Datasets:

Modalities:
Image
Text
Formats:
parquet
ArXiv:
Libraries:
Datasets
Dask
License:
fgranqvist commited on
Commit
227b063
1 Parent(s): a37706e

Upload folder using huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +20 -1
README.md CHANGED
@@ -74,12 +74,31 @@ This repo also provide scripts to explore the images and labels in more detail.
74
 
75
  ## Prepare the dataset in HDF5
76
  We found a single HDF5 file to be efficient for FL.
77
- If you want to process the dataset for general usage in FL, we recommend using [this preprocessing script](https://github.com/apple/pfl-research/blob/develop/benchmarks/dataset/flair/prepare_dataset.py) to construct a HDF5 file.
78
 
79
  By default the script will group the images and labels by train/val/test split and then by user ids, making it suitable for federated learning experiments.
80
  With the flag `--not_group_data_by_user`, the script will simply group the images and labels by train/val/test split and ignore the user ids, which is the typical setup for centralized training. \
81
  ⚠️ Warning: the hdf5 file take up to ~80GB disk space to store after processing.
82
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
83
  ## Disclaimer
84
  The annotations and Apple’s other rights in the dataset are licensed under CC-BY-NC 4.0 license.
85
  The images are copyright of the respective owners, the license terms of which can be found using the links provided in ATTRIBUTIONS.TXT (by matching the Image ID).
 
74
 
75
  ## Prepare the dataset in HDF5
76
  We found a single HDF5 file to be efficient for FL.
77
+ If you want to process the dataset for general usage in FL, we recommend using [this preprocessing script](https://github.com/apple/pfl-research/blob/develop/benchmarks/dataset/flair/download_preprocess.py) to construct a HDF5 file.
78
 
79
  By default the script will group the images and labels by train/val/test split and then by user ids, making it suitable for federated learning experiments.
80
  With the flag `--not_group_data_by_user`, the script will simply group the images and labels by train/val/test split and ignore the user ids, which is the typical setup for centralized training. \
81
  ⚠️ Warning: the hdf5 file take up to ~80GB disk space to store after processing.
82
 
83
+ ## Use dataset directly with HuggingFace
84
+
85
+ The dataset can also be used with the `datasets` package.
86
+ To group datapoints by user, simply construct a mapping and then query the dataset by index:
87
+ ```
88
+ from datasets import load_dataset
89
+ from collections import defaultdict
90
+ ds = load_dataset('apple/flair', split='val')
91
+
92
+ user_to_ix = defaultdict(list)
93
+ for i, record in enumerate(ds):
94
+ user_to_ix[record['user_id']].append(i)
95
+
96
+ def load_user_data(user_id):
97
+ return [ds[i] for i in user_to_ix[user_id]]
98
+
99
+ load_user_data('81594342@N00')
100
+ ```
101
+
102
  ## Disclaimer
103
  The annotations and Apple’s other rights in the dataset are licensed under CC-BY-NC 4.0 license.
104
  The images are copyright of the respective owners, the license terms of which can be found using the links provided in ATTRIBUTIONS.TXT (by matching the Image ID).