Why the build model part of these two files (P_cause.py and P_emotion.py) is the same, it should be different from the paper.

[guo-huojian]

[bbruceyuan]

General Response

hi, they are actually different. The reason is that the source code is not very readable of the build_model part. The order of return values of function build_model is different in P_emotion.py and P_cause.py. In P_emotion.py, it's return pred_pos, pred_cause, reg. And in P_cause.py, it's return pred_cause, pred_pos, reg. As a result, model implementation is the same with paper. It's just an issue about the naming problem. If you want to find a readable version of ECPE implementation, my PyTorch implementation might be helpful, and the link is https://github.com/bbruceyuan/ECPE-PyTorch.

Detail explanation

the scope name in P_emotion.py and P_cause.py are different.

u will find in P_cause.py, the author predict cause first and concatenates predict cause logits with clause representation to predict emotion.

While in P_emotion.py, the author predicts emotion first and concatenates predict emotion logits with clause representation to predict cause.