Implementation MPDA with pytorch

This is my implementation of the paper MPDA(On-Device Learning for Model Personalization with Large-Scale Cloud-Coordinated Domain Adaption) with pytorch.

Config

First you need to config the config.yml in the root path.

MovieLens Dataset Preprocess

generate users with train json file and users with train and test data json file

nohup python -u scripts/preprocess/movielens/generate_user_with_train_and_test.py > ./log/generate_user_with_train_and_test.log 2>&1 &

generate user and item mapping

python scripts/preprocess/movielens/generate_mapping_file.py

generate recall item pairs

python scripts/preprocess/movielens/generate_recall_item_pairs.py

Amazon Dataset Preprocess

generate rating.csv with raw_data.json

python scripts/preprocess/amazon/generate_raw_data.py

Initial Model

train global model NCF on MovieLens

nohup python -u scripts/train_global_model.py -model=NCF -epochs=10 -dataset=movielens -device=cuda:2 > ./log/train_global_model.log 2>&1 &

train mask model on MovieLens

nohup python -u scripts/train_mask_model.py -device=cuda:2 > ./log/train_mask_model.log 2>&1 &

transfer model NCF

bash ./commands/ncf_movielens_50_random.sh