Test method of sampling graph and gcn model on pure graph task
Example of running
Run graphsaint model on flicker dataset with sampling and loss normalization.
CUDA_VISIBLE_DEVICES=1 python main.py --dataset='reddit' --eval_sample=1 --batch_size=3000
Dataset support:
- Flickr
- Yelp(exist error)
- PPI(exist error)
Sample support:
- Graphsaint random walker sampler.
- Random node sampler.
TODO 06-07:
- Fix bias in sampling evaluation. Done
- Tensorboard for visualization.
- More sampler and more gcn net.
- Write own dataset loader for custom data.