search

otiliastr / coper

Contextual Parameter Generation for Knowledge Graph Link Prediction
22 stars 7 forks source link

The result of WN18RR is different from the paper. #4

Open Koenigsegg-One1 opened 3 years ago

Koenigsegg-One1 commented 3 years ago

Dear Prof: I can't get the result of CoPER-ConvE on the dataset WN18RR. The parameters I used during the training process are just the parameters from the source file named 'config_WN18RR_cpg.yaml'. If there is something I missed, could you please point it out? Thanks a lot.

gstoica27 commented 3 years ago

Hi, Sorry for such a late response. If you're still facing this issue, could you please let us know what your results are? I reran the experiment with the WN18RR config and am obtaining 42.25% Hits@1. This is slightly better than what we reported.

Also what environment are you using? Unfortunately we have been seeing that different cuda/tf/etc.. versions can slightly improve/decrease performance.

I ran our experiment in python 3.7.7 with tf 1.15 and cuda 10.1