search

allanj / ner_incomplete_annotation

129 stars 33 forks source link

I am interested in implementing the PyTorch version of this project. #1

Closed kajyuuen closed 5 years ago

kajyuuen commented 5 years ago

I implemented your "Hard method" using PyTorch. But, I couldn't get the performance written in your paper.

Therefore, I would like to ask you about the project of the PyTorch implementation. And, if you have time, I'd really appreciate it if you could give me your opinion on my implementation.

allanj commented 5 years ago

Thanks. I will try to finish it over the weekend. Please also point me to the link of your implementation, I will also take a look once I finished it.

kajyuuen commented 5 years ago

I really appreciate it. My implementation is as follows: https://github.com/kajyuuen/Fuzzy-LSTM-CRF

I'm looking forward to you finishing your implementation.

allanj commented 5 years ago

I'm fixing some of the bugs. will need some more days. Sorry for the delay

allanj commented 5 years ago

@kajyuuen I have implemented a hard approach but I didn't integrate the soft approach yet. You can check it out and let me know if you have any questions. Feel free to open the issue again.

By using PyTorch, I change the batch size to 10 and use SGD as the optimizer. Thus, the number of epochs is set to 50 for convergence. But I didn't go through the hyperparameters because my preliminary experimental results show it should be able to achieve the results in the paper without further tuning.

I will also look into your github repo.

allanj commented 5 years ago

Your code (hard_bilstm_crf) looks good to me. I might need to run your code to check though.