Using Autoencoders for Knowledge Graph Embedding

Introduction

This is the Pytorch implementaion of some knowledge graph embedding models based on adversarial autoencoders. And we have test these models with drug-drug interaction datasets.

Dependencies

This project uses Python 3.5.3, with the following lib dependencies:

• Pytorch 1.3

• Numpy 1.15.1

• Sklearn 0.21

Usage

1. If you need to test your own dataset, please put it in ./dataset directory and format it according to our template.
2. Then you can run commend as follows to train/test/valid the models. For example, this command train a RotatE model on Deepddi dataset.

python3 -u main_rotate.py -ne 1000 -D_lr 0.5 -G_lr 0.001 -reg 0.3 -dataset Deepddi -emb_dim 200 -neg_ratio 1 -batch_size 512 -save_each 100 -discriminator_range 1