The results in the paper cannot be reproduced

XingBowen714 / DARER

PyTorch source Code for our ACL 2022 paper: DARER: Dual-task Temporal Relational Recurrent Reasoning Network for Joint Dialog Sentiment Classification and Act Recognition.

14 stars 2 forks source link

The results in the paper cannot be reproduced #2

Open chenxiaoting-0613 opened 2 years ago

chenxiaoting-0613 commented 2 years ago

The reproduced results are quite different from the results given in the paper. I tried to adjust the parameters, and the results have improved, but there is still a gap with the results in the paper. Based on this problem, do you have any good suggestions?

XingBowen714 commented 2 years ago

On different computation devices and environments, usually, the optimal settings of hyperparameters are different. Maybe you can tune the hyperparameters in your own environment. Some other researchers, who communicated with me via email about this problem, can reproduce the experiment results after tuning the hyperparameters.