Hetero$^2$Net

Implementation of Hetero$^2$Net from the "Hetero$^2$Net: Heterophily-aware Representation Learning on Heterogenerous Graphs" paper.

Requirements

• torch 2.0.1
• torch_geometric 2.4.0
• torch-cluster 1.6.1+pt20cu117
• torch-scatter 2.1.1+pt20cu117
• torch-sparse 0.6.17+pt20cu117
• torch-spline-conv 1.2.2+pt20cu117
• texttable 1.6.7
• termcolor 1.1.0
• scikit-learn 1.0.2
• scipy 1.10.1
• numpy 1.22.4
• tqdm 4.64.0
• CUDA 11.7

Reproduction

DBLP

python main.py --dataset DBLP --dropout 0.5 --lr 0.001 --beta 0.2 --alpha 0.1

IMDB

python main.py --dataset IMDB --dropout 0.9 --lr 0.005 --alpha 0.1 --mask_lp --p 0.6

ACM

python main.py --dataset ACM --epochs 100 --monitor loss --mask_lp --p 0.7

OGB

python main.py --dataset MAG --num_neighbors 15 15 --hidden 256 --beta 0.2 --mini_batch --mask_lp --p 1.0

RCDD

python main.py --dataset RCDD --metrics ap micro-f1 macro-f1 --dropout 0.5 --epochs 100 --num_neighbors 15 15 --hidden 256 --beta 0.2 --mini_batch --mask_lp