LiuYuemei111
commented
3 years ago 您是直接采用默认参数运行的么？ 我刚运行了一下Complex，结果与readme中报告的结果是差不多的
(ptuning) wanghuadong@node1:~/OpenKE$ python train_transe_FB15K237.py Input Files Path : ./benchmarks/FB15K237/ The toolkit is importing datasets. The total of relations is 237. The total of entities is 14541. The total of train triples is 272115. Input Files Path : ./benchmarks/FB15K237/ The total of test triples is 20466. The total of valid triples is 17535. Finish initializing... Epoch 999 | loss: 2.381454: 100%|████████████████████████████████████████████████████████████████████████████| 1000/1000 [20:19<00:00, 1.22s/it] 100%|████████████████████████████████████████████████████████████████████████████████████████████████████| 20466/20466 [00:18<00:00, 1091.37it/s] no type constraint results: metric: MRR MR hit@10 hit@3 hit@1 l(raw): 0.089113 561.105835 0.204779 0.086583 0.033470 r(raw): 0.251904 165.952652 0.441171 0.271817 0.160608 averaged(raw): 0.170508 363.529236 0.322975 0.179200 0.097039
l(filter): 0.189979 308.011200 0.362748 0.213720 0.105492 r(filter): 0.390304 139.963348 0.594400 0.441268 0.283495 averaged(filter): 0.290141 223.987274 0.478574 0.327494 0.194493 0.478574 0.4785742163658142
我跑了Complex，SImplE，DistMult，Anonogy ，效果都远达不到readme中的表格的效果。各位有谁可以公开超参数吗？