Really need example or docs to implement CRF layer in Tensorflow Keras.

[ngoquanghuy99]

I'm confused about lots of PRs or Issues and I haven't got the solution yet!
Please provide a clear doc or example to deal with it or someone help me to add CRF layer to this BiLSTM network, i'd appreciate it!

input = Input(shape=(max_length,))
word_embedding_size = 100
model = Embedding(input_dim=vocab_size, output_dim=word_embedding_size, input_length=max_length)(input)
model = Bidirectional(LSTM(units=word_embedding_size, 
                           return_sequences=True, 
                           dropout=0.5, 
                           recurrent_dropout=0.5, 
                           kernel_initializer=k.initializers.he_normal()))(model)
model = LSTM(units=word_embedding_size * 2, 
             return_sequences=True, 
             dropout=0.5, 
             recurrent_dropout=0.5, 
             kernel_initializer=k.initializers.he_normal())(model)
model = TimeDistributed(Dense(tag_size, activation="relu"))(model)

[andreacimino]

I have been struggling too, it would be nice to have a minimal example.
With the keras-contrib package I was able to make it work without any issues.

[ngoquanghuy99]

I have been struggling too, it would be nice to have a minimal example.
With the keras-contrib package I was able to make it work without any issues.

I'm not able to add CRF layer by using keras-contrib package (since it was deprecated), though i solved the problem!

[andreacimino]

I have been struggling too, it would be nice to have a minimal example.
With the keras-contrib package I was able to make it work without any issues.

I'm not able to add CRF layer by using keras-contrib package (since it was deprecated), though i solved the problem!

Hello,

I tried to follow the the tests provided in the keras addons package, and did not manage to add the layer.
Could you please tell us why you closed the issue?
Could you please post (as mentioned) a very basic way to add a CRF layer to a LSTM (as in the first comment) ? The documentation and the examples are lacking this information.

Thank you.

[ngoquanghuy99]

I have been struggling too, it would be nice to have a minimal example.
With the keras-contrib package I was able to make it work without any issues.

I'm not able to add CRF layer by using keras-contrib package (since it was deprecated), though i solved the problem!

Hello,

I tried to follow the the tests provided in the keras addons package, and did not manage to add the layer.
Could you please tell us why you closed the issue?
Could you please post (as mentioned) a very basic way to add a CRF layer to a LSTM (as in the first comment) ? The documentation and the examples are lacking this information.

Thank you.

Sorry, i was thinking you'd solve the problem by using 'keras-contrib' package.
As far as i know, it was deprecated last year, here. They moved it to tensorflow/addons then removed it later. Now everyone sends bunch of pull requests (absolutely confusing).
By the way, my solution is to use tf2crf
or this way #1769

[Harsh188]

@ngoquanghuy99 If you are interested in helping create a CRF tutorial for addons please consider checking out #337. I've been unsuccessful in implementing CRF for a minimal tutorial since I'm fairly inexperienced with it.

[luozhouyang]

hey guys, have a look to luozhouyang/keras-crf, which is a more elegant and convenient CRF built on tensorflow-addons.