adam1679 / connecte Goto Github PK

Thank you very much for your contribution. I really appreciate your work and the excellent performance.

I have cloned your code and run it on my server, but currently I faced difficulty to reproduce the desired result with TRT (Hit-at-10 = 79.9) mentioned in your paper.

Here my test result: 2021-05-30 06:17:31,973 - main - INFO - [ 800] VALID: MRR = 0.21/0.55, Mean Rank = 47.56/38.99, Hits@1 = 7.81/44.45, Hits@3 = 22.78/60.63, Hits@10 = 54.18/75.75
This result is generated by the original settings without any code modification.

It seems that the code is not your latest version. I am wondering if you can release your latest version, even if there are bugs or fine tunes need to be solved. This will help us a lot.

Your reply or code update will be greatly appreciated!

Thanks for your contribution, your work inspires me a lot. But I have some trouble with designing the optimizer when I try to reproduce your work. So it will help me a lot if you could provide more details about optimizer.py.

For example, the data in "FB15k_Entity_Type_train.txt" has two columns which are entity_id and its corresponding type.

It seems like the type columns' content are same with relation content in normal FB15k.
I am not sure if the relations information are seemd as entity type info directly.
Or the type information is crawled from the Freebase website?

Thanks.

thanks for your contribution, but now i want to predict type for unlabeled entity, if that means i cannot use this model.

how can i get source code from ETE，thanks!