KE-X

Overview

Dependencies

• Compatible with PyTorch 1.8 and Python 3.7.x.
• Dependencies can be installed using requirements.txt.

Prepare:

• Obtain pretraining entity and relation embeddings for fb15k-237, NELL-995, and YAGO3-10 datasets. We recommend using OpenKE. You can also use your own embeddings, and please make sure that the indices of entities and relations correspond to the official dataset.
• Save the pretrained embeddings as .npy format with numpy, and rename the files as ent_embeddings.npy and rel_embeddings.npy.
• Remove the above two files to main/{datset}/{model}_preTrain folder.

Train:

python ./main/FB15K237/FB15K237_run.py --action train --model transe --relation_id 0 --batch_size 100 --epoch 100

Explain:

# Get metrics for three explanation methods
python ./main/FB15K237/FB15K237_run.py --action explain --model transe --relation_id 0

Visual:

# Get visual graph for target node
python ./main/FB15K237/FB15K237_run.py --action visual --model transe --relation_id 0 --node {target_node}

Visual example