dmlc--dgl
Minjie Wang
565f0c88fc
* refactor graph conv * docs & tests * fix lint * fix lint * fix lint * fix lint script * fix lint * Update * Style fix * Fix style * Fix style * Fix gpu case * Fix for gpu case * Hotfix edgesoftmax docs * Handle repeated features * Add docstring * Set default arguments * Remove dropout from nn.conv * Fix * add util fn for renaming * revert gcn_spmv.py * mx folder * fix wierd bug * fix mx * fix lint
Graph Attention Networks (GAT)
- Paper link: https://arxiv.org/abs/1710.10903
- Author's code repo (in Tensorflow): https://github.com/PetarV-/GAT.
- Popular pytorch implementation: https://github.com/Diego999/pyGAT.
Dependencies
- torch v1.0: the autograd support for sparse mm is only available in v1.0.
- requests
- sklearn
pip install torch==1.0.0 requests
How to run
Run with following:
python train.py --dataset=cora --gpu=0
python train.py --dataset=citeseer --gpu=0
python train.py --dataset=pubmed --gpu=0 --num-out-heads=8 --weight-decay=0.001
python train_ppi.py --gpu=0
Results
| Dataset | Test Accuracy | Time(s) | Baseline#1 times(s) | Baseline#2 times(s) |
|---|---|---|---|---|
| Cora | 84.0% | 0.0127 | 0.0982 (7.7x) | 0.0424 (3.3x) |
| Citeseer | 70.7% | 0.0123 | n/a | n/a |
| Pubmed | 78.1% | 0.0302 | n/a | n/a |
- All the accuracy numbers are obtained after 300 epochs.
- The time measures how long it takes to train one epoch.
- All time is measured on EC2 p3.2xlarge instance w/ V100 GPU.
- Baseline#1: https://github.com/PetarV-/GAT.
- Baseline#2: https://github.com/Diego999/pyGAT.