pip install requirement.txt

python setup.py develop

import lazygcn

set dataset directory

os.environ["GNN_DATASET_DIR"] = "<your dataset directory>"

dataset = data.Dataset('pubmed')

Supported datasets:

• Cora
• Citeseer
• Pubmed
• Flickr, PPI, PPI-Large, Reddit, Yelp and Amazon2M

There are several predefined training class, which prepare data and create pre-defined models and optimizer.

• Models: GCN, MLP, RecGCN [GraphSAGE, GAT]*
• Layers: GConv, Linear, RecGConv
• Minibatch: Full, Random, Split
• NodeSampler: LayerWise, LADIES, Exact, FullGraph [FastGCN, NodeWise]*

See notebooks folder for examples and lazygcn/training for pre-defined trainer.