Inductive Graph-based Matrix Completion

Paper link: https://arxiv.org/abs/1904.12058 Author's code: https://github.com/muhanzhang/IGMC

Dependencies

• PyTorch 1.2+
• DGL 0.5 (nightly version)

Data

Supported datasets: ml-100k, ml-1m

How to run

• ml-100k

python3 train.py --data_name ml-100k --testing \
                 --batch_size 32 --edge_dropout 0.2 --max_nodes_per_hop 200 --train_epochs 80 \
                 --device 0

• ml-1m

python3 train_multi_gpu.py --data_name ml-1m --testing \
                --batch_size 32 --edge_dropout 0. --max_nodes_per_hop 100 --train_epochs 40 \
                --train_log_interval 1000 --valid_log_interval 5 --train_lr_decay_step 20 \
                --gpu 0,1,2,3

Results

Dataset	Our code best of epochs	Author code best of epochs / ensembled
ml-100k	0.9053	0.9053 / 0.9051
ml-1m	0.8679	0.8685 / 0.8558