difference between results of transE(paper) and transE(our)

thunlp / KB2E

Knowledge Graph Embeddings including TransE, TransH, TransR and PTransE

MIT License

1.4k stars 449 forks source link

difference between results of transE(paper) and transE(our) #24

Closed wanpifeng4ever closed 7 years ago

wanpifeng4ever commented 7 years ago

Hi, could anyone please explain what causes the difference between transE(paper) and transE(our) in the evaluation results? Why transE(our) outperforms transE(paper) on hit@10 of FB15K that much?

zhangyi85 commented 7 years ago

第一，两人的代码不同，结果肯定有差异第二，根据林博士在paperweekly上的视频，原作者可能在实验上没有优化参数（包括维度，间隔等）

wanpifeng4ever commented 7 years ago

谢谢指教！可否再请教一下，貌似transR没有论文说的表现那么好？我的实验中，transR的表现比transE和transH都要差很多，不知道是否是我的代码有问题还是transR的训练难收敛？

Mrlyk423 commented 7 years ago

TransR需要使用TransE进行初始化，否则收敛不太好。