keras-crf 0.0.3

A more elegant and convenient CRF built on tensorflow-addons.

keras-crf

A more elegant and convenient CRF built on tensorflow-addons.

Python Compatibility is limited to tensorflow/addons, you can check the compatibility from it's home page.

Installation

pip install keras-crf

Usage

Here is an example to show you how to build a CRF model easily:

import tensorflow as tf

from keras_crf import CRF, CRFLoss, CRFAccuracy


sequence_input = tf.keras.layers.Input(shape=(None,), dtype=tf.int32, name='sequence_input')
sequence_mask = tf.keras.layers.Lambda(lambda x: tf.greater(x, 0))(sequence_input)
outputs = tf.keras.layers.Embedding(100, 128)(sequence_input)
outputs = tf.keras.layers.Dense(256)(outputs)
crf = CRF(7)
# mask is important to compute sequence length in CRF
outputs = crf(outputs, mask=sequence_mask)
model = tf.keras.Model(inputs=sequence_input, outputs=outputs)
model.compile(
    loss=CRFLoss(crf),
    metrics=[CRFAccuracy(crf)],
    optimizer=tf.keras.optimizers.Adam(5e-5)
    )
model.summary()

The model summary:

Model: "functional_1"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
sequence_input (InputLayer)  [(None, None)]            0         
_________________________________________________________________
embedding (Embedding)        (None, None, 128)         12800     
_________________________________________________________________
dense (Dense)                (None, None, 256)         33024     
_________________________________________________________________
crf (CRF)                    (None, None)              1862      
=================================================================
Total params: 47,686
Trainable params: 47,686
Non-trainable params: 0
_________________________________________________________________