dgai91/crcnn-pytorch

crcnn-pytorch's Issues

can't achieve the same performance

Hi,
I read the same paper about CRCNN and find your implementation. I noticed that you achieved f1 score of 70% on test data at ~50 epochs. However, I'm running your code but can't achieve a similar performance:(

Could you please share some details like:
- if you pre-trained the embeddings yourself with word2vec model on SemEval dataset, or use other embeddings? (I'm using pre-trained glove.6B.50d)
- are the hyper-parameters in your code the final ones you use to achieve 70% f1 score?
- can you remember the version of PyTorch in use? (I'm using ver 0.3 released in Dec 2017)

Thank you.

There is no such a file called attention, so we cannot get the embeddings.txt and the words.lst

There is no such a file called attention, so we cannot get the embeddings.txt and the words.lst. But they are all in need in the train.py. So, where can I get it? Thanks a lot!