edgify 0.0.1

Personalize DL models on the edge.

Model Personalization on Edge

Improving the performance of DL models for individual users by re-training on a user's data on the edge.

QMNIST Dataset

Divide the datset by the writer ID.

python preprocess/filter_by_user.py --download --dataset=train
python preprocess/filter_by_user.py --download --dataset=test

The resulting user-specific datasets will be under data/QMNIST/train and data/QMNIST/test. The file naming convention is w-<witer_id>.pth.

To load the dataset of a specific writer:

import torch
from torch.utils.data import DataLoader

dataset = torch.load('data/QMNIST/train/w-<writer_id>.pth')
dataloader = DataLoader(dataset)

License

BSD 3-Clause License.