readme.md

Graph Neural Networks with Feature-wise Linear Modulation (GNN-FiLM)

• Paper link: https://arxiv.org/abs/1906.12192
• Author's code repo: [https://github.com/ Microsoft/tf-gnn-samples](https://github.com/ Microsoft/tf-gnn-samples)

Dataset Statics

Dataset	#Graphs	# Nodes	# Edges	# Classes
PPI	20	44906	1226368	121

Refer to Planetoid.

Results

TL_BACKEND="tensorflow" python film_trainer.py --dataset ppi --lr 0.001 --l2_coef 5e-4 --drop_rate 0.1 --hidden_dim 160  --num_layers 4
TL_BACKEND="torch" python film_trainer.py --dataset ppi --lr 0.001 --l2_coef 5e-4 --drop_rate 0.1 --hidden_dim 160 --num_layers 4
TL_BACKEND="paddle" python film_trainer.py --dataset ppi --lr 0.001 --l2_coef 5e-4 --drop_rate 0.1 --hidden_dim 160 --num_layers 4

Dataset	Paper	Our(tf)	Our(torch)	Our(pd)
PPI	84.1±0.5	70±0.7	94±0.7	70±0.7