ink-usc / cpl Goto Github PK

Did you release the Pytorch version?

I read the paper and have a question on how to assign a reward to the extractor? The reasoner gets the reward 1 if it reaches the correct target entity, and the intermediate reward is 0. You mention that the extractor receives the reward from the reasoner in the step-wise. But how can reasoner give the extractor reward in each of the steps, since the reasoner can only get the reward in the end-step?

I can't run these codes. It seems that this repo lacks some necessary checkpoint files. Besides, I don't know how to use data dealers to generate datasets (e.g., after_dataset in this file). Could you provide a detailed instruction for running these codes? Thank you very much.