Can't save the CRF model about emnlp2017-bilstm-cnn-crf HOT 3 CLOSED

Hi davidvscl,
I think the code will not work with Keras 2.1.1. The API changed significantly so that the CRF layer must be changed significantly. If you have done this, it would be great if you could share the code.

Sadly I cannot provide any insights how to port the code to 2.1.1 as I didn't yet have the time to run it with Keras 2

from emnlp2017-bilstm-cnn-crf.

Not sure why this happens.

You could change line 381 to:
'init': 'glorot_uniform'

Hopefully this fixes the issue.

Which keras version are you using?

from emnlp2017-bilstm-cnn-crf.

Hi! Same problem here! Changing line 381 as you suggested throws this error:

  File /BiLSTM.py, line 383, in evaluate
    self.trainModel()
  File /BiLSTM.py, line 123, in trainModel
    self.model.train_on_batch(nnInput, labels)   
  File /lib/python3.5/site-packages/keras/engine/training.py, line 1808, in train_on_batch
    check_batch_axis=True)
  File /lib/python3.5/site-packages/keras/engine/training.py, line 1411, in _standardize_user_data
    exception_prefix='target')
  File /lib/python3.5/site-packages/keras/engine/training.py, line 153, in _standardize_input_data
    str(array.shape))
ValueError: Error when checking target: expected chain_crf_1 to have shape (None, None, 4) but got array with shape (2, 139, 1)

PD: I'm using Keras 2.1.1 (I'm trying to move code from Keras 1.2.2 to last one)

from emnlp2017-bilstm-cnn-crf.