mgnni

MGNNI: Multiscale Graph Neural Networks with Implicit Layers (NGNNI)

• Paper link: https://arxiv.org/abs/2210.08353
• Author's code repo: https://github.com/liu-jc/MGNNI. Note that the original code is implemented with Pytorch for the paper.

Dataset Statics

Dataset	# Nodes	# Edges	# Classes
Cornell	183	280	5
Texas	183	295	5
Wisconsin	251	466	5

Refer to WebKB.

Results

TL_BACKEND="paddle" python mgnni_trainner.py --dataset cornell --lr 0.5 --l2_coef 5e-6 --model MGNNI_m_att --ks [1,2] --epochs 300
TL_BACKEND="paddle" python mgnni_trainner.py --dataset texas --lr 0.5 --l2_coef 5e-6 --model MGNNI_m_att --ks [1,2] --epochs 300
TL_BACKEND="paddle" python mgnni_trainner.py --dataset wisconsin --lr 0.5 --l2_coef 5e-6 --model MGNNI_m_att --ks [1,2] --epochs 300
TL_BACKEND="tensorflow" python mgnni_trainner.py --dataset cornell --lr 0.5 --l2_coef 5e-6 --model MGNNI_m_att --ks [1,2] --epochs 300
TL_BACKEND="tensorflow" python mgnni_trainner.py --dataset texas --lr 0.5 --l2_coef 5e-6 --model MGNNI_m_att --ks [1,2] --epochs 300
TL_BACKEND="tensorflow" python mgnni_trainner.py --dataset wisconsin --lr 0.5 --l2_coef 5e-6 --model MGNNI_m_att --ks [1,2] --epochs 300
TL_BACKEND="torch" python mgnni_trainner.py --dataset cornell --lr 0.5 --l2_coef 5e-6 --model MGNNI_m_att --ks [1,2] --epochs 300
TL_BACKEND="torch" python mgnni_trainner.py --dataset texas --lr 0.5 --l2_coef 5e-6 --model MGNNI_m_att --ks [1,2] --epochs 300
TL_BACKEND="torch" python mgnni_trainner.py --dataset wisconsin --lr 0.5 --l2_coef 5e-6 --model MGNNI_m_att --ks [1,2] --epochs 300 

Dataset	Paper	Our(pd)	Our(tf)	Our(th)	Our(ms)
Cornell	78.38	78.92±1.21	78.38±0.00	78.38±0.00
Texas	81.08	85.95±1.21	83.78±0.00	84.86±1.48
Wisconsin	82.35	83.53±1.07	82.35±0.00	83.92±0.88