB,binary_matrix,labels,scoring = scorer,cv = 5 )\nprint('Cross validation on Binary matrix by stemming and keeping only keywords appearing in at least 5%% of the documents with a mean kappa score of %f and variance of %f' % (scores.mean(),scores.var()))", "Cross validation on Binary matrix by stemming and keeping only keywords appearing in at least 5% of the documents with a mean kappa score of 0.868418 and variance of 0.000270\n" ], [ "scores = cross_val_score(NB,count_matrix,labels,scoring = scorer,cv = 5 )\nprint('Cross validation on Count matrix by stemming and keeping only keywords appearing in at least 5%% of the documents with a mean kappa score of %f and variance of %f' % (scores.mean(),scores.var()))", "Cross validation on Count matrix by stemming and keeping only keywords appearing in at least 5% of the documents with a mean kappa score of 0.944600 and variance of 0.001127\n" ], [ "scores = cross_val_score(NB,tfidf_matrix,labels,scoring = scorer,cv = 5 )\nprint('Cross validation on TF-IDF matrix by stemming and keeping only keywords appearing in at least 5%% of the documents with a mean kappa score of %f and variance of %f' % (scores.mean(),scores.var()))", "Cross validation on TF-IDF matrix by stemming and keeping only keywords appearing in at least 5% of the documents with a mean kappa score of 0.901758 and variance of 0.001610\n" ], [ "# Check the stemmed final vocabulary\ntfidf_vectorizer.get_feature_names()", "_____no_output_____" ] ], [ [ "We can see that the performance slightly decreases with the stemming. Probably, because now when we are keeping words that appear in only 5% of the documents, we have more words than before, as before words with different endings were counted separately and now they are grouped together. So to correct for this we should increase our 5% threshold to take this effect into account.", "_____no_output_____" ] ], [ [ "# Preprocess the documents by stemming the words and keeping only words that appear in at least 10% of the documents:\nbinary_vectorizer = TfidfVectorizer(input=u'content', analyzer=u'word', binary = True,\n tokenizer=Tokenize_text_value, stop_words = 'english'\n , min_df = 0.1, preprocessor = preprocess) # initialize the binary vectorizer\ncount_vectorizer = TfidfVectorizer(input=u'content', analyzer=u'word', use_idf=False, smooth_idf=False,\n tokenizer=Tokenize_text_value, stop_words = 'english'\n , min_df = 0.1, preprocessor = preprocess) # initialize the count vectorizer\ntfidf_vectorizer = TfidfVectorizer(input=u'content', analyzer=u'word', use_idf=True, smooth_idf=True,\n tokenizer=Tokenize_text_value, stop_words = 'english'\n , min_df = 0.1, preprocessor = preprocess) # initialize the tf-idf vectorizer\nbinary_matrix = binary_vectorizer.fit_transform(data) # fit & transform\ncount_matrix = count_vectorizer.fit_transform(data) # fit & transform\ntfidf_matrix = tfidf_vectorizer.fit_transform(data) # fit & transform", "_____no_output_____" ], [ "scores = cross_val_score(NB,binary_matrix,labels,scoring = scorer,cv = 5 )\nprint('Cross validation on Binary matrix by stemming with a mean kappa score of %f and variance of %f' % (scores.mean(),scores.var()))", "Cross validation on Binary matrix by stemming with a mean kappa score of 0.893254 and variance of 0.000962\n" ], [ "scores = cross_val_score(NB,count_matrix,labels,scoring = scorer,cv = 5 )\nprint('Cross validation on Count matrix by stemming with a mean kappa score of %f and variance of %f' % (scores.mean(),scores.var()))", "Cross validation on Count matrix by stemming with a mean kappa score of 0.951618 and variance of 0.001134\n" ], [ "scores = cross_val_score(NB,tfidf_matrix,labels,scoring = scorer,cv = 5 )\nprint('Cross validation on TF-IDF matrix by stemming with a mean kappa score of %f and variance of %f' % (scores.mean(),scores.var()))", "Cross validation on TF-IDF matrix by stemming with a mean kappa score of 0.937502 and variance of 0.000565\n" ], [ "# Check the stemmed final vocabulary\ntfidf_vectorizer.get_feature_names()", "_____no_output_____" ] ], [ [ "Now we have the same or a bit higher performance as before.", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown" ]

[ [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code" ], [ "markdown" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "def bin_search_first_2(seq, val):\n \"\"\"\n A binary search returning the index of the first occurrence of a value val within a increasingly ordered sequence\n seq. If val is not part of seq, the index of the first value exceeding val is returned, if any. If all elements of\n seq are smaller than val, the len of seq is returned.\n @param seq: the increasingly ordered sequence to search in\n @param val: the value to search for\n @return: the lowest index r, such that r == len(seq) or seq[r] >= val\n \"\"\"\n lo, hi = -1, len(seq)\n while lo + 1 != hi:\n m = (lo + hi) // 2\n if seq[m] < val:\n lo = m\n else:\n hi = m\n return hi\n", "_____no_output_____" ], [ "def bin_search_first(list, search_val):\n lower, higher = -1, len(list)\n while lower + 1 != higher:\n center = (lower + higher) // 2\n if list[center] < search_val:\n lower = center\n else:\n higher = center\n return higher", "_____no_output_____" ], [ " print(bin_search_first([1, 1, 2, 3, 5, 8, 13, 14, 15, 22, 24, 24, 24, 30, 31], 24))\n print(bin_search_first([1, 1, 2, 3, 5, 8, 13, 14, 15, 22, 24, 24, 24, 30, 31], 0))\n print(bin_search_first([1, 1, 2, 3, 5, 8, 13, 14, 15, 22, 24, 24, 24, 30, 31], 1))\n print(bin_search_first([1, 1, 2, 3, 5, 8, 13, 14, 15, 22, 24, 24, 24, 30, 31], 32))\n print(bin_search_first([1, 1, 2, 3, 5, 8, 13, 14, 15, 22, 24, 24, 24, 30, 31], 31))\n", "10\n0\n0\n15\n14\n" ], [ "def bin_search(lst, search_val, lower, higher):\n if lower + 1 == higher:\n return higher\n\n center = (lower + higher) // 2\n if lst[center] < search_val:\n return bin_search(lst, search_val, center, higher)\n else:\n return bin_search(lst, search_val, lower, center)\n\ndef bin_search_start(lst, search_val):\n return bin_search(lst, search_val, -1, len(lst))", "_____no_output_____" ], [ "print(bin_search_start([1, 1, 2, 3, 5, 8, 13, 14, 15, 22, 24, 24, 24, 30, 31], 24))\nprint(bin_search_start([1, 1, 2, 3, 5, 8, 13, 14, 15, 22, 24, 24, 24, 30, 31], 0))\nprint(bin_search_start([1, 1, 2, 3, 5, 8, 13, 14, 15, 22, 24, 24, 24, 30, 31], 1))\nprint(bin_search_start([1, 1, 2, 3, 5, 8, 13, 14, 15, 22, 24, 24, 24, 30, 31], 32))\nprint(bin_search_start([1, 1, 2, 3, 5, 8, 13, 14, 15, 22, 24, 24, 24, 30, 31], 31))", "10\n0\n0\n15\n14\n" ], [ "import timeit\n", "_____no_output_____" ], [ "%timeit bin_search_start([1, 1, 2, 3, 5, 8, 13, 14, 15, 22, 24, 24, 24, 30, 31], 24)\n", "1.12 µs ± 11 ns per loop (mean ± std. dev. of 7 runs, 1000000 loops each)\n" ], [ "%timeit bin_search_first([1, 1, 2, 3, 5, 8, 13, 14, 15, 22, 24, 24, 24, 30, 31], 24)\n", "719 ns ± 22 ns per loop (mean ± std. dev. of 7 runs, 1000000 loops each)\n" ], [ "def inverse(func, y, lo, hi):\n \"\"\"\n Computes the value of y of the inverse of the monotonously increasing function func.\n\n >>> inverse(lambda x: x ** 2, 13, 0, 13) ** 2\n 13.000000000000002\n >>> 2 ** inverse(lambda x: 2 ** x, 13, 0, 13)\n 13.0\n\n @param func: the function to invert, must be increasing monotonously\n @param y: the value of the function to search for\n @param lo: a lower bound of the result, i.e. a value such that f(lo) < y\n @param hi: an upper bound of the result, i.e. a value such that f(hi) >= y\n @return: the smallest float x, such that f(x) >= y\n \"\"\"\n m = (lo + hi) / 2\n while m != lo and m != hi:\n if func(m) < y:\n lo = m\n else:\n hi = m\n m = (lo + hi) / 2\n return hi\n\n\ndef _test_inverse():\n print(inverse(lambda x: x ** 2, 13, 0, 13) ** 2)\n print(2 ** inverse(lambda x: 2 ** x, 13, 0, 13))\n", "_____no_output_____" ], [ "_test_inverse()", "13.000000000000002\n13.0\n" ], [ "def inverse(func, y,)", "_____no_output_____" ] ] ]

[ "code" ]

[ [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "%load_ext lab_black", "_____no_output_____" ], [ "! pip install smart_open[s3]\n! pip install xgboost\n! pip install pyathena[pandas]\n! pip install nb_black", "Collecting smart_open[s3]\n Downloading smart_open-4.1.2-py3-none-any.whl (111 kB)\n\u001b[K |████████████████████████████████| 111 kB 2.9 MB/s eta 0:00:01\n\u001b[?25hRequirement already satisfied: boto3; extra == \"s3\" in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from smart_open[s3]) (1.9.201)\nRequirement already satisfied: botocore<1.13.0,>=1.12.201 in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from boto3; extra == \"s3\"->smart_open[s3]) (1.12.204)\nRequirement already satisfied: s3transfer<0.3.0,>=0.2.0 in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from boto3; extra == \"s3\"->smart_open[s3]) (0.2.1)\nRequirement already satisfied: jmespath<1.0.0,>=0.7.1 in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from boto3; extra == \"s3\"->smart_open[s3]) (0.9.4)\nRequirement already satisfied: docutils<0.15,>=0.10 in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from botocore<1.13.0,>=1.12.201->boto3; extra == \"s3\"->smart_open[s3]) (0.14)\nRequirement already satisfied: python-dateutil<3.0.0,>=2.1; python_version >= \"2.7\" in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from botocore<1.13.0,>=1.12.201->boto3; extra == \"s3\"->smart_open[s3]) (2.8.0)\nRequirement already satisfied: urllib3<1.26,>=1.20; python_version >= \"3.4\" in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from botocore<1.13.0,>=1.12.201->boto3; extra == \"s3\"->smart_open[s3]) (1.25.3)\nRequirement already satisfied: six>=1.5 in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from python-dateutil<3.0.0,>=2.1; python_version >= \"2.7\"->botocore<1.13.0,>=1.12.201->boto3; extra == \"s3\"->smart_open[s3]) (1.12.0)\nInstalling collected packages: smart-open\nSuccessfully installed smart-open-4.1.2\n\u001b[33mWARNING: You are using pip version 20.1; however, version 20.3.3 is available.\nYou should consider upgrading via the '/usr/local/bin/python3 -m pip install --upgrade pip' command.\u001b[0m\nCollecting xgboost\n Downloading xgboost-1.3.3-py3-none-macosx_10_14_x86_64.macosx_10_15_x86_64.macosx_11_0_x86_64.whl (1.2 MB)\n\u001b[K |████████████████████████████████| 1.2 MB 5.0 MB/s eta 0:00:01\n\u001b[?25hRequirement already satisfied: scipy in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from xgboost) (1.3.1)\nRequirement already satisfied: numpy in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from xgboost) (1.17.0)\nInstalling collected packages: xgboost\nSuccessfully installed xgboost-1.3.3\n\u001b[33mWARNING: You are using pip version 20.1; however, version 20.3.3 is available.\nYou should consider upgrading via the '/usr/local/bin/python3 -m pip install --upgrade pip' command.\u001b[0m\nRequirement already satisfied: pyathena[pandas] in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (1.10.5)\nRequirement already satisfied: tenacity>=4.1.0 in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from pyathena[pandas]) (6.2.0)\nRequirement already satisfied: future in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from pyathena[pandas]) (0.18.2)\nRequirement already satisfied: botocore>=1.5.52 in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from pyathena[pandas]) (1.12.204)\nRequirement already satisfied: boto3>=1.4.4 in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from pyathena[pandas]) (1.9.201)\nRequirement already satisfied: pandas>=0.24.0; extra == \"pandas\" in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from pyathena[pandas]) (0.25.0)\nRequirement already satisfied: pyarrow>=0.15.0; extra == \"pandas\" in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from pyathena[pandas]) (0.15.0)\nRequirement already satisfied: six>=1.9.0 in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from tenacity>=4.1.0->pyathena[pandas]) (1.12.0)\nRequirement already satisfied: jmespath<1.0.0,>=0.7.1 in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from botocore>=1.5.52->pyathena[pandas]) (0.9.4)\nRequirement already satisfied: docutils<0.15,>=0.10 in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from botocore>=1.5.52->pyathena[pandas]) (0.14)\nRequirement already satisfied: python-dateutil<3.0.0,>=2.1; python_version >= \"2.7\" in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from botocore>=1.5.52->pyathena[pandas]) (2.8.0)\nRequirement already satisfied: urllib3<1.26,>=1.20; python_version >= \"3.4\" in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from botocore>=1.5.52->pyathena[pandas]) (1.25.3)\nRequirement already satisfied: s3transfer<0.3.0,>=0.2.0 in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from boto3>=1.4.4->pyathena[pandas]) (0.2.1)\nRequirement already satisfied: pytz>=2017.2 in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from pandas>=0.24.0; extra == \"pandas\"->pyathena[pandas]) (2019.2)\nRequirement already satisfied: numpy>=1.13.3 in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from pandas>=0.24.0; extra == \"pandas\"->pyathena[pandas]) (1.17.0)\n\u001b[33mWARNING: You are using pip version 20.1; however, version 20.3.3 is available.\nYou should consider upgrading via the '/usr/local/bin/python3 -m pip install --upgrade pip' command.\u001b[0m\nRequirement already satisfied: nb_black in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (1.0.7)\nRequirement already satisfied: ipython in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from nb_black) (7.7.0)\nRequirement already satisfied: black>='19.3'; python_version >= \"3.6\" in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from nb_black) (19.3b0)\nRequirement already satisfied: traitlets>=4.2 in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from ipython->nb_black) (4.3.2)\nRequirement already satisfied: appnope; sys_platform == \"darwin\" in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from ipython->nb_black) (0.1.0)\nRequirement already satisfied: pygments in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from ipython->nb_black) (2.4.2)\nRequirement already satisfied: pickleshare in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from ipython->nb_black) (0.7.5)\nRequirement already satisfied: setuptools>=18.5 in /Users/zwl/Library/Python/3.7/lib/python/site-packages (from ipython->nb_black) (46.4.0)\nRequirement already satisfied: jedi>=0.10 in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from ipython->nb_black) (0.14.1)\nRequirement already satisfied: decorator in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from ipython->nb_black) (4.4.0)\nRequirement already satisfied: backcall in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from ipython->nb_black) (0.1.0)\nRequirement already satisfied: pexpect; sys_platform != \"win32\" in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from ipython->nb_black) (4.7.0)\nRequirement already satisfied: prompt-toolkit<2.1.0,>=2.0.0 in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from ipython->nb_black) (2.0.9)\nRequirement already satisfied: click>=6.5 in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from black>='19.3'; python_version >= \"3.6\"->nb_black) (7.0)\nRequirement already satisfied: appdirs in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from black>='19.3'; python_version >= \"3.6\"->nb_black) (1.4.3)\nRequirement already satisfied: attrs>=18.1.0 in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from black>='19.3'; python_version >= \"3.6\"->nb_black) (19.1.0)\nRequirement already satisfied: toml>=0.9.4 in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from black>='19.3'; python_version >= \"3.6\"->nb_black) (0.10.0)\nRequirement already satisfied: ipython-genutils in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from traitlets>=4.2->ipython->nb_black) (0.2.0)\nRequirement already satisfied: six in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from traitlets>=4.2->ipython->nb_black) (1.12.0)\nRequirement already satisfied: parso>=0.5.0 in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from jedi>=0.10->ipython->nb_black) (0.5.1)\nRequirement already satisfied: ptyprocess>=0.5 in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from pexpect; sys_platform != \"win32\"->ipython->nb_black) (0.6.0)\nRequirement already satisfied: wcwidth in /Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages (from prompt-toolkit<2.1.0,>=2.0.0->ipython->nb_black) (0.1.7)\n\u001b[33mWARNING: You are using pip version 20.1; however, version 20.3.3 is available.\nYou should consider upgrading via the '/usr/local/bin/python3 -m pip install --upgrade pip' command.\u001b[0m\n" ] ] ]

[ "code" ]

[ [ "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Threads are useful for IO bound processes. Speed Up the processing.\nimport time\nimport threading\n\nstart = time.perf_counter()\n\n\ndef do_something():\n print(f'Sleeping 1 Second...')\n time.sleep(1)\n print(f'Done Sleeping...')\n\n \nt1 = threading.Thread(target=do_something)\nt2 = threading.Thread(target=do_something)\n\nt1.start()\nt2.start()\n\nt1.join()\nt2.join() # Means that complete the task before moving ahead\n\nfinish = time.perf_counter()\n\nprint(f'Finished in {round(finish-start, 2)} second(s)')", "Sleeping 1 Second...\nSleeping 1 Second...\nDone Sleeping...\nDone Sleeping...\nFinished in 1.01 second(s)\n" ], [ "import time\nimport threading\n\nstart = time.perf_counter()\n\n\ndef do_something(seconds):\n print(f'Sleeping {seconds} Second(s)...')\n time.sleep(seconds)\n print(f'Done Sleeping...')\n\nthreads = []\n \nfor _ in range(10):\n t = threading.Thread(target=do_something, args=[1.5])\n t.start()\n threads.append(t)\n \nfor thread in threads:\n thread.join()\n\nfinish = time.perf_counter()\n\nprint(f'Finished in {round(finish-start, 2)} second(s)')", "Sleeping 1.5 Second(s)...\nSleeping 1.5 Second(s)...\nSleeping 1.5 Second(s)...\nSleeping 1.5 Second(s)...\nSleeping 1.5 Second(s)...\nSleeping 1.5 Second(s)...\nSleeping 1.5 Second(s)...\nSleeping 1.5 Second(s)...\nSleeping 1.5 Second(s)...\nSleeping 1.5 Second(s)...\nDone Sleeping...\nDone Sleeping...\nDone Sleeping...\nDone Sleeping...\nDone Sleeping...\nDone Sleeping...\nDone Sleeping...\nDone Sleeping...\nDone Sleeping...\nDone Sleeping...\nFinished in 1.53 second(s)\n" ], [ "import time\nfrom concurrent.futures import ThreadPoolExecutor as executor\nfrom concurrent.futures import as_completed\n\nstart = time.perf_counter()\n\n\ndef do_something(seconds):\n print(f'Sleeping {seconds} Second(s)...')\n time.sleep(seconds)\n return f'Done Sleeping...{seconds}'\n\n\nwith executor():\n secs = [5,4,3,2,1]\n results = [executor().submit(do_something, sec) for sec in secs]\n \n for f in as_completed(results):\n print(f.result())\n\n\nfinish = time.perf_counter()\n\nprint(f'Finished in {round(finish-start, 2)} second(s)')", "Sleeping 5 Second(s)...Sleeping 4 Second(s)...\n\nSleeping 3 Second(s)...\nSleeping 2 Second(s)...\nSleeping 1 Second(s)...\nDone Sleeping...1\nDone Sleeping...2\nDone Sleeping...3\nDone Sleeping...4\nDone Sleeping...5\nFinished in 5.01 second(s)\n" ], [ "import time\nfrom concurrent.futures import ThreadPoolExecutor as executor\nfrom concurrent.futures import as_completed\n\nstart = time.perf_counter()\n\ndef do_something(seconds):\n print(f'Sleeping {seconds} Second(s)...')\n time.sleep(seconds)\n return f'Done Sleeping...{seconds}'\n\n\nwith executor():\n secs = [5,4,3,2,1]\n # map returns the results in the order they were started\n results = executor().map(do_something, secs)\n \n for result in results:\n print(result)\n\n\nfinish = time.perf_counter()\n\nprint(f'Finished in {round(finish-start, 2)} second(s)')", "Sleeping 5 Second(s)...\nSleeping 4 Second(s)...\nSleeping 3 Second(s)...\nSleeping 2 Second(s)...\nSleeping 1 Second(s)...\nDone Sleeping...5\nDone Sleeping...4\nDone Sleeping...3\nDone Sleeping...2\nDone Sleeping...1\nFinished in 5.01 second(s)\n" ], [ "# Real Example\nimport requests\nimport time\nimport concurrent.futures\n\nimg_urls = [\n 'https://images.unsplash.com/photo-1516117172878-fd2c41f4a759',\n 'https://images.unsplash.com/photo-1532009324734-20a7a5813719',\n 'https://images.unsplash.com/photo-1524429656589-6633a470097c',\n 'https://images.unsplash.com/photo-1513938709626-033611b8cc03',\n 'https://images.unsplash.com/photo-1530224264768-7ff8c1789d79',\n]\n\nt1 = time.perf_counter()\n\n\nfor img_url in img_urls:\n img_bytes = requests.get(img_url).content\n img_name = img_url.split('/')[3]\n img_name = f'{img_name}.jpg'\n with open(img_name, 'wb') as img_file:\n img_file.write(img_bytes)\n print(f'{img_name} was downloaded...')\n \nt2 = time.perf_counter()\n\nprint(f'Finished in {t2-t1} seconds')", "photo-1516117172878-fd2c41f4a759.jpg was downloaded...\nphoto-1532009324734-20a7a5813719.jpg was downloaded...\nphoto-1524429656589-6633a470097c.jpg was downloaded...\nphoto-1513938709626-033611b8cc03.jpg was downloaded...\nphoto-1530224264768-7ff8c1789d79.jpg was downloaded...\nFinished in 82.39902849999999 seconds\n" ], [ "# Real Example with threads\nimg_urls = [\n 'https://images.unsplash.com/photo-1516117172878-fd2c41f4a759',\n 'https://images.unsplash.com/photo-1532009324734-20a7a5813719',\n 'https://images.unsplash.com/photo-1524429656589-6633a470097c',\n 'https://images.unsplash.com/photo-1513938709626-033611b8cc03',\n 'https://images.unsplash.com/photo-1530224264768-7ff8c1789d79',\n]\n\nt1 = time.perf_counter()\n\n\ndef download_image(img_url):\n img_bytes = requests.get(img_url).content\n img_name = img_url.split('/')[3]\n img_name = f'{img_name}.jpg'\n with open(img_name, 'wb') as img_file:\n img_file.write(img_bytes)\n print(f'{img_name} was downloaded...')\n \nwith concurrent.futures.ThreadPoolExecutor() as executor:\n executor.map(download_image, img_urls)\n \nt2 = time.perf_counter()\n\nprint(f'Finished in {t2-t1} seconds')", "photo-1516117172878-fd2c41f4a759.jpg was downloaded...\nphoto-1532009324734-20a7a5813719.jpg was downloaded...\nphoto-1524429656589-6633a470097c.jpg was downloaded...\nphoto-1513938709626-033611b8cc03.jpg was downloaded...\nphoto-1530224264768-7ff8c1789d79.jpg was downloaded...\nFinished in 37.34158899999966 seconds\n" ] ] ]

[ "code" ]

[ [ "code", "code", "code", "code", "code", "code" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "This notebook was prepared by [Donne Martin](https://github.com/donnemartin). Source and license info is on [GitHub](https://github.com/donnemartin/interactive-coding-challenges).", "_____no_output_____" ], [ "# Challenge Notebook", "_____no_output_____" ], [ "## Problem: Implement a priority queue backed by an array.\n\n* [Constraints](#Constraints)\n* [Test Cases](#Test-Cases)\n* [Algorithm](#Algorithm)\n* [Code](#Code)\n* [Unit Test](#Unit-Test)\n* [Solution Notebook](#Solution-Notebook)", "_____no_output_____" ], [ "## Constraints\n\n* Do we expect the methods to be insert, extract_min, and decrease_key?\n * Yes\n* Can we assume there aren't any duplicate keys?\n * Yes\n* Do we need to validate inputs?\n * No\n* Can we assume this fits memory?\n * Yes", "_____no_output_____" ], [ "## Test Cases\n\n### insert\n\n* `insert` general case -> inserted node\n\n### extract_min\n\n* `extract_min` from an empty list -> None\n* `extract_min` general case -> min node\n\n### decrease_key\n\n* `decrease_key` an invalid key -> None\n* `decrease_key` general case -> updated node", "_____no_output_____" ], [ "## Algorithm\n\nRefer to the [Solution Notebook](priority_queue_solution.ipynb). If you are stuck and need a hint, the solution notebook's algorithm discussion might be a good place to start.", "_____no_output_____" ], [ "## Code", "_____no_output_____" ] ], [ [ "class PriorityQueueNode(object):\n\n def __init__(self, obj, key):\n self.obj = obj\n self.key = key\n\n def __repr__(self):\n return str(self.obj) + ': ' + str(self.key)\n\n\nclass PriorityQueue(object):\n\n def __init__(self):\n self.array = []\n\n def __len__(self):\n return len(self.array)\n\n def insert(self, node):\n for k in range(len(self.array)):\n if self.array[k].key < node.key:\n \n # TODO: Implement me\n pass\n\n def extract_min(self):\n if not self.array:\n return None\n else:\n \n # TODO: Implement me\n pass\n\n def decrease_key(self, obj, new_key):\n # TODO: Implement me\n pass", "_____no_output_____" ] ], [ [ "## Unit Test", "_____no_output_____" ], [ "**The following unit test is expected to fail until you solve the challenge.**", "_____no_output_____" ] ], [ [ "# %load test_priority_queue.py\nimport unittest\n\n\nclass TestPriorityQueue(unittest.TestCase):\n\n def test_priority_queue(self):\n priority_queue = PriorityQueue()\n self.assertEqual(priority_queue.extract_min(), None)\n priority_queue.insert(PriorityQueueNode('a', 20))\n priority_queue.insert(PriorityQueueNode('b', 5))\n priority_queue.insert(PriorityQueueNode('c', 15))\n priority_queue.insert(PriorityQueueNode('d', 22))\n priority_queue.insert(PriorityQueueNode('e', 40))\n priority_queue.insert(PriorityQueueNode('f', 3))\n priority_queue.decrease_key('f', 2)\n priority_queue.decrease_key('a', 19)\n mins = []\n while priority_queue.array:\n mins.append(priority_queue.extract_min().key)\n self.assertEqual(mins, [2, 5, 15, 19, 22, 40])\n print('Success: test_min_heap')\n\n\ndef main():\n test = TestPriorityQueue()\n test.test_priority_queue()\n\n\nif __name__ == '__main__':\n main()", "_____no_output_____" ] ], [ [ "## Solution Notebook\n\nReview the [Solution Notebook](priority_queue_solution.ipynb) for a discussion on algorithms and code solutions.", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown" ]

[ [ "markdown", "markdown", "markdown", "markdown", "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# pIC50 Test", "_____no_output_____" ] ], [ [ "import numpy as np\nimport torch\nimport seaborn as sns\nimport malt\nimport pandas as pd\nimport dgllife", "Using backend: pytorch\n" ], [ "from dgllife.utils import smiles_to_bigraph, CanonicalAtomFeaturizer, CanonicalBondFeaturizer\ndf = pd.read_csv('../../../data/moonshot_pIC50.csv', index_col=0)\n\ndgllife_dataset = dgllife.data.csv_dataset.MoleculeCSVDataset(\n df=df,\n smiles_to_graph=smiles_to_bigraph,\n node_featurizer=CanonicalAtomFeaturizer(),\n edge_featurizer=CanonicalBondFeaturizer(),\n smiles_column='SMILES',\n task_names=[\n # 'MW', 'cLogP', 'r_inhibition_at_20_uM',\n # 'r_inhibition_at_50_uM', 'r_avg_IC50', 'f_inhibition_at_20_uM',\n # 'f_inhibition_at_50_uM', 'f_avg_IC50', 'relative_solubility_at_20_uM',\n # 'relative_solubility_at_100_uM', 'trypsin_IC50',\n 'f_avg_pIC50'\n ],\n init_mask=False,\n cache_file_path='../../../data/moonshot_pIC50.bin'\n)", "Processing dgl graphs from scratch...\nProcessing molecule 1000/2260\nProcessing molecule 2000/2260\n" ], [ "sns.displot(dgllife_dataset.labels.numpy()[dgllife_dataset.labels.numpy() > 4.005])", "_____no_output_____" ], [ "from malt.data.collections import _dataset_from_dgllife\n\ndata = _dataset_from_dgllife(dgllife_dataset)\n\n# mask data if it's at the limit of detection\ndata_masked = data[list(np.flatnonzero(np.array(data.y) > 4.005))]\ndata.shuffle(seed=2666)\nds_tr, ds_vl, ds_te = data_masked[:1500].split([8, 1, 1])", "_____no_output_____" ] ], [ [ "Make model", "_____no_output_____" ] ], [ [ "model_choice = 'nn' # 'nn'\nif model_choice == \"gp\":\n model = malt.models.supervised_model.GaussianProcessSupervisedModel(\n representation=malt.models.representation.DGLRepresentation(\n out_features=128,\n ),\n regressor=malt.models.regressor.ExactGaussianProcessRegressor(\n in_features=128, out_features=2,\n ),\n likelihood=malt.models.likelihood.HeteroschedasticGaussianLikelihood(),\n )\n\n\nelif model_choice == \"nn\":\n model = malt.models.supervised_model.SimpleSupervisedModel(\n representation=malt.models.representation.DGLRepresentation(\n out_features=128,\n ),\n regressor=malt.models.regressor.NeuralNetworkRegressor(\n in_features=128, out_features=1,\n ),\n likelihood=malt.models.likelihood.HomoschedasticGaussianLikelihood(),\n )", "_____no_output_____" ] ], [ [ "Train and evaluate.", "_____no_output_____" ] ], [ [ "trainer = malt.trainer.get_default_trainer(\n without_player=True,\n batch_size=32,\n n_epochs=3000,\n learning_rate=1e-3\n)\nmodel = trainer(model, ds_tr)\n\nr2 = malt.metrics.supervised_metrics.R2()(model, ds_te)\nprint(r2)\n\nrmse = malt.metrics.supervised_metrics.RMSE()(model, ds_te)\nprint(rmse)", "_____no_output_____" ], [ "ds_te_loader = ds_te.view(batch_size=len(ds_te))\ng, y = next(iter(ds_te_loader))\ny_hat = model.condition(g).mean\ng = sns.jointplot(x = ds_te.y, y = y_hat.detach().numpy())\ng.set_axis_labels('y', '\\hat{y}')", "_____no_output_____" ], [ "import torch\nimport dgl\nimport malt\nimport argparse\n\nparser = argparse.ArgumentParser()\nparser.add_argument(\"--data\", type=str, default=\"esol\")\nparser.add_argument(\"--model\", type=str, default=\"nn\")\n\nargs = parser.parse_args([])\n\ndata = getattr(malt.data.collections, args.data)()\ndata.shuffle(seed=2666)\nds_tr, ds_vl, ds_te = data.split([8, 1, 1])\n\nif args.model == \"gp\":\n model = malt.models.supervised_model.GaussianProcessSupervisedModel(\n representation=malt.models.representation.DGLRepresentation(\n out_features=128,\n ),\n regressor=malt.models.regressor.ExactGaussianProcessRegressor(\n in_features=128, out_features=2,\n ),\n likelihood=malt.models.likelihood.HeteroschedasticGaussianLikelihood(),\n )\n\n\nelif args.model == \"nn\":\n model = malt.models.supervised_model.SimpleSupervisedModel(\n representation=malt.models.representation.DGLRepresentation(\n out_features=128,\n ),\n regressor=malt.models.regressor.NeuralNetworkRegressor(\n in_features=128, out_features=1,\n ),\n likelihood=malt.models.likelihood.HomoschedasticGaussianLikelihood(),\n )\n\n\ntrainer = malt.trainer.get_default_trainer(without_player=True, batch_size=len(ds_tr), n_epochs=3000, learning_rate=1e-3)\nmodel = trainer(model, ds_tr)\n\nr2 = malt.metrics.supervised_metrics.R2()(model, ds_te)\nprint(r2)\n\nrmse = malt.metrics.supervised_metrics.RMSE()(model, ds_te)\nprint(rmse)", "Processing dgl graphs from scratch...\nProcessing molecule 1000/1128\ntensor(0.9217, grad_fn=<RsubBackward1>)\ntensor(2.8730, grad_fn=<SqrtBackward0>)\n" ], [ "ds_te_loader = ds_te.view(batch_size=len(ds_te))\ng, y = next(iter(ds_te_loader))\ny_hat = model.condition(g).mean\ng = sns.jointplot(x = ds_te.y, y = y_hat.detach().numpy())\ng.set_axis_labels('y', '\\hat{y}')", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown" ], [ "code", "code", "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Inference and Validation\n\nNow that you have a trained network, you can use it for making predictions. This is typically called **inference**, a term borrowed from statistics. However, neural networks have a tendency to perform *too well* on the training data and aren't able to generalize to data that hasn't been seen before. This is called **overfitting** and it impairs inference performance. To test for overfitting while training, we measure the performance on data not in the training set called the **validation** set. We avoid overfitting through regularization such as dropout while monitoring the validation performance during training. In this notebook, I'll show you how to do this in PyTorch. \n\nAs usual, let's start by loading the dataset through torchvision. You'll learn more about torchvision and loading data in a later part. This time we'll be taking advantage of the test set which you can get by setting `train=False` here:\n\n```python\ntestset = datasets.FashionMNIST('~/.pytorch/F_MNIST_data/', download=True, train=False, transform=transform)\n```\n\nThe test set contains images just like the training set. Typically you'll see 10-20% of the original dataset held out for testing and validation with the rest being used for training.", "_____no_output_____" ] ], [ [ "import torch\nfrom torchvision import datasets, transforms\n\n# Define a transform to normalize the data\ntransform = transforms.Compose([transforms.ToTensor(),\n transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])\n# Download and load the training data\ntrainset = datasets.FashionMNIST('~/.pytorch/F_MNIST_data/', download=True, train=True, transform=transform)\ntrainloader = torch.utils.data.DataLoader(trainset, batch_size=64, shuffle=True)\n\n# Download and load the test data\ntestset = datasets.FashionMNIST('~/.pytorch/F_MNIST_data/', download=True, train=False, transform=transform)\ntestloader = torch.utils.data.DataLoader(testset, batch_size=64, shuffle=True)", "_____no_output_____" ] ], [ [ "Here I'll create a model like normal, using the same one from my solution for part 4.", "_____no_output_____" ] ], [ [ "from torch import nn, optim\nimport torch.nn.functional as F\n\nclass Classifier(nn.Module):\n def __init__(self):\n super().__init__()\n self.fc1 = nn.Linear(784, 256)\n self.fc2 = nn.Linear(256, 128)\n self.fc3 = nn.Linear(128, 64)\n self.fc4 = nn.Linear(64, 10)\n \n def forward(self, x):\n # make sure input tensor is flattened\n x = x.view(x.shape[0], -1)\n \n x = F.relu(self.fc1(x))\n x = F.relu(self.fc2(x))\n x = F.relu(self.fc3(x))\n x = F.log_softmax(self.fc4(x), dim=1)\n \n return x", "_____no_output_____" ] ], [ [ "The goal of validation is to measure the model's performance on data that isn't part of the training set. Performance here is up to the developer to define though. Typically this is just accuracy, the percentage of classes the network predicted correctly. Other options are [precision and recall](https://en.wikipedia.org/wiki/Precision_and_recall#Definition_(classification_context)) and top-5 error rate. We'll focus on accuracy here. First I'll do a forward pass with one batch from the test set.", "_____no_output_____" ] ], [ [ "model = Classifier()\n\nimages, labels = next(iter(testloader))\n# Get the class probabilities\nps = torch.exp(model(images))\n# Make sure the shape is appropriate, we should get 10 class probabilities for 64 examples\nprint(ps.shape)", "torch.Size([64, 10])\n" ] ], [ [ "With the probabilities, we can get the most likely class using the `ps.topk` method. This returns the $k$ highest values. Since we just want the most likely class, we can use `ps.topk(1)`. This returns a tuple of the top-$k$ values and the top-$k$ indices. If the highest value is the fifth element, we'll get back 4 as the index.", "_____no_output_____" ] ], [ [ "top_p, top_class = ps.topk(1, dim=1)\n# Look at the most likely classes for the first 10 examples\nprint(top_class[:10,:])", "tensor([[0],\n [3],\n [0],\n [0],\n [0],\n [0],\n [0],\n [0],\n [0],\n [0]])\n" ] ], [ [ "Now we can check if the predicted classes match the labels. This is simple to do by equating `top_class` and `labels`, but we have to be careful of the shapes. Here `top_class` is a 2D tensor with shape `(64, 1)` while `labels` is 1D with shape `(64)`. To get the equality to work out the way we want, `top_class` and `labels` must have the same shape.\n\nIf we do\n\n```python\nequals = top_class == labels\n```\n\n`equals` will have shape `(64, 64)`, try it yourself. What it's doing is comparing the one element in each row of `top_class` with each element in `labels` which returns 64 True/False boolean values for each row.", "_____no_output_____" ] ], [ [ "equals = top_class == labels.view(*top_class.shape)", "_____no_output_____" ] ], [ [ "Now we need to calculate the percentage of correct predictions. `equals` has binary values, either 0 or 1. This means that if we just sum up all the values and divide by the number of values, we get the percentage of correct predictions. This is the same operation as taking the mean, so we can get the accuracy with a call to `torch.mean`. If only it was that simple. If you try `torch.mean(equals)`, you'll get an error\n\n```\nRuntimeError: mean is not implemented for type torch.ByteTensor\n```\n\nThis happens because `equals` has type `torch.ByteTensor` but `torch.mean` isn't implement for tensors with that type. So we'll need to convert `equals` to a float tensor. Note that when we take `torch.mean` it returns a scalar tensor, to get the actual value as a float we'll need to do `accuracy.item()`.", "_____no_output_____" ] ], [ [ "accuracy = torch.mean(equals.type(torch.FloatTensor))\nprint(f'Accuracy: {accuracy.item()*100}%')", "Accuracy: 20.3125%\n" ] ], [ [ "The network is untrained so it's making random guesses and we should see an accuracy around 10%. Now let's train our network and include our validation pass so we can measure how well the network is performing on the test set. Since we're not updating our parameters in the validation pass, we can speed up our code by turning off gradients using `torch.no_grad()`:\n\n```python\n# turn off gradients\nwith torch.no_grad():\n # validation pass here\n for images, labels in testloader:\n ...\n```\n\n>**Exercise:** Implement the validation loop below and print out the total accuracy after the loop. You can largely copy and paste the code from above, but I suggest typing it in because writing it out yourself is essential for building the skill. In general you'll always learn more by typing it rather than copy-pasting. You should be able to get an accuracy above 80%.", "_____no_output_____" ] ], [ [ "model = Classifier()\ncriterion = nn.NLLLoss()\noptimizer = optim.Adam(model.parameters(), lr=0.003)\n\nepochs = 30\nsteps = 0\n\ntrain_losses, test_losses = [], []\nfor e in range(epochs):\n running_loss = 0\n for images, labels in trainloader:\n \n optimizer.zero_grad()\n \n log_ps = model(images)\n loss = criterion(log_ps, labels)\n loss.backward()\n optimizer.step()\n \n running_loss += loss.item()\n \n else:\n ## TODO: Implement the validation pass and print out the validation accuracy\n test_loss = 0\n accuracy = 0\n with torch.no_grad():\n for images, labels in testloader:\n log_ps = model(images)\n test_loss += criterion(log_ps, labels)\n \n ps = torch.exp(log_ps)\n top_p, top_class = ps.topk(1, dim=1)\n equals = top_class == labels.view(*top_class.shape)\n accuracy += torch.mean(equals.type(torch.FloatTensor))\n \n train_losses.append(running_loss/len(trainloader))\n test_losses.append(test_loss/len(testloader))\n \n \n print(\"Epoch: {}/{}.. \".format(e+1, epochs),\n \"Training Loss: {:.3f}.. \".format(running_loss/len(trainloader)),\n \"Test Loss: {:.3f}.. \".format(test_loss/len(testloader)),\n \"Test Accuracy: {:.3f}\".format(accuracy/len(testloader)))\n ", "Epoch: 1/30.. Training Loss: 0.511.. Test Loss: 0.455.. Test Accuracy: 0.839\nEpoch: 2/30.. Training Loss: 0.392.. Test Loss: 0.411.. Test Accuracy: 0.846\nEpoch: 3/30.. Training Loss: 0.353.. Test Loss: 0.403.. Test Accuracy: 0.853\nEpoch: 4/30.. Training Loss: 0.333.. Test Loss: 0.376.. Test Accuracy: 0.867\nEpoch: 5/30.. Training Loss: 0.318.. Test Loss: 0.392.. Test Accuracy: 0.857\nEpoch: 6/30.. Training Loss: 0.303.. Test Loss: 0.388.. Test Accuracy: 0.864\nEpoch: 7/30.. Training Loss: 0.295.. Test Loss: 0.363.. Test Accuracy: 0.871\nEpoch: 8/30.. Training Loss: 0.282.. Test Loss: 0.352.. Test Accuracy: 0.877\nEpoch: 9/30.. Training Loss: 0.279.. Test Loss: 0.396.. Test Accuracy: 0.863\nEpoch: 10/30.. Training Loss: 0.269.. Test Loss: 0.363.. Test Accuracy: 0.874\nEpoch: 11/30.. Training Loss: 0.260.. Test Loss: 0.385.. Test Accuracy: 0.869\nEpoch: 12/30.. Training Loss: 0.249.. Test Loss: 0.382.. Test Accuracy: 0.879\nEpoch: 13/30.. Training Loss: 0.247.. Test Loss: 0.376.. Test Accuracy: 0.876\nEpoch: 14/30.. Training Loss: 0.244.. Test Loss: 0.381.. Test Accuracy: 0.878\nEpoch: 15/30.. Training Loss: 0.235.. Test Loss: 0.369.. Test Accuracy: 0.880\nEpoch: 16/30.. Training Loss: 0.228.. Test Loss: 0.389.. Test Accuracy: 0.877\nEpoch: 17/30.. Training Loss: 0.232.. Test Loss: 0.380.. Test Accuracy: 0.877\nEpoch: 18/30.. Training Loss: 0.220.. Test Loss: 0.379.. Test Accuracy: 0.880\nEpoch: 19/30.. Training Loss: 0.217.. Test Loss: 0.383.. Test Accuracy: 0.883\nEpoch: 20/30.. Training Loss: 0.214.. Test Loss: 0.371.. Test Accuracy: 0.880\nEpoch: 21/30.. Training Loss: 0.211.. Test Loss: 0.423.. Test Accuracy: 0.873\nEpoch: 22/30.. Training Loss: 0.207.. Test Loss: 0.401.. Test Accuracy: 0.885\nEpoch: 23/30.. Training Loss: 0.203.. Test Loss: 0.414.. Test Accuracy: 0.876\nEpoch: 24/30.. Training Loss: 0.203.. Test Loss: 0.401.. Test Accuracy: 0.880\nEpoch: 25/30.. Training Loss: 0.200.. Test Loss: 0.442.. Test Accuracy: 0.874\nEpoch: 26/30.. Training Loss: 0.196.. Test Loss: 0.393.. Test Accuracy: 0.886\nEpoch: 27/30.. Training Loss: 0.190.. Test Loss: 0.421.. Test Accuracy: 0.883\nEpoch: 28/30.. Training Loss: 0.187.. Test Loss: 0.408.. Test Accuracy: 0.878\nEpoch: 29/30.. Training Loss: 0.185.. Test Loss: 0.423.. Test Accuracy: 0.880\nEpoch: 30/30.. Training Loss: 0.180.. Test Loss: 0.466.. Test Accuracy: 0.874\n" ] ], [ [ "## Overfitting\n\nIf we look at the training and validation losses as we train the network, we can see a phenomenon known as overfitting.\n\n<img src='assets/overfitting.png' width=450px>\n\nThe network learns the training set better and better, resulting in lower training losses. However, it starts having problems generalizing to data outside the training set leading to the validation loss increasing. The ultimate goal of any deep learning model is to make predictions on new data, so we should strive to get the lowest validation loss possible. One option is to use the version of the model with the lowest validation loss, here the one around 8-10 training epochs. This strategy is called *early-stopping*. In practice, you'd save the model frequently as you're training then later choose the model with the lowest validation loss.\n\nThe most common method to reduce overfitting (outside of early-stopping) is *dropout*, where we randomly drop input units. This forces the network to share information between weights, increasing it's ability to generalize to new data. Adding dropout in PyTorch is straightforward using the [`nn.Dropout`](https://pytorch.org/docs/stable/nn.html#torch.nn.Dropout) module.\n\n```python\nclass Classifier(nn.Module):\n def __init__(self):\n super().__init__()\n self.fc1 = nn.Linear(784, 256)\n self.fc2 = nn.Linear(256, 128)\n self.fc3 = nn.Linear(128, 64)\n self.fc4 = nn.Linear(64, 10)\n \n # Dropout module with 0.2 drop probability\n self.dropout = nn.Dropout(p=0.2)\n \n def forward(self, x):\n # make sure input tensor is flattened\n x = x.view(x.shape[0], -1)\n \n # Now with dropout\n x = self.dropout(F.relu(self.fc1(x)))\n x = self.dropout(F.relu(self.fc2(x)))\n x = self.dropout(F.relu(self.fc3(x)))\n \n # output so no dropout here\n x = F.log_softmax(self.fc4(x), dim=1)\n \n return x\n```\n\nDuring training we want to use dropout to prevent overfitting, but during inference we want to use the entire network. So, we need to turn off dropout during validation, testing, and whenever we're using the network to make predictions. To do this, you use `model.eval()`. This sets the model to evaluation mode where the dropout probability is 0. You can turn dropout back on by setting the model to train mode with `model.train()`. In general, the pattern for the validation loop will look like this, where you turn off gradients, set the model to evaluation mode, calculate the validation loss and metric, then set the model back to train mode.\n\n```python\n# turn off gradients\nwith torch.no_grad():\n \n # set model to evaluation mode\n model.eval()\n \n # validation pass here\n for images, labels in testloader:\n ...\n\n# set model back to train mode\nmodel.train()\n```", "_____no_output_____" ], [ "> **Exercise:** Add dropout to your model and train it on Fashion-MNIST again. See if you can get a lower validation loss or higher accuracy.", "_____no_output_____" ] ], [ [ "## TODO: Define your model with dropout added\nclass Classifier(nn.Module):\n def __init__(self):\n super().__init__()\n self.fc1 = nn.Linear(784, 256)\n self.fc2 = nn.Linear(256, 128)\n self.fc3 = nn.Linear(128, 64)\n self.fc4 = nn.Linear(64, 10)\n \n self.dropout = nn.Dropout(p=0.2)\n \n def forward(self,x):\n x = x.view(x.shape[0], -1)\n\n # Now with dropout\n x = self.dropout(F.relu(self.fc1(x)))\n x = self.dropout(F.relu(self.fc2(x)))\n x = self.dropout(F.relu(self.fc3(x)))\n\n # output so no dropout here\n x = F.log_softmax(self.fc4(x), dim=1)\n\n return x\n ", "_____no_output_____" ], [ "model = Classifier()\ncriterion = nn.NLLLoss()\noptimizer = optim.Adam(model.parameters(), lr=0.003)\n\nepochs = 30\nsteps = 0\n\ntrain_losses, test_losses = [], []\nfor e in range(epochs):\n running_loss = 0\n for images, labels in trainloader:\n \n optimizer.zero_grad()\n \n log_ps = model(images)\n loss = criterion(log_ps, labels)\n loss.backward()\n optimizer.step()\n \n running_loss += loss.item()\n \n else:\n test_loss = 0\n accuracy = 0\n \n # Turn off gradients for validation, saves memory and computations\n with torch.no_grad():\n model.eval()\n for images, labels in testloader:\n log_ps = model(images)\n test_loss += criterion(log_ps, labels)\n \n ps = torch.exp(log_ps)\n top_p, top_class = ps.topk(1, dim=1)\n equals = top_class == labels.view(*top_class.shape)\n accuracy += torch.mean(equals.type(torch.FloatTensor))\n \n model.train()\n \n train_losses.append(running_loss/len(trainloader))\n test_losses.append(test_loss/len(testloader))\n\n print(\"Epoch: {}/{}.. \".format(e+1, epochs),\n \"Training Loss: {:.3f}.. \".format(running_loss/len(trainloader)),\n \"Test Loss: {:.3f}.. \".format(test_loss/len(testloader)),\n \"Test Accuracy: {:.3f}\".format(accuracy/len(testloader)))", "Epoch: 1/30.. Training Loss: 0.611.. Test Loss: 0.471.. Test Accuracy: 0.830\nEpoch: 2/30.. Training Loss: 0.482.. Test Loss: 0.439.. Test Accuracy: 0.841\nEpoch: 3/30.. Training Loss: 0.449.. Test Loss: 0.406.. Test Accuracy: 0.852\nEpoch: 4/30.. Training Loss: 0.433.. Test Loss: 0.438.. Test Accuracy: 0.841\nEpoch: 5/30.. Training Loss: 0.421.. Test Loss: 0.409.. Test Accuracy: 0.852\nEpoch: 6/30.. Training Loss: 0.411.. Test Loss: 0.416.. Test Accuracy: 0.845\nEpoch: 7/30.. Training Loss: 0.406.. Test Loss: 0.401.. Test Accuracy: 0.856\nEpoch: 8/30.. Training Loss: 0.399.. Test Loss: 0.406.. Test Accuracy: 0.850\nEpoch: 9/30.. Training Loss: 0.399.. Test Loss: 0.412.. Test Accuracy: 0.855\nEpoch: 10/30.. Training Loss: 0.389.. Test Loss: 0.394.. Test Accuracy: 0.864\nEpoch: 11/30.. Training Loss: 0.385.. Test Loss: 0.378.. Test Accuracy: 0.869\nEpoch: 12/30.. Training Loss: 0.382.. Test Loss: 0.385.. Test Accuracy: 0.866\nEpoch: 13/30.. Training Loss: 0.380.. Test Loss: 0.384.. Test Accuracy: 0.867\nEpoch: 14/30.. Training Loss: 0.374.. Test Loss: 0.377.. Test Accuracy: 0.867\nEpoch: 15/30.. Training Loss: 0.373.. Test Loss: 0.382.. Test Accuracy: 0.869\nEpoch: 16/30.. Training Loss: 0.374.. Test Loss: 0.372.. Test Accuracy: 0.872\nEpoch: 17/30.. Training Loss: 0.372.. Test Loss: 0.384.. Test Accuracy: 0.870\nEpoch: 18/30.. Training Loss: 0.370.. Test Loss: 0.389.. Test Accuracy: 0.869\nEpoch: 19/30.. Training Loss: 0.358.. Test Loss: 0.380.. Test Accuracy: 0.871\nEpoch: 20/30.. Training Loss: 0.357.. Test Loss: 0.384.. Test Accuracy: 0.871\nEpoch: 21/30.. Training Loss: 0.357.. Test Loss: 0.388.. Test Accuracy: 0.865\nEpoch: 22/30.. Training Loss: 0.353.. Test Loss: 0.388.. Test Accuracy: 0.875\nEpoch: 23/30.. Training Loss: 0.362.. Test Loss: 0.373.. Test Accuracy: 0.874\nEpoch: 24/30.. Training Loss: 0.346.. Test Loss: 0.383.. Test Accuracy: 0.873\nEpoch: 25/30.. Training Loss: 0.347.. Test Loss: 0.382.. Test Accuracy: 0.871\nEpoch: 26/30.. Training Loss: 0.343.. Test Loss: 0.390.. Test Accuracy: 0.867\nEpoch: 27/30.. Training Loss: 0.344.. Test Loss: 0.386.. Test Accuracy: 0.868\nEpoch: 28/30.. Training Loss: 0.342.. Test Loss: 0.392.. Test Accuracy: 0.867\nEpoch: 29/30.. Training Loss: 0.346.. Test Loss: 0.402.. Test Accuracy: 0.867\nEpoch: 30/30.. Training Loss: 0.341.. Test Loss: 0.373.. Test Accuracy: 0.874\n" ], [ "%matplotlib inline\n%config InlineBackend.figure_format = 'retina'\n\nimport matplotlib.pyplot as plt", "_____no_output_____" ], [ "plt.plot(train_losses, label='Training loss')\nplt.plot(test_losses, label='Validation loss')\nplt.legend(frameon=False)", "_____no_output_____" ] ], [ [ "## Inference\n\nNow that the model is trained, we can use it for inference. We've done this before, but now we need to remember to set the model in inference mode with `model.eval()`. You'll also want to turn off autograd with the `torch.no_grad()` context.", "_____no_output_____" ] ], [ [ "# Import helper module (should be in the repo)\nimport helper\n\n# Test out your network!\n\nmodel.eval()\n\ndataiter = iter(testloader)\nimages, labels = dataiter.next()\nimg = images[0]\n# Convert 2D image to 1D vector\nimg = img.view(1, 784)\n\n# Calculate the class probabilities (softmax) for img\nwith torch.no_grad():\n output = model.forward(img)\n\nps = torch.exp(output)\n\n# Plot the image and probabilities\nhelper.view_classify(img.view(1, 28, 28), ps, version='Fashion')", "_____no_output_____" ] ], [ [ "## Next Up!\n\nIn the next part, I'll show you how to save your trained models. In general, you won't want to train a model everytime you need it. Instead, you'll train once, save it, then load the model when you want to train more or use if for inference.", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown" ]

[ [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code", "code", "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "# Requirements", "_____no_output_____" ] ], [ [ "# !pip install --upgrade transformers bertviz checklist", "_____no_output_____" ] ], [ [ "# Data loading", "_____no_output_____" ] ], [ [ "# !rm -rf ru_news_cause_v1.tsv*\n# !wget https://www.dropbox.com/s/kcxnhjzfut4guut/ru_news_cause_v1.tsv.tar.gz\n# !tar -xzvf ru_news_cause_v1.tsv.tar.gz", "_____no_output_____" ], [ "# !cat ru_news_cause_v1.tsv | wc -l\n# !head ru_news_cause_v1.tsv", "_____no_output_____" ] ], [ [ "# GPTCause", "_____no_output_____" ] ], [ [ "from transformers import GPT2LMHeadModel, GPT2TokenizerFast\ndevice = 'cuda'\nmodel_id = 'sberbank-ai/rugpt3small_based_on_gpt2' \n \nmodel = GPT2LMHeadModel.from_pretrained(model_id).to(device)\ntokenizer = GPT2TokenizerFast.from_pretrained(model_id)", "2021-08-01 15:30:34.664821: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcudart.so.11.0\n" ], [ "import torch\nmax_length = model.config.n_positions\n\ndef gpt_assess(s1, s2):\n encodings = tokenizer(f'{s1} {s2}', return_tensors='pt')\n with torch.no_grad():\n outputs = model(encodings.input_ids.to(device), labels=encodings.input_ids.to(device))\n log_likelihood = outputs[0] * encodings.input_ids.size(1)\n return log_likelihood.detach().cpu().numpy()\n\ndef gpt_assess_pair(s1,s2):\n ppl1 = gpt_assess(s1,s2)\n ppl2 = gpt_assess(s2,s1)\n if ppl1<ppl2:\n return 0, ppl1/ppl2\n else:\n return 1, ppl1/ppl2", "_____no_output_____" ], [ "print(gpt_assess('Привет!', 'Как дела?'))\nprint(gpt_assess('Как дела?', 'Привет!'))\nprint(gpt_assess_pair('Как дела?', 'Привет!'))\nprint(gpt_assess_pair('Привет!', 'Как дела?'))\n", "11.083624\n23.637806\n(1, 2.1326785)\n(0, 0.46889395)\n" ] ], [ [ "## Scoring", "_____no_output_____" ] ], [ [ "import csv\n\nlabels = []\ntexts = []\npreds = []\nconfs = []\n\nwith open(\"ru_news_cause_v1.tsv\", \"r\", encoding='utf-8') as r:\n reader = csv.reader(r, delimiter=\"\\t\")\n header = next(reader)\n for row in reader:\n r = dict(zip(header, row))\n if float(r[\"confidence\"]) < 0.69:\n continue\n result = r[\"result\"]\n mapping = {\n \"left_right_cause\": 0,\n \"left_right_cancel\": 0,\n \"right_left_cause\": 1,\n \"right_left_cancel\": 1\n }\n if result not in mapping:\n continue\n r[\"label\"] = mapping[result]\n \n labels.append(r['label'])\n texts.append( (r[\"left_title\"], r[\"right_title\"] ) )\n p, c = gpt_assess_pair( r[\"left_title\"], r[\"right_title\"] )\n preds.append( p )\n confs.append( c )\n", "_____no_output_____" ], [ "from collections import Counter\n\nprint('labels', Counter(labels))\nprint('preds', Counter(preds))", "labels Counter({1: 763, 0: 706})\npreds Counter({1: 742, 0: 727})\n" ], [ "import matplotlib.pyplot as plt\n\nplt.hist(confs)\nplt.show()", "huggingface/tokenizers: The current process just got forked, after parallelism has already been used. Disabling parallelism to avoid deadlocks...\nTo disable this warning, you can either:\n\t- Avoid using `tokenizers` before the fork if possible\n\t- Explicitly set the environment variable TOKENIZERS_PARALLELISM=(true | false)\nhuggingface/tokenizers: The current process just got forked, after parallelism has already been used. Disabling parallelism to avoid deadlocks...\nTo disable this warning, you can either:\n\t- Avoid using `tokenizers` before the fork if possible\n\t- Explicitly set the environment variable TOKENIZERS_PARALLELISM=(true | false)\n" ], [ "from sklearn.metrics import classification_report, balanced_accuracy_score, confusion_matrix\ny_true = labels\ny_pred = preds\n\nprint(classification_report(y_true, y_pred))\nprint('balanced_accuracy_score', balanced_accuracy_score(y_true, y_pred))\nprint('\\nconfusion_matrix\\n',confusion_matrix(y_true, y_pred))\n\n", " precision recall f1-score support\n\n 0 0.58 0.60 0.59 706\n 1 0.62 0.60 0.61 763\n\n accuracy 0.60 1469\n macro avg 0.60 0.60 0.60 1469\nweighted avg 0.60 0.60 0.60 1469\n\nbalanced_accuracy_score 0.5976343938308228\n\nconfusion_matrix\n [[421 285]\n [306 457]]\n" ], [ "import numpy as np\n\nconfidence_th = .1\n\nmask = np.array(list(map(lambda x:abs(x-1.), confs)))>confidence_th\n\ny_true = np.array(labels)[mask]\ny_pred = np.array(preds)[mask]\n\nprint(classification_report(y_true, y_pred))\nprint('balanced_accuracy_score', balanced_accuracy_score(y_true, y_pred))\nprint('\\nconfusion_matrix\\n',confusion_matrix(y_true, y_pred))\n\n", " precision recall f1-score support\n\n 0 0.73 0.67 0.70 139\n 1 0.72 0.78 0.75 153\n\n accuracy 0.73 292\n macro avg 0.73 0.72 0.72 292\nweighted avg 0.73 0.73 0.73 292\n\nbalanced_accuracy_score 0.7234212629896083\n\nconfusion_matrix\n [[ 93 46]\n [ 34 119]]\n" ], [ "from sklearn.metrics import f1_score\n\nxs = []\nys = []\nfor th_idx in range(250):\n th = th_idx/1000.\n mask = np.array(list(map(lambda x:abs(x-1.), confs)))>th\n\n y_true = np.array(labels)[mask]\n y_pred = np.array(preds)[mask]\n xs.append( th )\n ys.append( f1_score(y_true, y_pred) )\n", "_____no_output_____" ], [ "import matplotlib.pyplot as plt\n\nplt.plot(xs, ys)\nplt.suptitle('f1 by conf_th')\nplt.show()", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "from abc import ABC\nimport cmath\n\nclass DiscriminantStrategy(ABC):\n def calculate_discriminant(self, a, b, c):\n pass\n\n\nclass OrdinaryDiscriminantStrategy(DiscriminantStrategy):\n def calculate_discriminant(self, a, b, c):\n first_item = b*b\n second_item = 4*a*c\n return first_item - second_item\n\n\nclass RealDiscriminantStrategy(DiscriminantStrategy):\n def calculate_discriminant(self, a, b, c):\n first_item = b*b\n second_item = 4*a*c\n if second_item > first_item:\n return float('nan')\n else:\n return first_item - second_item\n\n\nclass QuadraticEquationSolver:\n def __init__(self, strategy):\n self.strategy = strategy\n\n def solve(self, a, b, c):\n \"\"\" Returns a pair of complex (!) values \"\"\"\n discriminant = complex(strategy.calculate_discriminant(a,b,c), 0)\n root_dis = cmath.sqrt(discriminant)\n return (\n (-b + root_dis) / (2*a),\n (-b - root_dis) / (2*a)\n )", "_____no_output_____" ], [ "strategy = RealDiscriminantStrategy()\nsolver = QuadraticEquationSolver(strategy)\nresults = solver.solve(1, 4, 5)", "_____no_output_____" ], [ "print(results)", "((nan+nanj), (nan+nanj))\n" ] ] ]

[ "code" ]

[ [ "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# CS224N Assignment 1: Exploring Word Vectors (25 Points)\n### <font color='blue'> Due 4:30pm, Tue Jan 14 </font>\n\nWelcome to CS224n! \n\nBefore you start, make sure you read the README.txt in the same directory as this notebook. You will find many provided codes in the notebook. We highly encourage you to read and understand the provided codes as part of the learning :-)", "_____no_output_____" ] ], [ [ "# All Import Statements Defined Here\n# Note: Do not add to this list.\n# ----------------\n\nimport sys\nassert sys.version_info[0]==3\nassert sys.version_info[1] >= 5\n\nfrom gensim.models import KeyedVectors\nfrom gensim.test.utils import datapath\nimport pprint\nimport matplotlib.pyplot as plt\nplt.rcParams['figure.figsize'] = [10, 5]\nimport nltk\nnltk.download('reuters')\nfrom nltk.corpus import reuters\nimport numpy as np\nimport random\nimport scipy as sp\nfrom sklearn.decomposition import TruncatedSVD\nfrom sklearn.decomposition import PCA\n\nSTART_TOKEN = '<START>'\nEND_TOKEN = '<END>'\n\nnp.random.seed(0)\nrandom.seed(0)\n# ----------------", "[nltk_data] Downloading package reuters to\n[nltk_data] /usr/local/share/nltk_data...\n[nltk_data] Package reuters is already up-to-date!\n" ] ], [ [ "## Word Vectors\n\nWord Vectors are often used as a fundamental component for downstream NLP tasks, e.g. question answering, text generation, translation, etc., so it is important to build some intuitions as to their strengths and weaknesses. Here, you will explore two types of word vectors: those derived from *co-occurrence matrices*, and those derived via *GloVe*. \n\n**Assignment Notes:** Please make sure to save the notebook as you go along. Submission Instructions are located at the bottom of the notebook.\n\n**Note on Terminology:** The terms \"word vectors\" and \"word embeddings\" are often used interchangeably. The term \"embedding\" refers to the fact that we are encoding aspects of a word's meaning in a lower dimensional space. As [Wikipedia](https://en.wikipedia.org/wiki/Word_embedding) states, \"*conceptually it involves a mathematical embedding from a space with one dimension per word to a continuous vector space with a much lower dimension*\".", "_____no_output_____" ], [ "## Part 1: Count-Based Word Vectors (10 points)\n\nMost word vector models start from the following idea:\n\n*You shall know a word by the company it keeps ([Firth, J. R. 1957:11](https://en.wikipedia.org/wiki/John_Rupert_Firth))*\n\nMany word vector implementations are driven by the idea that similar words, i.e., (near) synonyms, will be used in similar contexts. As a result, similar words will often be spoken or written along with a shared subset of words, i.e., contexts. By examining these contexts, we can try to develop embeddings for our words. With this intuition in mind, many \"old school\" approaches to constructing word vectors relied on word counts. Here we elaborate upon one of those strategies, *co-occurrence matrices* (for more information, see [here](http://web.stanford.edu/class/cs124/lec/vectorsemantics.video.pdf) or [here](https://medium.com/data-science-group-iitr/word-embedding-2d05d270b285)).", "_____no_output_____" ], [ "### Co-Occurrence\n\nA co-occurrence matrix counts how often things co-occur in some environment. Given some word $w_i$ occurring in the document, we consider the *context window* surrounding $w_i$. Supposing our fixed window size is $n$, then this is the $n$ preceding and $n$ subsequent words in that document, i.e. words $w_{i-n} \\dots w_{i-1}$ and $w_{i+1} \\dots w_{i+n}$. We build a *co-occurrence matrix* $M$, which is a symmetric word-by-word matrix in which $M_{ij}$ is the number of times $w_j$ appears inside $w_i$'s window among all documents.\n\n**Example: Co-Occurrence with Fixed Window of n=1**:\n\nDocument 1: \"all that glitters is not gold\"\n\nDocument 2: \"all is well that ends well\"\n\n\n| * | `<START>` | all | that | glitters | is | not | gold | well | ends | `<END>` |\n|----------|-------|-----|------|----------|------|------|-------|------|------|-----|\n| `<START>` | 0 | 2 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |\n| all | 2 | 0 | 1 | 0 | 1 | 0 | 0 | 0 | 0 | 0 |\n| that | 0 | 1 | 0 | 1 | 0 | 0 | 0 | 1 | 1 | 0 |\n| glitters | 0 | 0 | 1 | 0 | 1 | 0 | 0 | 0 | 0 | 0 |\n| is | 0 | 1 | 0 | 1 | 0 | 1 | 0 | 1 | 0 | 0 |\n| not | 0 | 0 | 0 | 0 | 1 | 0 | 1 | 0 | 0 | 0 |\n| gold | 0 | 0 | 0 | 0 | 0 | 1 | 0 | 0 | 0 | 1 |\n| well | 0 | 0 | 1 | 0 | 1 | 0 | 0 | 0 | 1 | 1 |\n| ends | 0 | 0 | 1 | 0 | 0 | 0 | 0 | 1 | 0 | 0 |\n| `<END>` | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 1 | 0 | 0 |\n\n**Note:** In NLP, we often add `<START>` and `<END>` tokens to represent the beginning and end of sentences, paragraphs or documents. In thise case we imagine `<START>` and `<END>` tokens encapsulating each document, e.g., \"`<START>` All that glitters is not gold `<END>`\", and include these tokens in our co-occurrence counts.\n\nThe rows (or columns) of this matrix provide one type of word vectors (those based on word-word co-occurrence), but the vectors will be large in general (linear in the number of distinct words in a corpus). Thus, our next step is to run *dimensionality reduction*. In particular, we will run *SVD (Singular Value Decomposition)*, which is a kind of generalized *PCA (Principal Components Analysis)* to select the top $k$ principal components. Here's a visualization of dimensionality reduction with SVD. In this picture our co-occurrence matrix is $A$ with $n$ rows corresponding to $n$ words. We obtain a full matrix decomposition, with the singular values ordered in the diagonal $S$ matrix, and our new, shorter length-$k$ word vectors in $U_k$.\n\n![Picture of an SVD](./imgs/svd.png \"SVD\")\n\nThis reduced-dimensionality co-occurrence representation preserves semantic relationships between words, e.g. *doctor* and *hospital* will be closer than *doctor* and *dog*. \n\n**Notes:** If you can barely remember what an eigenvalue is, here's [a slow, friendly introduction to SVD](https://davetang.org/file/Singular_Value_Decomposition_Tutorial.pdf). If you want to learn more thoroughly about PCA or SVD, feel free to check out lectures [7](https://web.stanford.edu/class/cs168/l/l7.pdf), [8](http://theory.stanford.edu/~tim/s15/l/l8.pdf), and [9](https://web.stanford.edu/class/cs168/l/l9.pdf) of CS168. These course notes provide a great high-level treatment of these general purpose algorithms. Though, for the purpose of this class, you only need to know how to extract the k-dimensional embeddings by utilizing pre-programmed implementations of these algorithms from the numpy, scipy, or sklearn python packages. In practice, it is challenging to apply full SVD to large corpora because of the memory needed to perform PCA or SVD. However, if you only want the top $k$ vector components for relatively small $k$ — known as [Truncated SVD](https://en.wikipedia.org/wiki/Singular_value_decomposition#Truncated_SVD) — then there are reasonably scalable techniques to compute those iteratively.", "_____no_output_____" ], [ "### Plotting Co-Occurrence Word Embeddings\n\nHere, we will be using the Reuters (business and financial news) corpus. If you haven't run the import cell at the top of this page, please run it now (click it and press SHIFT-RETURN). The corpus consists of 10,788 news documents totaling 1.3 million words. These documents span 90 categories and are split into train and test. For more details, please see https://www.nltk.org/book/ch02.html. We provide a `read_corpus` function below that pulls out only articles from the \"crude\" (i.e. news articles about oil, gas, etc.) category. The function also adds `<START>` and `<END>` tokens to each of the documents, and lowercases words. You do **not** have to perform any other kind of pre-processing.", "_____no_output_____" ] ], [ [ "def read_corpus(category=\"crude\"):\n \"\"\" Read files from the specified Reuter's category.\n Params:\n category (string): category name\n Return:\n list of lists, with words from each of the processed files\n \"\"\"\n files = reuters.fileids(category)\n return [[START_TOKEN] + [w.lower() for w in list(reuters.words(f))] + [END_TOKEN] for f in files]\n", "_____no_output_____" ] ], [ [ "Let's have a look what these documents are like….", "_____no_output_____" ] ], [ [ "reuters_corpus = read_corpus()\npprint.pprint(reuters_corpus[:3], compact=True, width=100)", "[['<START>', 'japan', 'to', 'revise', 'long', '-', 'term', 'energy', 'demand', 'downwards', 'the',\n 'ministry', 'of', 'international', 'trade', 'and', 'industry', '(', 'miti', ')', 'will', 'revise',\n 'its', 'long', '-', 'term', 'energy', 'supply', '/', 'demand', 'outlook', 'by', 'august', 'to',\n 'meet', 'a', 'forecast', 'downtrend', 'in', 'japanese', 'energy', 'demand', ',', 'ministry',\n 'officials', 'said', '.', 'miti', 'is', 'expected', 'to', 'lower', 'the', 'projection', 'for',\n 'primary', 'energy', 'supplies', 'in', 'the', 'year', '2000', 'to', '550', 'mln', 'kilolitres',\n '(', 'kl', ')', 'from', '600', 'mln', ',', 'they', 'said', '.', 'the', 'decision', 'follows',\n 'the', 'emergence', 'of', 'structural', 'changes', 'in', 'japanese', 'industry', 'following',\n 'the', 'rise', 'in', 'the', 'value', 'of', 'the', 'yen', 'and', 'a', 'decline', 'in', 'domestic',\n 'electric', 'power', 'demand', '.', 'miti', 'is', 'planning', 'to', 'work', 'out', 'a', 'revised',\n 'energy', 'supply', '/', 'demand', 'outlook', 'through', 'deliberations', 'of', 'committee',\n 'meetings', 'of', 'the', 'agency', 'of', 'natural', 'resources', 'and', 'energy', ',', 'the',\n 'officials', 'said', '.', 'they', 'said', 'miti', 'will', 'also', 'review', 'the', 'breakdown',\n 'of', 'energy', 'supply', 'sources', ',', 'including', 'oil', ',', 'nuclear', ',', 'coal', 'and',\n 'natural', 'gas', '.', 'nuclear', 'energy', 'provided', 'the', 'bulk', 'of', 'japan', \"'\", 's',\n 'electric', 'power', 'in', 'the', 'fiscal', 'year', 'ended', 'march', '31', ',', 'supplying',\n 'an', 'estimated', '27', 'pct', 'on', 'a', 'kilowatt', '/', 'hour', 'basis', ',', 'followed',\n 'by', 'oil', '(', '23', 'pct', ')', 'and', 'liquefied', 'natural', 'gas', '(', '21', 'pct', '),',\n 'they', 'noted', '.', '<END>'],\n ['<START>', 'energy', '/', 'u', '.', 's', '.', 'petrochemical', 'industry', 'cheap', 'oil',\n 'feedstocks', ',', 'the', 'weakened', 'u', '.', 's', '.', 'dollar', 'and', 'a', 'plant',\n 'utilization', 'rate', 'approaching', '90', 'pct', 'will', 'propel', 'the', 'streamlined', 'u',\n '.', 's', '.', 'petrochemical', 'industry', 'to', 'record', 'profits', 'this', 'year', ',',\n 'with', 'growth', 'expected', 'through', 'at', 'least', '1990', ',', 'major', 'company',\n 'executives', 'predicted', '.', 'this', 'bullish', 'outlook', 'for', 'chemical', 'manufacturing',\n 'and', 'an', 'industrywide', 'move', 'to', 'shed', 'unrelated', 'businesses', 'has', 'prompted',\n 'gaf', 'corp', '&', 'lt', ';', 'gaf', '>,', 'privately', '-', 'held', 'cain', 'chemical', 'inc',\n ',', 'and', 'other', 'firms', 'to', 'aggressively', 'seek', 'acquisitions', 'of', 'petrochemical',\n 'plants', '.', 'oil', 'companies', 'such', 'as', 'ashland', 'oil', 'inc', '&', 'lt', ';', 'ash',\n '>,', 'the', 'kentucky', '-', 'based', 'oil', 'refiner', 'and', 'marketer', ',', 'are', 'also',\n 'shopping', 'for', 'money', '-', 'making', 'petrochemical', 'businesses', 'to', 'buy', '.', '\"',\n 'i', 'see', 'us', 'poised', 'at', 'the', 'threshold', 'of', 'a', 'golden', 'period', ',\"', 'said',\n 'paul', 'oreffice', ',', 'chairman', 'of', 'giant', 'dow', 'chemical', 'co', '&', 'lt', ';',\n 'dow', '>,', 'adding', ',', '\"', 'there', \"'\", 's', 'no', 'major', 'plant', 'capacity', 'being',\n 'added', 'around', 'the', 'world', 'now', '.', 'the', 'whole', 'game', 'is', 'bringing', 'out',\n 'new', 'products', 'and', 'improving', 'the', 'old', 'ones', '.\"', 'analysts', 'say', 'the',\n 'chemical', 'industry', \"'\", 's', 'biggest', 'customers', ',', 'automobile', 'manufacturers',\n 'and', 'home', 'builders', 'that', 'use', 'a', 'lot', 'of', 'paints', 'and', 'plastics', ',',\n 'are', 'expected', 'to', 'buy', 'quantities', 'this', 'year', '.', 'u', '.', 's', '.',\n 'petrochemical', 'plants', 'are', 'currently', 'operating', 'at', 'about', '90', 'pct',\n 'capacity', ',', 'reflecting', 'tighter', 'supply', 'that', 'could', 'hike', 'product', 'prices',\n 'by', '30', 'to', '40', 'pct', 'this', 'year', ',', 'said', 'john', 'dosher', ',', 'managing',\n 'director', 'of', 'pace', 'consultants', 'inc', 'of', 'houston', '.', 'demand', 'for', 'some',\n 'products', 'such', 'as', 'styrene', 'could', 'push', 'profit', 'margins', 'up', 'by', 'as',\n 'much', 'as', '300', 'pct', ',', 'he', 'said', '.', 'oreffice', ',', 'speaking', 'at', 'a',\n 'meeting', 'of', 'chemical', 'engineers', 'in', 'houston', ',', 'said', 'dow', 'would', 'easily',\n 'top', 'the', '741', 'mln', 'dlrs', 'it', 'earned', 'last', 'year', 'and', 'predicted', 'it',\n 'would', 'have', 'the', 'best', 'year', 'in', 'its', 'history', '.', 'in', '1985', ',', 'when',\n 'oil', 'prices', 'were', 'still', 'above', '25', 'dlrs', 'a', 'barrel', 'and', 'chemical',\n 'exports', 'were', 'adversely', 'affected', 'by', 'the', 'strong', 'u', '.', 's', '.', 'dollar',\n ',', 'dow', 'had', 'profits', 'of', '58', 'mln', 'dlrs', '.', '\"', 'i', 'believe', 'the',\n 'entire', 'chemical', 'industry', 'is', 'headed', 'for', 'a', 'record', 'year', 'or', 'close',\n 'to', 'it', ',\"', 'oreffice', 'said', '.', 'gaf', 'chairman', 'samuel', 'heyman', 'estimated',\n 'that', 'the', 'u', '.', 's', '.', 'chemical', 'industry', 'would', 'report', 'a', '20', 'pct',\n 'gain', 'in', 'profits', 'during', '1987', '.', 'last', 'year', ',', 'the', 'domestic',\n 'industry', 'earned', 'a', 'total', 'of', '13', 'billion', 'dlrs', ',', 'a', '54', 'pct', 'leap',\n 'from', '1985', '.', 'the', 'turn', 'in', 'the', 'fortunes', 'of', 'the', 'once', '-', 'sickly',\n 'chemical', 'industry', 'has', 'been', 'brought', 'about', 'by', 'a', 'combination', 'of', 'luck',\n 'and', 'planning', ',', 'said', 'pace', \"'\", 's', 'john', 'dosher', '.', 'dosher', 'said', 'last',\n 'year', \"'\", 's', 'fall', 'in', 'oil', 'prices', 'made', 'feedstocks', 'dramatically', 'cheaper',\n 'and', 'at', 'the', 'same', 'time', 'the', 'american', 'dollar', 'was', 'weakening', 'against',\n 'foreign', 'currencies', '.', 'that', 'helped', 'boost', 'u', '.', 's', '.', 'chemical',\n 'exports', '.', 'also', 'helping', 'to', 'bring', 'supply', 'and', 'demand', 'into', 'balance',\n 'has', 'been', 'the', 'gradual', 'market', 'absorption', 'of', 'the', 'extra', 'chemical',\n 'manufacturing', 'capacity', 'created', 'by', 'middle', 'eastern', 'oil', 'producers', 'in',\n 'the', 'early', '1980s', '.', 'finally', ',', 'virtually', 'all', 'major', 'u', '.', 's', '.',\n 'chemical', 'manufacturers', 'have', 'embarked', 'on', 'an', 'extensive', 'corporate',\n 'restructuring', 'program', 'to', 'mothball', 'inefficient', 'plants', ',', 'trim', 'the',\n 'payroll', 'and', 'eliminate', 'unrelated', 'businesses', '.', 'the', 'restructuring', 'touched',\n 'off', 'a', 'flurry', 'of', 'friendly', 'and', 'hostile', 'takeover', 'attempts', '.', 'gaf', ',',\n 'which', 'made', 'an', 'unsuccessful', 'attempt', 'in', '1985', 'to', 'acquire', 'union',\n 'carbide', 'corp', '&', 'lt', ';', 'uk', '>,', 'recently', 'offered', 'three', 'billion', 'dlrs',\n 'for', 'borg', 'warner', 'corp', '&', 'lt', ';', 'bor', '>,', 'a', 'chicago', 'manufacturer',\n 'of', 'plastics', 'and', 'chemicals', '.', 'another', 'industry', 'powerhouse', ',', 'w', '.',\n 'r', '.', 'grace', '&', 'lt', ';', 'gra', '>', 'has', 'divested', 'its', 'retailing', ',',\n 'restaurant', 'and', 'fertilizer', 'businesses', 'to', 'raise', 'cash', 'for', 'chemical',\n 'acquisitions', '.', 'but', 'some', 'experts', 'worry', 'that', 'the', 'chemical', 'industry',\n 'may', 'be', 'headed', 'for', 'trouble', 'if', 'companies', 'continue', 'turning', 'their',\n 'back', 'on', 'the', 'manufacturing', 'of', 'staple', 'petrochemical', 'commodities', ',', 'such',\n 'as', 'ethylene', ',', 'in', 'favor', 'of', 'more', 'profitable', 'specialty', 'chemicals',\n 'that', 'are', 'custom', '-', 'designed', 'for', 'a', 'small', 'group', 'of', 'buyers', '.', '\"',\n 'companies', 'like', 'dupont', '&', 'lt', ';', 'dd', '>', 'and', 'monsanto', 'co', '&', 'lt', ';',\n 'mtc', '>', 'spent', 'the', 'past', 'two', 'or', 'three', 'years', 'trying', 'to', 'get', 'out',\n 'of', 'the', 'commodity', 'chemical', 'business', 'in', 'reaction', 'to', 'how', 'badly', 'the',\n 'market', 'had', 'deteriorated', ',\"', 'dosher', 'said', '.', '\"', 'but', 'i', 'think', 'they',\n 'will', 'eventually', 'kill', 'the', 'margins', 'on', 'the', 'profitable', 'chemicals', 'in',\n 'the', 'niche', 'market', '.\"', 'some', 'top', 'chemical', 'executives', 'share', 'the',\n 'concern', '.', '\"', 'the', 'challenge', 'for', 'our', 'industry', 'is', 'to', 'keep', 'from',\n 'getting', 'carried', 'away', 'and', 'repeating', 'past', 'mistakes', ',\"', 'gaf', \"'\", 's',\n 'heyman', 'cautioned', '.', '\"', 'the', 'shift', 'from', 'commodity', 'chemicals', 'may', 'be',\n 'ill', '-', 'advised', '.', 'specialty', 'businesses', 'do', 'not', 'stay', 'special', 'long',\n '.\"', 'houston', '-', 'based', 'cain', 'chemical', ',', 'created', 'this', 'month', 'by', 'the',\n 'sterling', 'investment', 'banking', 'group', ',', 'believes', 'it', 'can', 'generate', '700',\n 'mln', 'dlrs', 'in', 'annual', 'sales', 'by', 'bucking', 'the', 'industry', 'trend', '.',\n 'chairman', 'gordon', 'cain', ',', 'who', 'previously', 'led', 'a', 'leveraged', 'buyout', 'of',\n 'dupont', \"'\", 's', 'conoco', 'inc', \"'\", 's', 'chemical', 'business', ',', 'has', 'spent', '1',\n '.', '1', 'billion', 'dlrs', 'since', 'january', 'to', 'buy', 'seven', 'petrochemical', 'plants',\n 'along', 'the', 'texas', 'gulf', 'coast', '.', 'the', 'plants', 'produce', 'only', 'basic',\n 'commodity', 'petrochemicals', 'that', 'are', 'the', 'building', 'blocks', 'of', 'specialty',\n 'products', '.', '\"', 'this', 'kind', 'of', 'commodity', 'chemical', 'business', 'will', 'never',\n 'be', 'a', 'glamorous', ',', 'high', '-', 'margin', 'business', ',\"', 'cain', 'said', ',',\n 'adding', 'that', 'demand', 'is', 'expected', 'to', 'grow', 'by', 'about', 'three', 'pct',\n 'annually', '.', 'garo', 'armen', ',', 'an', 'analyst', 'with', 'dean', 'witter', 'reynolds', ',',\n 'said', 'chemical', 'makers', 'have', 'also', 'benefitted', 'by', 'increasing', 'demand', 'for',\n 'plastics', 'as', 'prices', 'become', 'more', 'competitive', 'with', 'aluminum', ',', 'wood',\n 'and', 'steel', 'products', '.', 'armen', 'estimated', 'the', 'upturn', 'in', 'the', 'chemical',\n 'business', 'could', 'last', 'as', 'long', 'as', 'four', 'or', 'five', 'years', ',', 'provided',\n 'the', 'u', '.', 's', '.', 'economy', 'continues', 'its', 'modest', 'rate', 'of', 'growth', '.',\n '<END>'],\n ['<START>', 'turkey', 'calls', 'for', 'dialogue', 'to', 'solve', 'dispute', 'turkey', 'said',\n 'today', 'its', 'disputes', 'with', 'greece', ',', 'including', 'rights', 'on', 'the',\n 'continental', 'shelf', 'in', 'the', 'aegean', 'sea', ',', 'should', 'be', 'solved', 'through',\n 'negotiations', '.', 'a', 'foreign', 'ministry', 'statement', 'said', 'the', 'latest', 'crisis',\n 'between', 'the', 'two', 'nato', 'members', 'stemmed', 'from', 'the', 'continental', 'shelf',\n 'dispute', 'and', 'an', 'agreement', 'on', 'this', 'issue', 'would', 'effect', 'the', 'security',\n ',', 'economy', 'and', 'other', 'rights', 'of', 'both', 'countries', '.', '\"', 'as', 'the',\n 'issue', 'is', 'basicly', 'political', ',', 'a', 'solution', 'can', 'only', 'be', 'found', 'by',\n 'bilateral', 'negotiations', ',\"', 'the', 'statement', 'said', '.', 'greece', 'has', 'repeatedly',\n 'said', 'the', 'issue', 'was', 'legal', 'and', 'could', 'be', 'solved', 'at', 'the',\n 'international', 'court', 'of', 'justice', '.', 'the', 'two', 'countries', 'approached', 'armed',\n 'confrontation', 'last', 'month', 'after', 'greece', 'announced', 'it', 'planned', 'oil',\n 'exploration', 'work', 'in', 'the', 'aegean', 'and', 'turkey', 'said', 'it', 'would', 'also',\n 'search', 'for', 'oil', '.', 'a', 'face', '-', 'off', 'was', 'averted', 'when', 'turkey',\n 'confined', 'its', 'research', 'to', 'territorrial', 'waters', '.', '\"', 'the', 'latest',\n 'crises', 'created', 'an', 'historic', 'opportunity', 'to', 'solve', 'the', 'disputes', 'between',\n 'the', 'two', 'countries', ',\"', 'the', 'foreign', 'ministry', 'statement', 'said', '.', 'turkey',\n \"'\", 's', 'ambassador', 'in', 'athens', ',', 'nazmi', 'akiman', ',', 'was', 'due', 'to', 'meet',\n 'prime', 'minister', 'andreas', 'papandreou', 'today', 'for', 'the', 'greek', 'reply', 'to', 'a',\n 'message', 'sent', 'last', 'week', 'by', 'turkish', 'prime', 'minister', 'turgut', 'ozal', '.',\n 'the', 'contents', 'of', 'the', 'message', 'were', 'not', 'disclosed', '.', '<END>']]\n" ] ], [ [ "### Question 1.1: Implement `distinct_words` [code] (2 points)\n\nWrite a method to work out the distinct words (word types) that occur in the corpus. You can do this with `for` loops, but it's more efficient to do it with Python list comprehensions. In particular, [this](https://coderwall.com/p/rcmaea/flatten-a-list-of-lists-in-one-line-in-python) may be useful to flatten a list of lists. If you're not familiar with Python list comprehensions in general, here's [more information](https://python-3-patterns-idioms-test.readthedocs.io/en/latest/Comprehensions.html).\n\nYou may find it useful to use [Python sets](https://www.w3schools.com/python/python_sets.asp) to remove duplicate words.", "_____no_output_____" ] ], [ [ "def distinct_words(corpus):\n \"\"\" Determine a list of distinct words for the corpus.\n Params:\n corpus (list of list of strings): corpus of documents\n Return:\n corpus_words (list of strings): list of distinct words across the corpus, sorted (using python 'sorted' function)\n num_corpus_words (integer): number of distinct words across the corpus\n \"\"\"\n \n # ------------------\n # Write your implementation here.\n corpus_words = list(sorted(set([token for sentences in corpus for token in sentences])))\n num_corpus_words = len(corpus_words)\n # ------------------\n\n return corpus_words, num_corpus_words", "_____no_output_____" ], [ "# ---------------------\n# Run this sanity check\n# Note that this not an exhaustive check for correctness.\n# ---------------------\n\n# Define toy corpus\ntest_corpus = [\"{} All that glitters isn't gold {}\".format(START_TOKEN, END_TOKEN).split(\" \"), \"{} All's well that ends well {}\".format(START_TOKEN, END_TOKEN).split(\" \")]\ntest_corpus_words, num_corpus_words = distinct_words(test_corpus)\n\n# Correct answers\nans_test_corpus_words = sorted([START_TOKEN, \"All\", \"ends\", \"that\", \"gold\", \"All's\", \"glitters\", \"isn't\", \"well\", END_TOKEN])\nans_num_corpus_words = len(ans_test_corpus_words)\n\n# Test correct number of words\nassert(num_corpus_words == ans_num_corpus_words), \"Incorrect number of distinct words. Correct: {}. Yours: {}\".format(ans_num_corpus_words, num_corpus_words)\n\n# Test correct words\nassert (test_corpus_words == ans_test_corpus_words), \"Incorrect corpus_words.\\nCorrect: {}\\nYours: {}\".format(str(ans_test_corpus_words), str(test_corpus_words))\n\n# Print Success\nprint (\"-\" * 80)\nprint(\"Passed All Tests!\")\nprint (\"-\" * 80)", "--------------------------------------------------------------------------------\nPassed All Tests!\n--------------------------------------------------------------------------------\n" ] ], [ [ "### Question 1.2: Implement `compute_co_occurrence_matrix` [code] (3 points)\n\nWrite a method that constructs a co-occurrence matrix for a certain window-size $n$ (with a default of 4), considering words $n$ before and $n$ after the word in the center of the window. Here, we start to use `numpy (np)` to represent vectors, matrices, and tensors. If you're not familiar with NumPy, there's a NumPy tutorial in the second half of this cs231n [Python NumPy tutorial](http://cs231n.github.io/python-numpy-tutorial/).\n", "_____no_output_____" ] ], [ [ "def compute_co_occurrence_matrix(corpus, window_size=4):\n \"\"\" Compute co-occurrence matrix for the given corpus and window_size (default of 4).\n \n Note: Each word in a document should be at the center of a window. Words near edges will have a smaller\n number of co-occurring words.\n \n For example, if we take the document \"<START> All that glitters is not gold <END>\" with window size of 4,\n \"All\" will co-occur with \"<START>\", \"that\", \"glitters\", \"is\", and \"not\".\n \n Params:\n corpus (list of list of strings): corpus of documents\n window_size (int): size of context window\n Return:\n M (a symmetric numpy matrix of shape (number of unique words in the corpus , number of unique words in the corpus)): \n Co-occurence matrix of word counts. \n The ordering of the words in the rows/columns should be the same as the ordering of the words given by the distinct_words function.\n word2Ind (dict): dictionary that maps word to index (i.e. row/column number) for matrix M.\n \"\"\"\n words, num_words = distinct_words(corpus)\n\n # ------------------\n # Write your implementation here.\n M = np.zeros((num_words, num_words), dtype=np.int)\n co_time = {ii: [] for ii in range(num_words)}\n word2Ind = dict(zip(words, range(num_words)))\n for sent in corpus:\n for idx, center in enumerate(sent):\n center_id = word2Ind[center]\n context = sent[max(0, idx - window_size):idx + window_size + 1]\n context_id = [word2Ind[ii] for ii in context if ii != center]\n co_time[center_id].extend(context_id)\n for center, co_list in co_time.items():\n unique, counts = np.unique(co_list, return_counts=True)\n co_map = dict(zip(unique, counts))\n for context, time in co_map.items():\n M[center][context] = time\n\n # ------------------\n\n return M, word2Ind", "_____no_output_____" ], [ "# ---------------------\n# Run this sanity check\n# Note that this is not an exhaustive check for correctness.\n# ---------------------\n\n# Define toy corpus and get student's co-occurrence matrix\ntest_corpus = [\"{} All that glitters isn't gold {}\".format(START_TOKEN, END_TOKEN).split(\" \"), \"{} All's well that ends well {}\".format(START_TOKEN, END_TOKEN).split(\" \")]\nM_test, word2Ind_test = compute_co_occurrence_matrix(test_corpus, window_size=1)\n\n# Correct M and word2Ind\nM_test_ans = np.array( \n [[0., 0., 0., 0., 0., 0., 1., 0., 0., 1.,],\n [0., 0., 1., 1., 0., 0., 0., 0., 0., 0.,],\n [0., 1., 0., 0., 0., 0., 0., 0., 1., 0.,],\n [0., 1., 0., 0., 0., 0., 0., 0., 0., 1.,],\n [0., 0., 0., 0., 0., 0., 0., 0., 1., 1.,],\n [0., 0., 0., 0., 0., 0., 0., 1., 1., 0.,],\n [1., 0., 0., 0., 0., 0., 0., 1., 0., 0.,],\n [0., 0., 0., 0., 0., 1., 1., 0., 0., 0.,],\n [0., 0., 1., 0., 1., 1., 0., 0., 0., 1.,],\n [1., 0., 0., 1., 1., 0., 0., 0., 1., 0.,]]\n)\nans_test_corpus_words = sorted([START_TOKEN, \"All\", \"ends\", \"that\", \"gold\", \"All's\", \"glitters\", \"isn't\", \"well\", END_TOKEN])\nword2Ind_ans = dict(zip(ans_test_corpus_words, range(len(ans_test_corpus_words))))\n\n# Test correct word2Ind\nassert (word2Ind_ans == word2Ind_test), \"Your word2Ind is incorrect:\\nCorrect: {}\\nYours: {}\".format(word2Ind_ans, word2Ind_test)\n\n# Test correct M shape\nassert (M_test.shape == M_test_ans.shape), \"M matrix has incorrect shape.\\nCorrect: {}\\nYours: {}\".format(M_test.shape, M_test_ans.shape)\n\n# Test correct M values\nfor w1 in word2Ind_ans.keys():\n idx1 = word2Ind_ans[w1]\n for w2 in word2Ind_ans.keys():\n idx2 = word2Ind_ans[w2]\n student = M_test[idx1, idx2]\n correct = M_test_ans[idx1, idx2]\n if student != correct:\n print(\"Correct M:\")\n print(M_test_ans)\n print(\"Your M: \")\n print(M_test)\n raise AssertionError(\"Incorrect count at index ({}, {})=({}, {}) in matrix M. Yours has {} but should have {}.\".format(idx1, idx2, w1, w2, student, correct))\n\n# Print Success\nprint (\"-\" * 80)\nprint(\"Passed All Tests!\")\nprint (\"-\" * 80)", "--------------------------------------------------------------------------------\nPassed All Tests!\n--------------------------------------------------------------------------------\n" ] ], [ [ "### Question 1.3: Implement `reduce_to_k_dim` [code] (1 point)\n\nConstruct a method that performs dimensionality reduction on the matrix to produce k-dimensional embeddings. Use SVD to take the top k components and produce a new matrix of k-dimensional embeddings. \n\n**Note:** All of numpy, scipy, and scikit-learn (`sklearn`) provide *some* implementation of SVD, but only scipy and sklearn provide an implementation of Truncated SVD, and only sklearn provides an efficient randomized algorithm for calculating large-scale Truncated SVD. So please use [sklearn.decomposition.TruncatedSVD](https://scikit-learn.org/stable/modules/generated/sklearn.decomposition.TruncatedSVD.html).", "_____no_output_____" ] ], [ [ "def reduce_to_k_dim(M, k=2):\n \"\"\" Reduce a co-occurence count matrix of dimensionality (num_corpus_words, num_corpus_words)\n to a matrix of dimensionality (num_corpus_words, k) using the following SVD function from Scikit-Learn:\n - http://scikit-learn.org/stable/modules/generated/sklearn.decomposition.TruncatedSVD.html\n \n Params:\n M (numpy matrix of shape (number of unique words in the corpus , number of unique words in the corpus)): co-occurence matrix of word counts\n k (int): embedding size of each word after dimension reduction\n Return:\n M_reduced (numpy matrix of shape (number of corpus words, k)): matrix of k-dimensioal word embeddings.\n In terms of the SVD from math class, this actually returns U * S\n \"\"\" \n n_iters = 10 # Use this parameter in your call to `TruncatedSVD`\n print(\"Running Truncated SVD over %i words...\" % (M.shape[0]))\n \n # ------------------\n # Write your implementation here.\n svd = TruncatedSVD(n_components=k, n_iter=n_iters)\n svd.fit(M.T)\n M_reduced = svd.components_.T\n # ------------------\n\n print(\"Done.\")\n return M_reduced", "_____no_output_____" ], [ "# ---------------------\n# Run this sanity check\n# Note that this is not an exhaustive check for correctness \n# In fact we only check that your M_reduced has the right dimensions.\n# ---------------------\n\n# Define toy corpus and run student code\ntest_corpus = [\"{} All that glitters isn't gold {}\".format(START_TOKEN, END_TOKEN).split(\" \"), \"{} All's well that ends well {}\".format(START_TOKEN, END_TOKEN).split(\" \")]\nM_test, word2Ind_test = compute_co_occurrence_matrix(test_corpus, window_size=1)\nM_test_reduced = reduce_to_k_dim(M_test, k=2)\n\n# Test proper dimensions\nassert (M_test_reduced.shape[0] == 10), \"M_reduced has {} rows; should have {}\".format(M_test_reduced.shape[0], 10)\nassert (M_test_reduced.shape[1] == 2), \"M_reduced has {} columns; should have {}\".format(M_test_reduced.shape[1], 2)\n\n# Print Success\nprint (\"-\" * 80)\nprint(\"Passed All Tests!\")\nprint (\"-\" * 80)", "Running Truncated SVD over 10 words...\nDone.\n--------------------------------------------------------------------------------\nPassed All Tests!\n--------------------------------------------------------------------------------\n" ] ], [ [ "### Question 1.4: Implement `plot_embeddings` [code] (1 point)\n\nHere you will write a function to plot a set of 2D vectors in 2D space. For graphs, we will use Matplotlib (`plt`).\n\nFor this example, you may find it useful to adapt [this code](https://www.pythonmembers.club/2018/05/08/matplotlib-scatter-plot-annotate-set-text-at-label-each-point/). In the future, a good way to make a plot is to look at [the Matplotlib gallery](https://matplotlib.org/gallery/index.html), find a plot that looks somewhat like what you want, and adapt the code they give.", "_____no_output_____" ] ], [ [ "def plot_embeddings(M_reduced, word2Ind, words):\n \"\"\" Plot in a scatterplot the embeddings of the words specified in the list \"words\".\n NOTE: do not plot all the words listed in M_reduced / word2Ind.\n Include a label next to each point.\n \n Params:\n M_reduced (numpy matrix of shape (number of unique words in the corpus , 2)): matrix of 2-dimensioal word embeddings\n word2Ind (dict): dictionary that maps word to indices for matrix M\n words (list of strings): words whose embeddings we want to visualize\n \"\"\"\n\n # ------------------\n # Write your implementation here.\n for w in words:\n x = M_reduced[word2Ind[w]][0]\n y = M_reduced[word2Ind[w]][1]\n plt.scatter(x, y, marker='x', color='red')\n plt.text(x, y, w)\n plt.show()\n # ------------------", "_____no_output_____" ], [ "# ---------------------\n# Run this sanity check\n# Note that this is not an exhaustive check for correctness.\n# The plot produced should look like the \"test solution plot\" depicted below. \n# ---------------------\n\nprint (\"-\" * 80)\nprint (\"Outputted Plot:\")\n\nM_reduced_plot_test = np.array([[1, 1], [-1, -1], [1, -1], [-1, 1], [0, 0]])\nword2Ind_plot_test = {'test1': 0, 'test2': 1, 'test3': 2, 'test4': 3, 'test5': 4}\nwords = ['test1', 'test2', 'test3', 'test4', 'test5']\nplot_embeddings(M_reduced_plot_test, word2Ind_plot_test, words)\n\nprint (\"-\" * 80)", "--------------------------------------------------------------------------------\nOutputted Plot:\n" ] ], [ [ "<font color=red>**Test Plot Solution**</font>\n<br>\n<img src=\"./imgs/test_plot.png\" width=40% style=\"float: left;\"> </img>\n", "_____no_output_____" ], [ "### Question 1.5: Co-Occurrence Plot Analysis [written] (3 points)\n\nNow we will put together all the parts you have written! We will compute the co-occurrence matrix with fixed window of 4 (the default window size), over the Reuters \"crude\" (oil) corpus. Then we will use TruncatedSVD to compute 2-dimensional embeddings of each word. TruncatedSVD returns U\\*S, so we need to normalize the returned vectors, so that all the vectors will appear around the unit circle (therefore closeness is directional closeness). **Note**: The line of code below that does the normalizing uses the NumPy concept of *broadcasting*. If you don't know about broadcasting, check out\n[Computation on Arrays: Broadcasting by Jake VanderPlas](https://jakevdp.github.io/PythonDataScienceHandbook/02.05-computation-on-arrays-broadcasting.html).\n\nRun the below cell to produce the plot. It'll probably take a few seconds to run. What clusters together in 2-dimensional embedding space? What doesn't cluster together that you might think should have? **Note:** \"bpd\" stands for \"barrels per day\" and is a commonly used abbreviation in crude oil topic articles.", "_____no_output_____" ] ], [ [ "# -----------------------------\n# Run This Cell to Produce Your Plot\n# ------------------------------\nreuters_corpus = read_corpus()\nM_co_occurrence, word2Ind_co_occurrence = compute_co_occurrence_matrix(reuters_corpus)\nM_reduced_co_occurrence = reduce_to_k_dim(M_co_occurrence, k=2)\n\n# Rescale (normalize) the rows to make them each of unit-length\nM_lengths = np.linalg.norm(M_reduced_co_occurrence, axis=1)\nM_normalized = M_reduced_co_occurrence / M_lengths[:, np.newaxis] # broadcasting\n\nwords = ['barrels', 'bpd', 'ecuador', 'energy', 'industry', 'kuwait', 'oil', 'output', 'petroleum', 'venezuela']\n\nplot_embeddings(M_normalized, word2Ind_co_occurrence, words)", "Running Truncated SVD over 8185 words...\nDone.\n" ] ], [ [ "The 'ecuador', 'energy', 'kuwait', 'oil', 'output', 'venezuela' cluster together which is intuitive.\nAnd the 'bpd', 'barrels' doesn't cluster together which should have.\n", "_____no_output_____" ], [ "## Part 2: Prediction-Based Word Vectors (15 points)\n\nAs discussed in class, more recently prediction-based word vectors have demonstrated better performance, such as word2vec and GloVe (which also utilizes the benefit of counts). Here, we shall explore the embeddings produced by GloVe. Please revisit the class notes and lecture slides for more details on the word2vec and GloVe algorithms. If you're feeling adventurous, challenge yourself and try reading [GloVe's original paper](https://nlp.stanford.edu/pubs/glove.pdf).\n\nThen run the following cells to load the GloVe vectors into memory. **Note**: If this is your first time to run these cells, i.e. download the embedding model, it will take about 15 minutes to run. If you've run these cells before, rerunning them will load the model without redownloading it, which will take about 1 to 2 minutes.", "_____no_output_____" ] ], [ [ "def load_embedding_model():\n \"\"\" Load GloVe Vectors\n Return:\n wv_from_bin: All 400000 embeddings, each lengh 200\n \"\"\"\n import gensim.downloader as api\n wv_from_bin = api.load(\"glove-wiki-gigaword-200\")\n print(\"Loaded vocab size %i\" % len(wv_from_bin.vocab.keys()))\n return wv_from_bin", "_____no_output_____" ], [ "# -----------------------------------\n# Run Cell to Load Word Vectors\n# Note: This will take several minutes\n# -----------------------------------\nwv_from_bin = load_embedding_model()", "Loaded vocab size 400000\n" ] ], [ [ "#### Note: If you are receiving reset by peer error, rerun the cell to restart the download. ", "_____no_output_____" ], [ "### Reducing dimensionality of Word Embeddings\nLet's directly compare the GloVe embeddings to those of the co-occurrence matrix. In order to avoid running out of memory, we will work with a sample of 10000 GloVe vectors instead.\nRun the following cells to:\n\n1. Put 10000 Glove vectors into a matrix M\n2. Run reduce_to_k_dim (your Truncated SVD function) to reduce the vectors from 200-dimensional to 2-dimensional.", "_____no_output_____" ] ], [ [ "def get_matrix_of_vectors(wv_from_bin, required_words=['barrels', 'bpd', 'ecuador', 'energy', 'industry', 'kuwait', 'oil', 'output', 'petroleum', 'venezuela']):\n \"\"\" Put the GloVe vectors into a matrix M.\n Param:\n wv_from_bin: KeyedVectors object; the 400000 GloVe vectors loaded from file\n Return:\n M: numpy matrix shape (num words, 200) containing the vectors\n word2Ind: dictionary mapping each word to its row number in M\n \"\"\"\n import random\n words = list(wv_from_bin.vocab.keys())\n print(\"Shuffling words ...\")\n random.seed(224)\n random.shuffle(words)\n words = words[:10000]\n print(\"Putting %i words into word2Ind and matrix M...\" % len(words))\n word2Ind = {}\n M = []\n curInd = 0\n for w in words:\n try:\n M.append(wv_from_bin.word_vec(w))\n word2Ind[w] = curInd\n curInd += 1\n except KeyError:\n continue\n for w in required_words:\n if w in words:\n continue\n try:\n M.append(wv_from_bin.word_vec(w))\n word2Ind[w] = curInd\n curInd += 1\n except KeyError:\n continue\n M = np.stack(M)\n print(\"Done.\")\n return M, word2Ind", "_____no_output_____" ], [ "# -----------------------------------------------------------------\n# Run Cell to Reduce 200-Dimensional Word Embeddings to k Dimensions\n# Note: This should be quick to run\n# -----------------------------------------------------------------\nM, word2Ind = get_matrix_of_vectors(wv_from_bin)\nM_reduced = reduce_to_k_dim(M, k=2)\n\n# Rescale (normalize) the rows to make them each of unit-length\nM_lengths = np.linalg.norm(M_reduced, axis=1)\nM_reduced_normalized = M_reduced / M_lengths[:, np.newaxis] # broadcasting", "Shuffling words ...\nPutting 10000 words into word2Ind and matrix M...\nDone.\nRunning Truncated SVD over 10010 words...\nDone.\n" ] ], [ [ "**Note: If you are receiving out of memory issues on your local machine, try closing other applications to free more memory on your device. You may want to try restarting your machine so that you can free up extra memory. Then immediately run the jupyter notebook and see if you can load the word vectors properly. If you still have problems with loading the embeddings onto your local machine after this, please follow the Piazza instructions, as how to run remotely on Stanford Farmshare machines.**", "_____no_output_____" ], [ "### Question 2.1: GloVe Plot Analysis [written] (4 points)\n\nRun the cell below to plot the 2D GloVe embeddings for `['barrels', 'bpd', 'ecuador', 'energy', 'industry', 'kuwait', 'oil', 'output', 'petroleum', 'venezuela']`.\n\nWhat clusters together in 2-dimensional embedding space? What doesn't cluster together that you might think should have? How is the plot different from the one generated earlier from the co-occurrence matrix? What is a possible reason for causing the difference?", "_____no_output_____" ] ], [ [ "words = ['barrels', 'bpd', 'ecuador', 'energy', 'industry', 'kuwait', 'oil', 'output', 'petroleum', 'venezuela']\nplot_embeddings(M_reduced_normalized, word2Ind, words)", "_____no_output_____" ] ], [ [ "venezuela with ecuador and industry with energy cluster together. petroleum and oil shoulf cluster together.\nThe plot contine more semantic information than the co-occurrence matrix method.\nMaybe the counter of word is small in the data set.", "_____no_output_____" ], [ "### Cosine Similarity\nNow that we have word vectors, we need a way to quantify the similarity between individual words, according to these vectors. One such metric is cosine-similarity. We will be using this to find words that are \"close\" and \"far\" from one another.\n\nWe can think of n-dimensional vectors as points in n-dimensional space. If we take this perspective [L1](http://mathworld.wolfram.com/L1-Norm.html) and [L2](http://mathworld.wolfram.com/L2-Norm.html) Distances help quantify the amount of space \"we must travel\" to get between these two points. Another approach is to examine the angle between two vectors. From trigonometry we know that:\n\n<img src=\"./imgs/inner_product.png\" width=20% style=\"float: center;\"></img>\n\nInstead of computing the actual angle, we can leave the similarity in terms of $similarity = cos(\\Theta)$. Formally the [Cosine Similarity](https://en.wikipedia.org/wiki/Cosine_similarity) $s$ between two vectors $p$ and $q$ is defined as:\n\n$$s = \\frac{p \\cdot q}{||p|| ||q||}, \\textrm{ where } s \\in [-1, 1] $$ ", "_____no_output_____" ], [ "### Question 2.2: Words with Multiple Meanings (2 points) [code + written] \nPolysemes and homonyms are words that have more than one meaning (see this [wiki page](https://en.wikipedia.org/wiki/Polysemy) to learn more about the difference between polysemes and homonyms ). Find a word with at least 2 different meanings such that the top-10 most similar words (according to cosine similarity) contain related words from *both* meanings. For example, \"leaves\" has both \"vanishes\" and \"stalks\" in the top 10, and \"scoop\" has both \"handed_waffle_cone\" and \"lowdown\". You will probably need to try several polysemous or homonymic words before you find one. Please state the word you discover and the multiple meanings that occur in the top 10. Why do you think many of the polysemous or homonymic words you tried didn't work (i.e. the top-10 most similar words only contain **one** of the meanings of the words)?\n\n**Note**: You should use the `wv_from_bin.most_similar(word)` function to get the top 10 similar words. This function ranks all other words in the vocabulary with respect to their cosine similarity to the given word. For further assistance please check the __[GenSim documentation](https://radimrehurek.com/gensim/models/keyedvectors.html#gensim.models.keyedvectors.FastTextKeyedVectors.most_similar)__.", "_____no_output_____" ] ], [ [ "# ------------------\n# Write your implementation here.\nwv_from_bin.most_similar(\"run\")\n# ------------------", "_____no_output_____" ] ], [ [ "The run have running and start two meaning.", "_____no_output_____" ], [ "### Question 2.3: Synonyms & Antonyms (2 points) [code + written] \n\nWhen considering Cosine Similarity, it's often more convenient to think of Cosine Distance, which is simply 1 - Cosine Similarity.\n\nFind three words (w1,w2,w3) where w1 and w2 are synonyms and w1 and w3 are antonyms, but Cosine Distance(w1,w3) < Cosine Distance(w1,w2). For example, w1=\"happy\" is closer to w3=\"sad\" than to w2=\"cheerful\". \n\nOnce you have found your example, please give a possible explanation for why this counter-intuitive result may have happened.\n\nYou should use the the `wv_from_bin.distance(w1, w2)` function here in order to compute the cosine distance between two words. Please see the __[GenSim documentation](https://radimrehurek.com/gensim/models/keyedvectors.html#gensim.models.keyedvectors.FastTextKeyedVectors.distance)__ for further assistance.", "_____no_output_____" ] ], [ [ "# ------------------\n# Write your implementation here.\nw1 = \"design\"\nw2 = \"proposal\"\nw3 = \"borrow\"\nw12_dist = wv_from_bin.distance(w1, w2)\nw13_dist = wv_from_bin.distance(w1, w3)\nprint(\"Synonyms {}, {} have cosine distance: {:.2f}\".format(w1, w2, w12_dist))\nprint(\"Antonyms {}, {} have cosine distance: {:.2f}\".format(w1, w3, w13_dist))\n\n# ------------------", "Synonyms design, proposal have cosine distance: 0.69\nAntonyms design, borrow have cosine distance: 0.90\n" ] ], [ [ "Comapring proposal and design, the cosine distrance is 0.69. And the cosine distance between design and borrow is 0.9.", "_____no_output_____" ], [ "### Solving Analogies with Word Vectors\nWord vectors have been shown to *sometimes* exhibit the ability to solve analogies. \n\nAs an example, for the analogy \"man : king :: woman : x\" (read: man is to king as woman is to x), what is x?\n\nIn the cell below, we show you how to use word vectors to find x. The `most_similar` function finds words that are most similar to the words in the `positive` list and most dissimilar from the words in the `negative` list. The answer to the analogy will be the word ranked most similar (largest numerical value).\n\n**Note:** Further Documentation on the `most_similar` function can be found within the __[GenSim documentation](https://radimrehurek.com/gensim/models/keyedvectors.html#gensim.models.keyedvectors.FastTextKeyedVectors.most_similar)__.", "_____no_output_____" ] ], [ [ "# Run this cell to answer the analogy -- man : king :: woman : x\npprint.pprint(wv_from_bin.most_similar(positive=['woman', 'king'], negative=['man']))", "[('queen', 0.6978678703308105),\n ('princess', 0.6081745028495789),\n ('monarch', 0.5889754891395569),\n ('throne', 0.5775108933448792),\n ('prince', 0.5750998854637146),\n ('elizabeth', 0.546359658241272),\n ('daughter', 0.5399125814437866),\n ('kingdom', 0.5318052768707275),\n ('mother', 0.5168544054031372),\n ('crown', 0.5164472460746765)]\n" ] ], [ [ "### Question 2.4: Finding Analogies [code + written] (2 Points)\nFind an example of analogy that holds according to these vectors (i.e. the intended word is ranked top). In your solution please state the full analogy in the form x:y :: a:b. If you believe the analogy is complicated, explain why the analogy holds in one or two sentences.\n\n**Note**: You may have to try many analogies to find one that works!", "_____no_output_____" ] ], [ [ "# ------------------\n# Write your implementation here.\npprint.pprint(wv_from_bin.most_similar(positive=['woman', 'waitress'], negative=['man']))\n# ------------------", "[('barmaid', 0.6116799116134644),\n ('bartender', 0.5877381563186646),\n ('receptionist', 0.5782569646835327),\n ('waiter', 0.5508327484130859),\n ('waitresses', 0.5503603219985962),\n ('hostess', 0.5346562266349792),\n ('housekeeper', 0.5310243368148804),\n ('homemaker', 0.5298492908477783),\n ('prostitute', 0.5254124402999878),\n ('housewife', 0.5207685232162476)]\n" ] ], [ [ "woman:man::waitress:waiter, through the probability of waiter isn't the maximum.", "_____no_output_____" ], [ "### Question 2.5: Incorrect Analogy [code + written] (1 point)\nFind an example of analogy that does *not* hold according to these vectors. In your solution, state the intended analogy in the form x:y :: a:b, and state the (incorrect) value of b according to the word vectors.", "_____no_output_____" ] ], [ [ "# ------------------\n# Write your implementation here.\npprint.pprint(wv_from_bin.most_similar(positive=['high', 'jump'], negative=['low']))\n# ------------------", "[('jumping', 0.6205310225486755),\n ('jumps', 0.5840020775794983),\n ('leap', 0.5402169823646545),\n ('jumper', 0.4817255735397339),\n ('climb', 0.4797284007072449),\n ('bungee', 0.464731365442276),\n ('championships', 0.4643418788909912),\n ('jumped', 0.46396756172180176),\n ('triple', 0.4550389349460602),\n ('throw', 0.4516879916191101)]\n" ] ], [ [ "the high:low shoudle == jump:fall, but the fall not in the top 10 probability list.", "_____no_output_____" ], [ "### Question 2.6: Guided Analysis of Bias in Word Vectors [written] (1 point)\n\nIt's important to be cognizant of the biases (gender, race, sexual orientation etc.) implicit in our word embeddings. Bias can be dangerous because it can reinforce stereotypes through applications that employ these models.\n\nRun the cell below, to examine (a) which terms are most similar to \"woman\" and \"worker\" and most dissimilar to \"man\", and (b) which terms are most similar to \"man\" and \"worker\" and most dissimilar to \"woman\". Point out the difference between the list of female-associated words and the list of male-associated words, and explain how it is reflecting gender bias.", "_____no_output_____" ] ], [ [ "# Run this cell\n# Here `positive` indicates the list of words to be similar to and `negative` indicates the list of words to be\n# most dissimilar from.\npprint.pprint(wv_from_bin.most_similar(positive=['woman', 'worker'], negative=['man']))\nprint()\npprint.pprint(wv_from_bin.most_similar(positive=['man', 'worker'], negative=['woman']))", "[('employee', 0.6375863552093506),\n ('workers', 0.6068919897079468),\n ('nurse', 0.5837947726249695),\n ('pregnant', 0.5363885164260864),\n ('mother', 0.5321309566497803),\n ('employer', 0.5127025842666626),\n ('teacher', 0.5099576711654663),\n ('child', 0.5096741914749146),\n ('homemaker', 0.5019454956054688),\n ('nurses', 0.4970572590827942)]\n\n[('workers', 0.6113258004188538),\n ('employee', 0.5983108282089233),\n ('working', 0.5615328550338745),\n ('laborer', 0.5442320108413696),\n ('unemployed', 0.5368517637252808),\n ('job', 0.5278826951980591),\n ('work', 0.5223963260650635),\n ('mechanic', 0.5088937282562256),\n ('worked', 0.505452036857605),\n ('factory', 0.4940453767776489)]\n" ] ], [ [ "The word most similar to \"woman\" and \"worker\" and most dissimilar to \"man\" is nurses.\nMost nurses are woman.\n\nThe word most similar to \"man\" and \"worker\" and most dissimilar to \"woman\" is factory.\nThe factory have many worker man.\n", "_____no_output_____" ], [ "### Question 2.7: Independent Analysis of Bias in Word Vectors [code + written] (1 point)\n\nUse the `most_similar` function to find another case where some bias is exhibited by the vectors. Please briefly explain the example of bias that you discover.", "_____no_output_____" ] ], [ [ "# ------------------\n# Write your implementation here.\npprint.pprint(wv_from_bin.most_similar(positive=['elephant', 'skyscraper'], negative=['ant']))\nprint()\npprint.pprint(wv_from_bin.most_similar(positive=['motorcycle', 'car'], negative=['bicycle']))\n# ------------------", "[('tower', 0.49941301345825195),\n ('skyscrapers', 0.48599374294281006),\n ('tallest', 0.46377506852149963),\n ('statue', 0.4558914303779602),\n ('towers', 0.44428494572639465),\n ('40-story', 0.4247894287109375),\n ('monument', 0.4171640872955322),\n ('high-rise', 0.41149571537971497),\n ('bust', 0.408037006855011),\n ('gleaming', 0.40352344512939453)]\n\n[('cars', 0.6663373112678528),\n ('driver', 0.6263732314109802),\n ('vehicle', 0.6231670379638672),\n ('mercedes', 0.6017158627510071),\n ('truck', 0.581316351890564),\n ('driving', 0.5702999234199524),\n ('motorbike', 0.5668439865112305),\n ('bmw', 0.5605546236038208),\n ('vehicles', 0.5472753047943115),\n ('motorcycles', 0.5434989929199219)]\n" ] ], [ [ "elephant is very high and ant is very small, skyscraper is very high but the most similairty also high.\n\nThe motocycle and bicycle is two rounds.", "_____no_output_____" ], [ "### Question 2.8: Thinking About Bias [written] (2 points)\n\nWhat might be the causes of these biases in the word vectors? You should give least 2 explainations how bias get into the word vectors. How might you be able to investigate/test these causes?", "_____no_output_____" ], [ "From the training processing, in my opinion, the bias is base on the co-occurrence.\nDifference word have difference co-occurrence.", "_____no_output_____" ], [ "# <font color=\"blue\"> Submission Instructions</font>\n\n1. Click the Save button at the top of the Jupyter Notebook.\n2. Select Cell -> All Output -> Clear. This will clear all the outputs from all cells (but will keep the content of all cells). \n2. Select Cell -> Run All. This will run all the cells in order, and will take several minutes.\n3. Once you've rerun everything, select File -> Download as -> PDF via LaTeX (If you have trouble using \"PDF via LaTex\", you can also save the webpage as pdf. <font color='blue'> Make sure all your solutions especially the coding parts are displayed in the pdf</font>, it's okay if the provided codes get cut off because lines are not wrapped in code cells).\n4. Look at the PDF file and make sure all your solutions are there, displayed correctly. The PDF is the only thing your graders will see!\n5. Submit your PDF on Gradescope.", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown" ]

[ [ "markdown" ], [ "code" ], [ "markdown", "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code", "code" ], [ "markdown", "markdown" ], [ "code", "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown", "markdown", "markdown", "markdown" ] ]

[ "MIT" ]

[ "MIT" ]