Qi Zhu

> Hey Yen-Chen! > > I am not sure what has been changed but maybe the following happens with the reward: > > https://github.com/thu-coai/ConvLab-2/blob/master/convlab2/evaluator/multiwoz_eval.py#L417 > > If you look here,...

@YenChen-Wu @thenickben @sherlock1987 @ChrisGeishauser We have re-trained PPO and updated the results (PR #211), thanks @aaa123git !

抱歉，我们不维护这个工具了，可能需要你自己调试

如果在同一台电脑的话，试试用两个不同的浏览器分别登录两个账号，再进行匹配

Select主要用于标识跨领域的约束 ，比如这里(Select, Hotel, src_domain, Attraction)就是说hotel的位置要在src_domain即attraction的实体附近。在论文NLU的结果分析最后一部分，也有提到select与其他intent的不同

[preprocess.py.zip](https://github.com/thu-coai/CrossWOZ/files/8478573/preprocess.py.zip) 这是对话标注的部分源码，看看逻辑就好，数据处理的完整步骤不太好找了

> hey we'll look into this discrepancy in the next few days. thanks for your note. Any idea?

Sorry for the late reply. intent_logits are the logits **before** sigmoid layer, since [BCEWithLogitsLoss](https://pytorch.org/docs/stable/generated/torch.nn.BCEWithLogitsLoss.html) combines a Sigmoid layer and the BCELoss in one single class. Therefore, the logit>0 means output...

@AtmaHou We release [ConvLab-3](https://github.com/ConvLab/ConvLab-3), using python>=3.8, have a try!

@ShaneTian [ConvLab-3](https://github.com/ConvLab/ConvLab-3) release now, new models are available!