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.
pip install keras-crf
Here is an example to show you how to build a CRF model easily:
import tensorflow as tf
from keras_crf import CRFModel
# build backbone model, you can use large models like BERT
sequence_input = tf.keras.layers.Input(shape=(None,), dtype=tf.int32, name='sequence_input')
outputs = tf.keras.layers.Embedding(21128, 128)(sequence_input)
outputs = tf.keras.layers.Dense(256)(outputs)
base = tf.keras.Model(inputs=sequence_input, outputs=outputs)
# build CRFModel, 5 is num of tags
model = CRFModel(base, 5)
# no need to specify a loss for CRFModel, model will compute crf loss by itself
model.compile(
optimizer=tf.keras.optimizers.Adam(3e-4)
metrics=['acc'],
)
model.summary()
# you can now train this model
model.fit(dataset, epochs=10, callbacks=None)
The model summary:
Model: "crf_model"
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
sequence_input (InputLayer) [(None, None)] 0
__________________________________________________________________________________________________
embedding (Embedding) (None, None, 128) 2704384 sequence_input[0][0]
__________________________________________________________________________________________________
dense (Dense) (None, None, 256) 33024 embedding[0][0]
__________________________________________________________________________________________________
crf (CRF) [(None, None), (None 1320 dense[0][0]
__________________________________________________________________________________________________
decode_sequence (Lambda) (None, None) 0 crf[0][0]
__________________________________________________________________________________________________
potentials (Lambda) (None, None, 5) 0 crf[0][1]
__________________________________________________________________________________________________
sequence_length (Lambda) (None,) 0 crf[0][2]
__________________________________________________________________________________________________
kernel (Lambda) (5, 5) 0 crf[0][3]
==================================================================================================
Total params: 2,738,728
Trainable params: 2,738,728
Non-trainable params: 0
__________________________________________________________________________________________________