Datasets:

Formats:
parquet
Languages:
English
ArXiv:
Tags:
image
Libraries:
Datasets
Dask
License:
Nick Padgett commited on
Commit
a9b60c5
1 Parent(s): 47d10e2

Adding small lines for embeddings.

Browse files
Files changed (2) hide show
  1. README.md +2 -0
  2. tutorials/embeddings.md +20 -0
README.md CHANGED
@@ -47,6 +47,8 @@ The image files are all hosted in the AWS S3 bucket `pd12m`. The URLs to the ima
47
 
48
  [Downloading Images](./tutorials/images.md)
49
 
 
 
50
  # License
51
  The dataset is licensed under the [CDLA-Permissive-2.0](https://cdla.dev/permissive-2-0/).
52
 
 
47
 
48
  [Downloading Images](./tutorials/images.md)
49
 
50
+ [Working with the Embeddings](./tutorials/embeddings.md)
51
+
52
  # License
53
  The dataset is licensed under the [CDLA-Permissive-2.0](https://cdla.dev/permissive-2-0/).
54
 
tutorials/embeddings.md ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Working with the embeddings
2
+ The embeddings are available as numpy files, where each row is a 768-dimension floating point embedding. Each row is associated with it's matching index in the metadata files.
3
+
4
+ #### Open an embedding file
5
+ The files are in numpy format, and can be opened with `numpy` in Python.
6
+ ```python
7
+ import numpy as np
8
+ embeddings = np.load('pd12m.01.npy')
9
+ ```
10
+
11
+ #### Join Embeddings to Metadata
12
+ If you already have the metadata files loaded with pandas, you can join the embeddings to the dataframe simply.
13
+ ```python
14
+ df["embeddings"] = embeddings.tolist()
15
+ ```
16
+ Alternatively, you could use a 0-based index to access both the metadata and associated embedding.
17
+ ```python
18
+ i = 300
19
+ metadata = df.iloc[i]
20
+ embedding = embeddings[i]