search

ICT-GIMLab / SeHGNN

94 stars 16 forks source link

Request for ComplEx Embedding #3

Closed weikai-li closed 2 years ago

weikai-li commented 2 years ago

Hello! I generated the ComplEx embedding as you instructed and ran your code. I basically reproduced your results of "SeHGNN" but could not reproduce the results of "ComplEx + SeHGNN." My first-stage result is about 55.83(valid set) and 54.12(test set), much lower than yours. I wonder whether it has something to do with the ComplEx embedding. Could you please provide your generated ComplEx embedding? Thank you!

Yangxc13 commented 2 years ago

Thanks for your attention and help. Please change the '--max_step 10' (number of maximum training steps) to a large number such as 500000 in the file ./data/complex_nars/train_graph_emb.sh for embedding generation. This bug has been fixed in the lastest commit.

About why this bug occurred: Before releasing the codes, we decreased this number for quickly verifying the smoothness of the whole training process, but sorry we forgot to change it back.

weikai-li commented 2 years ago

Thank you! I have successfully reproduced your results.

weikai-li commented 2 years ago

Thank you! I have successfully reproduced your results.

在2022-08-09 @.***写道：

Thanks for your attention and help. Please change the '--max_step 10' (number of maximum training steps) to a large number such as 500000 in the file ./data/complex_nars/train_graph_emb.sh for embedding generation. This bug has been fixed in the lastest commit.

About why this bug occurred: Before releasing the codes, we decreased this number for quickly verifying the smoothness of the whole training process, but sorry we forgot to change it back.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>