The performance of pytorch-reimplement

[Onion12138]

It seems that the performance isn’t as good as expected when I apply the model to other Chinese Datasets. I wonder whether it is due to the model itself or the missing details in the pytorch reimplement. Looking forward to your reply, many Thanks.

[longlongman]

Sorry to hear about that, as you can see, I use the code on a Chinese Dataset(CMED) and achieve better results compared with the Keras version. I did not test the model on other Chinese Datasets, but I have done my best to keep the torch version as same as the Kears version.

[jxrjlxc02]

大佬，请问 f1 = 47.59，在真实场景能使用么，会不会偏低了些？

[longlongman]

确实不能用在实际场景，肯定是需要根据实际场景做调整的

[1qingqing1]

请问一般如何根据实际场景做调整呢，可否传授一些经验