tf implemented CRF compared with API crf

[heshenghuan]

Hello, recently I am studying CRF and I tried to using tensorflow to implement a linear-chain CRF.

I have read about you blog about this repo. And I also checked tensorflow's API, it has a Module for constructing a linear-chain CRF.

By reading your source code and tensorflow's source code, I found that both are using tag\_score for CRF decoding, I was wondering whether the two methods have a performance gap?

[heshenghuan]

And here is my repo about NER task using LSTM/Bi-LSTM/CNN-BiLSTM + CRF, but I used the API.

What confused me is that when I apply a small training data to do NER tasks, the model performance really poor (F1 only 2~3.7%). But then I tried to use the SIGHAN05 Chinese Word Segmentation data, and the performance was not bad (pku F1 about 90%).