MetaCF & Score-aware MetaCF

Some of the codes are adopted from MetaCF

Requirements

Packages

• python==3.6.3
• pytorch==1.1.0
• numpy==1.17.0
• scikit-learn==0.24.2

Environment

• Ubuntu, VERSION="16.04.7 LTS (Xenial Xerus)

How to run

Training

• train.sh
  • directly run to train our score-aware MetaCF model
  • uncomment --original_model, --dot_prod to run the original model
  • change --epoch to load saved model
  • other configs' discriptions are in help information

Evaluation

• evaluate.sh:
  • directly run to evaluate our score-aware MetaCF model
  • uncomment --original_model, --dot_prod to evaluate the original model
  • change --epoch to load saved model
  • other configs' discriptions are in help information
  • we early stop after 2 epoches, and we fix the rank prediction after the first epoch

Details of each script

dataset.py

• Initialize and load the original kindle dataset and our movielens dataset
• Subgraph sampling with/without probability

main.py

• Training and evaluation procedure

model.py

• The model of GCN

self-use test scripts

• read_log.py
• testcode.py