Repository navigation
Best parameters for numbers of sampled rules and important rules #2
Description
The setup in your paper introduces 2000 sampled rules and 1000 important rules and I confuse about which dataset these parameters refer to. And the command lines given in README are also different with setups in you paper. Could you please release the details about the numbers of sampled rules and important rules during pre-training and training process? Thanks a lot!
Activity
mnqu commentedon Jun 12, 2021
Thanks for your interest!
The current codes in the main branch give a simple implementation of RNNLogic, which is easier to use but the hyperparameters are different from those reported in the paper.
We have just published a new branch called torch, which corresponds to the implementation used in our experiment, and you could use the same hyperparameters as reported in the paper to run this model.
We will add more detailed README and comments for this new branch.
Thanks!
Meng
csholder commentedon Jun 13, 2021
Thank you for your reply, I will follow torch branch continuously. This work is really motivated to me, and I think it could reason facts with inductive reasoning method and deductive reasoning method simultaneously. I also focus on your other work and benefit a lot. Hope your better work!
Best wishes!
He Tao
mnqu commentedon Jun 14, 2021
Thanks He! Also look forward to your work in this direction : )
Best regards!
Meng