Replies: 1 comment 1 reply
-
The relation matching (RM) module is the key. If you include that and run for larger epochs, I think you would get better results. RM module can be found in the original paper's repository. I hope that this clears your doubt. Also, a hyperparameter search could also help to filter out the useful ones. |
Beta Was this translation helpful? Give feedback.
1 reply
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Hi,
I get 0.68 in metaQA half KG-hop1. I'm using your command with the same hyperparameter setting. After the first epoch, I get 0.68, and then the score stays the same. After 10 epochs, the training automatically ends.0.68 is too low, I do not know what's wrong with it? I would like to ask whether you get the result 83.9 in the paper?
Beta Was this translation helpful? Give feedback.
All reactions