File size: 15,933 Bytes
f3972ea |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 |
import numpy as np
import pickle
from torch.utils.data import Dataset, DataLoader
import os
import torch
from copy import deepcopy
from blimpy import Waterfall
from tqdm import tqdm
from copy import deepcopy
from sigpyproc.readers import FilReader
from torch import nn
def load_pickled_data(file_path):
with open(file_path, 'rb') as f:
data = pickle.load(f)
return data
# Custom dataset class
class CustomDataset(Dataset):
def __init__(self, data_dir, bit8=False, transform=None):
self.data_dir = data_dir
self.transform = transform
self.images = []
self.labels = []
self.classes = os.listdir(data_dir)
self.class_to_idx = {cls: idx for idx, cls in enumerate(self.classes)}
self.bit8 = bit8
# Load images and labels
for cls in self.classes:
class_dir = os.path.join(data_dir, cls)
for image_name in os.listdir(class_dir):
image_path = os.path.join(class_dir, image_name)
self.images.append(image_path)
self.labels.append(self.class_to_idx[cls])
def __len__(self):
return len(self.images)
def __getitem__(self, idx):
image_path = self.images[idx]
label = self.labels[idx]
# Load image
image = load_pickled_data(image_path)
if self.transform is not None:
if self.bit8 == True:
new_image = self.transform(torch.from_numpy(image['8_data']).type(torch.float32))
else:
new_image = self.transform(torch.from_numpy(image['data']))
# new_image = self.transform(image['data'])
return new_image, label
# Custom dataset class
class CustomDataset_Masked(Dataset):
def __init__(self, data_dir, transform=None):
self.data_dir = data_dir
self.transform = transform
self.images = []
self.labels = []
self.classes = os.listdir(data_dir)
self.class_to_idx = {cls: idx for idx, cls in enumerate(self.classes)}
# Load images and labels
for cls in self.classes:
class_dir = os.path.join(data_dir, cls)
for image_name in os.listdir(class_dir):
image_path = os.path.join(class_dir, image_name)
self.images.append(image_path)
self.labels.append(self.class_to_idx[cls])
def __len__(self):
return len(self.images)
def __getitem__(self, idx):
image_path = self.images[idx]
label = self.labels[idx]
# Load image
image = load_pickled_data(image_path)
if self.transform is not None:
if image['burst'].max() ==0:
new_burst = torch.from_numpy(image['burst'])
else:
new_burst = torch.from_numpy(image['burst']/image['burst'].max())
ind = new_burst > 0.1
ind_not = new_burst <= 0.1
new_burst[ind] = 1
new_burst[ind_not] = 0
new_image = self.transform(torch.from_numpy(image['data'].data))
new_burst_arr = torch.zeros_like(new_image)
new_burst_arr[ 0, :,:] = new_burst
new_burst_arr[ 1, :,:] = new_burst
new_burst_arr[ 2, :,:] = new_burst
return new_image, label, new_burst_arr
# Custom dataset class
class TestingDataset(Dataset):
def __init__(self, data_dir, bit8=False, transform=None):
self.data_dir = data_dir
self.transform = transform
self.images = []
self.labels = []
self.classes = os.listdir(data_dir)
self.class_to_idx = {cls: idx for idx, cls in enumerate(self.classes)}
self.bit8 = bit8
# Load images and labels
for cls in self.classes:
class_dir = os.path.join(data_dir, cls)
for image_name in os.listdir(class_dir):
image_path = os.path.join(class_dir, image_name)
self.images.append(image_path)
self.labels.append(self.class_to_idx[cls])
def __len__(self):
return len(self.images)
def __getitem__(self, idx):
image_path = self.images[idx]
label = self.labels[idx]
# Load image
image = load_pickled_data(image_path)
params = image['params']
if self.transform is not None:
params = image['params']
if self.bit8 == True:
new_image = self.transform(torch.from_numpy(image['8_data']).type(torch.float32))
else:
new_image = self.transform(torch.from_numpy(image['data']))
params['labels'] = label
return new_image, (label, params['dm'], params['freq_ref'], params['snr'], params['boxcard'])
# Custom dataset class
class SearchDataset(Dataset):
def __init__(self, data_dir, transform=None, pickle_data=False):
self.window_size = 2048
if pickle_data:
with open(data_dir, 'rb') as f:
self.d = pickle.load(f)
self.header = self.d['header']
self.images = self.crop(self.d['data'][:,0,:], self.window_size)
else:
self.obs = Waterfall(data_dir, max_load = 50)
self.header = self.obs.header
self.images = self.crop(self.obs.data[:,0,:], self.window_size)
self.transform = transform
self.SEC_PER_DAY = 86400
def crop(self, data, window_size = 2048):
n_samp = data.shape[0]//window_size
new_data = np.zeros((n_samp, window_size, 192 ))
for i in range(n_samp):
new_data[i, :,:] = data[ i*window_size : (i+1)*window_size, :]
return new_data
def __len__(self):
return self.images.shape[0]
def __getitem__(self, idx):
data = self.images[idx, :, :].T
tindex = idx * self.window_size
time = self.header['tsamp'] * tindex / self.SEC_PER_DAY + self.header['tstart']
if self.transform is not None:
new_image = self.transform(data)
return new_image, idx
# Custom dataset class
class SearchDataset_Sigproc(Dataset):
def __init__(self, data_dir, transform=None):
self.window_size = 2048
fil = FilReader(data_dir)
self.header = fil.header
# print("check shape ",fil.read_block(0, fil.header.nsamples).shape)
read_data = fil.read_block(0, fil.header.nsamples)[:,1024:-1024]
read_data = np.swapaxes(read_data, 0,-1)
self.images = self.crop(read_data, self.window_size)
self.transform = transform
self.SEC_PER_DAY = 86400
def crop(self, data, window_size = 2048):
n_samp = data.shape[0]//window_size
new_data = np.zeros((n_samp, window_size, 192 ))
for i in range(n_samp):
new_data[i, :,:] = data[ i*window_size : (i+1)*window_size, :]
return new_data
def __len__(self):
return self.images.shape[0]
def __getitem__(self, idx):
data = self.images[idx, :, :].T
tindex = idx * self.window_size
time = self.header.tsamp * tindex / self.SEC_PER_DAY + self.header.tstart
if self.transform is not None:
new_image = self.transform(torch.from_numpy(data))
return new_image, idx
# def renorm(data):
# shifted = data - data.min()
# shifted = shifted/shifted.max()
# return shifted
def renorm(data):
mean = torch.mean(data)
std = torch.std(data)
# Standardize the data
standardized_data = (data - mean) / std
return standardized_data
def transform(data):
copy_data = data.detach().clone()
rms = torch.std(data)
mean = torch.mean(data)
masks_rms = [-1, 5]
new_data = torch.zeros((len(masks_rms)+1, data.shape[0], data.shape[1]))
new_data[0,:,:] = renorm(torch.log10(copy_data+1e-10))
for i in range(1, len(masks_rms)+1):
scale = masks_rms[i-1]
copy_data = data.detach().clone() #deepcopy(data)
if scale < 0:
ind = copy_data < abs(scale) * rms + mean
copy_data[ind] = 0
else:
ind = copy_data > (scale) * rms + mean
copy_data[ind] = 0
new_data[i,:,:] = renorm(torch.log10(copy_data+1e-10))
new_data = new_data.type(torch.float32)
slices = torch.chunk(new_data, 8, dim=-1) # dim=1 is the height dimension
new_data = torch.stack(slices, dim=1) # New axis is inserted at dim=1
new_data = new_data.view(-1, new_data.size(2), new_data.size(3))
return new_data
def renorm_batched(data):
mean = torch.mean(data, dim=tuple(range(1, data.ndim)), keepdim=True)
std = torch.std(data, dim=tuple(range(1, data.ndim)), keepdim=True)
standardized_data = (data - mean) / std
return standardized_data
def transform_batched(data):
copy_data = data.detach().clone()
rms = torch.std(data, dim=tuple(range(1, data.ndim)), keepdim=True) # Batch-wise std
mean = torch.mean(data, dim=tuple(range(1, data.ndim)), keepdim=True) # Batch-wise mean
masks_rms = [-1, 5]
# Prepare the new_data tensor
num_masks = len(masks_rms) + 1
new_data = torch.zeros((num_masks, *data.shape), device=data.device) # Shape: (num_masks, batch_size, ..., ...)
# First layer: Apply renorm(log10(copy_data + epsilon))
new_data[0] = renorm_batched(torch.log10(copy_data + 1e-10))
for i, scale in enumerate(masks_rms, start=1):
copy_data = data.detach().clone()
# Apply masking based on the scale
if scale < 0:
ind = copy_data < abs(scale) * rms + mean
else:
ind = copy_data > scale * rms + mean
copy_data[ind] = 0
# Renormalize and log10 transform
new_data[i] = renorm_batched(torch.log10(copy_data + 1e-10))
# Convert to float32
new_data = new_data.type(torch.float32)
# Chunk along the last dimension and stack
slices = torch.chunk(new_data, 8, dim=-1) # Adjust for batch-wise slicing
new_data = torch.stack(slices, dim=2) # Insert a new axis at dim=1
new_data = torch.swapaxes(new_data, 0,1)
# Reshape into final format
new_data = new_data.reshape( new_data.size(0), 24, new_data.size(3), new_data.size(4)) # Flatten batch and mask dimensions
return new_data
class preproc(nn.Module):
def forward(self, x, flip=True):
if flip:
transform_batched(torch.flip(x, dims = (-2,)))
else:
transform_batched(x)
return template
# class preproc_debug(nn.Module):
# def forward(self, x):
# template = torch.zeros((32, 24, 192, 256))
# # for i in torch.arange(x.shape[0]): # Use a tensor-based range
# template[0,:,:,:] = transform_debug(torch.flip(x[0,:,:], dims = (0,)))
# template[1,:,:,:] = transform_debug(torch.flip(x[1,:,:], dims = (0,)))
# template[2,:,:,:] = transform_debug(torch.flip(x[2,:,:], dims = (0,)))
# template[3,:,:,:] = transform_debug(torch.flip(x[3,:,:], dims = (0,)))
# template[4,:,:,:] = transform_debug(torch.flip(x[4,:,:], dims = (0,)))
# template[5,:,:,:] = transform_debug(torch.flip(x[5,:,:], dims = (0,)))
# template[6,:,:,:] = transform_debug(torch.flip(x[6,:,:], dims = (0,)))
# template[7,:,:,:] = transform_debug(torch.flip(x[7,:,:], dims = (0,)))
# template[8,:,:,:] = transform_debug(torch.flip(x[8,:,:], dims = (0,)))
# template[9,:,:,:] = transform_debug(torch.flip(x[9,:,:], dims = (0,)))
# template[10,:,:,:] = transform_debug(torch.flip(x[10,:,:], dims = (0,)))
# template[11,:,:,:] = transform_debug(torch.flip(x[11,:,:], dims = (0,)))
# template[12,:,:,:] = transform_debug(torch.flip(x[12,:,:], dims = (0,)))
# template[13,:,:,:] = transform_debug(torch.flip(x[13,:,:], dims = (0,)))
# template[14,:,:,:] = transform_debug(torch.flip(x[14,:,:], dims = (0,)))
# template[15,:,:,:] = transform_debug(torch.flip(x[15,:,:], dims = (0,)))
# template[16,:,:,:] = transform_debug(torch.flip(x[16,:,:], dims = (0,)))
# template[17,:,:,:] = transform_debug(torch.flip(x[17,:,:], dims = (0,)))
# template[18,:,:,:] = transform_debug(torch.flip(x[18,:,:], dims = (0,)))
# template[19,:,:,:] = transform_debug(torch.flip(x[19,:,:], dims = (0,)))
# template[20,:,:,:] = transform_debug(torch.flip(x[20,:,:], dims = (0,)))
# template[21,:,:,:] = transform_debug(torch.flip(x[21,:,:], dims = (0,)))
# template[22,:,:,:] = transform_debug(torch.flip(x[22,:,:], dims = (0,)))
# template[23,:,:,:] = transform_debug(torch.flip(x[23,:,:], dims = (0,)))
# template[24,:,:,:] = transform_debug(torch.flip(x[24,:,:], dims = (0,)))
# template[25,:,:,:] = transform_debug(torch.flip(x[25,:,:], dims = (0,)))
# template[26,:,:,:] = transform_debug(torch.flip(x[26,:,:], dims = (0,)))
# template[27,:,:,:] = transform_debug(torch.flip(x[27,:,:], dims = (0,)))
# template[28,:,:,:] = transform_debug(torch.flip(x[28,:,:], dims = (0,)))
# template[29,:,:,:] = transform_debug(torch.flip(x[29,:,:], dims = (0,)))
# template[30,:,:,:] = transform_debug(torch.flip(x[30,:,:], dims = (0,)))
# template[31,:,:,:] = transform_debug(torch.flip(x[31,:,:], dims = (0,)))
# return template
# def transform_debug(data):
# copy_data = data.detach().clone()
# rms = torch.std(data)
# mean = torch.mean(data)
# masks_rms = [-1, 5]
# new_data = torch.zeros((len(masks_rms)+1, data.shape[0], data.shape[1]))
# new_data[0,:,:] = renorm(torch.log10(copy_data+1e-10))
# for i in range(1, len(masks_rms)+1):
# scale = masks_rms[i-1]
# copy_data = data.detach().clone()
# if scale < 0:
# ind = copy_data < abs(scale) * rms + mean
# copy_data[ind] = 0
# else:
# ind = copy_data > (scale) * rms + mean
# copy_data[ind] = 0
# new_data[i,:,:] = renorm(torch.log10(copy_data+1e-10))
# new_data = new_data.type(torch.float32)
# slices = torch.chunk(new_data, 8, dim=-1) # dim=1 is the height dimension
# new_data = torch.stack(slices, dim=1) # New axis is inserted at dim=1
# new_data = new_data.view(-1, new_data.size(2), new_data.size(3))
# return new_data
def renorm_batched(data):
mins = torch.amin(data, (-2, -1))
mins = mins.unsqueeze(1).unsqueeze(2)
mins = mins.expand(data.shape[0], 192, 2048)
shifted = data - mins
maxs = torch.amax(shifted, (-2, -1))
maxs = maxs.unsqueeze(1).unsqueeze(2)
maxs = maxs.expand(data.shape[0], 192, 2048)
shifted = shifted/maxs
return shifted
def transform_mask(data):
copy_data = deepcopy(data)
shift = copy_data - copy_data.min()
normalized_data = shift / shift.max()
new_data = np.zeros((3, data.shape[0], data.shape[1]))
for i in range(3):
new_data[i,:,:] = normalized_data
new_data = new_data.astype(np.float32)
return new_data
#Function to Convert to ONNX
def Convert_ONNX(model, saveloc, input_data_mock):
print("Saving to ONNX")
# set the model to inference mode
model.eval()
# Let's create a dummy input tensor
dummy_input = torch.autograd.Variable(input_data_mock)
# Export the model
torch.onnx.export(model, # model being run
dummy_input, # model input (or a tuple for multiple inputs)
saveloc, # where to save the model
input_names = ['modelInput'], # the model's input names
output_names = ['modelOutput'], # the model's output names
dynamic_axes={'modelInput' : {0 : 'batch_size'}, # variable length axes
'modelOutput' : {0 : 'batch_size'}} )
print(" ")
print('Model has been converted to ONNX')
|