Awesome
Hyper-parameters for reproducing the reported results are provided in the train_QuatE_dataset.py.
How to run
Requirements: Pytorch 1.4+
STEP:
- export LD_LIBRARY_PATH=/usr/local/lib:/usr/lib:/usr/local/lib64:/usr/lib64
- CUDA_VISIBLE_DEVICES=0 python3 train_QuatE_dataset.py
Citation
@article{zhang2019quaternion,
title={Quaternion Knowledge Graph Embedding},
author={Zhang, Shuai and Tay, Yi and Yao, Lina and Liu, Qi},
journal={arXiv preprint arXiv:1904.10281},
year={2019}
}
This code is based on the OpenKE project.