Composition-based Multi-Relational Graph Convolutional Networks Paper: Composition-based Multi-Relational Graph Convolutional Networks Author's code for link prediction: https://github.com/malllabiisc/CompGCN Dataset Statics Dataset #Nodes #Edges #Relations #Labeled AIFB 8,285 58,086 90 176 MUTAG 23,644 148,454 46 340 BGS 333,845 1,832,398 206 146 Results TL_BACKEND="pytorch" python comgcn_trainer.py --dataset aifb --l2 5e-5 --op sub TL_BACKEND="pytorch" python comgcn_trainer.py --dataset mutag --lr 0.015 --l2_coef 5e-2 --op sub TL_BACKEND="pytorch" python comgcn_trainer.py --dataset bgs --lr 0.0001 --l2_coef 5e-2 TL_BACKEND="paddle" python comgcn_trainer.py --dataset aifb --l2 5e-5 --op sub TL_BACKEND="paddle" python comgcn_trainer.py --dataset mutag --lr 0.015 --l2_coef 5e-2 --op sub TL_BACKEND="paddle" python comgcn_trainer.py --dataset bgs --lr 0.0001 --l2_coef 5e-2 Dataset Paper Our(th) Our(pd) AIFB / 88.89(±2.78) 87.22(±1.36) MUTAG 85.3(±1.2) 82.0(±2.66) 81.47(±2.73) BGS / 77.93(±4.14) 75.17(±4.02)