Composition-Based Multi-Relational Graph Convolutional Networks

An implement of CompGCN in Pytorch and DGL.

• Paper: ICLR 2020: Composition-Based Multi-Relational Graph Convolutional Networks
• Author's code: https://github.com/malllabiisc/CompGCN

Dependencies

1. install Python3
2. install requirements pip install -r requirements.txt

Train Model

To start training process:

python run.py --score_func conve --opn corr --gpu 0 --epoch 500 --batch 256 --n_layer 1

• --score_func denotes the link prediction score score function
  • conve : Convolutional 2D Knowledge Graph Embeddings
  • distmult : Embedding Entities and Relations for Learning and Inference in Knowledge Bases
• --opn is the composition operation used in CompGCN. It can take the following values:
  • sub for subtraction operation: Φ(e_s, e_r) = e_s - e_r
  • mult for multiplication operation: Φ(e_s, e_r) = e_s * e_r
  • corr for circular-correlation: Φ(e_s, e_r) = e_s ★ e_r
• --gpu for specifying the GPU to use
• --epoch for number of epochs
• --batch for batch size
• --n_layer for number of GCN Layers to use
• Rest of the arguments can be listed using python run.py -h

Test Result

FB15k-237