When i run the pretrain.py in pytorch0.4 ,the best result (hits@10) of TransE model is about 10% .

cai-lw / KBGAN

Code for "KBGAN: Adversarial Learning for Knowledge Graph Embeddings" https://arxiv.org/abs/1711.04071

MIT License

212 stars 57 forks source link

When i run the pretrain.py in pytorch0.4 ,the best result (hits@10) of TransE model is about 10% . #3

Closed xitongdashi closed 6 years ago

xitongdashi commented 6 years ago

I don't know where come to wrong. However, with pytorch0.2 or 0.3v, the result is right w.r.t your paper. In fact , with only little change , your code will work with pytorch0.4 , but the result is not the same . I get confused. Might you help me ? I know this question is beyond your work , but i want to figure out the problem.

cai-lw commented 6 years ago

PyTorch changed its API significantly in version 0.4, so many API calls in the program can be affected. I am sorry that I cannot spot the problem, as problems could be everywhere.

Also, please note that I have no plan of updating this project for PyTorch 0.4 support.

qiuyumo208 commented 5 years ago

Pytorch0.3v also works fine?