Error Display
new_model = load_model(“model.h5”)
Report an error:
1、keras load_model valueError: Unknown Layer :CRF
2、keras load_model valueError: Unknown loss function:crf_loss
bug fix
1、load_modelModify the source code：custom_objects = None change into def load_model(filepath, custom_objects, compile=True)：
2、new_model = load_model(“model.h5”,custom_objects={‘CRF': CRF,‘crf_loss': crf_loss,‘crf_viterbi_accuracy': crf_viterbi_accuracy}
After the above modifications, it is ready to run.
Additional knowledge:Building bilstm crf with keras
utilization/keras-team/keras-contribrealization of the crf layer.
Install keras-contrib
pip install git+/keras-team/
Code Example：
# coding: utf-8
from import Sequential
from import Embedding
from import LSTM
from import Bidirectional
from import Dense
from import TimeDistributed
from import Dropout
from keras_contrib. import CRF
from keras_contrib.utils import save_load_utils
VOCAB_SIZE = 2500
EMBEDDING_OUT_DIM = 128
TIME_STAMPS = 100
HIDDEN_UNITS = 200
DROPOUT_RATE = 0.3
NUM_CLASS = 5
def build_embedding_bilstm2_crf_model():
"""
Bidirectional LSTM with embedding + crf
"""
model = Sequential()
(Embedding(VOCAB_SIZE, output_dim=EMBEDDING_OUT_DIM, input_length=TIME_STAMPS))
(Bidirectional(LSTM(HIDDEN_UNITS, return_sequences=True)))
(Dropout(DROPOUT_RATE))
(Bidirectional(LSTM(HIDDEN_UNITS, return_sequences=True)))
(Dropout(DROPOUT_RATE))
(TimeDistributed(Dense(NUM_CLASS)))
crf_layer = CRF(NUM_CLASS)
(crf_layer)
('rmsprop', loss=crf_layer.loss_function, metrics=[crf_layer.accuracy])
return model
def save_embedding_bilstm2_crf_model(model, filename):
save_load_utils.save_all_weights(model,filename)
def load_embedding_bilstm2_crf_model(filename):
model = build_embedding_bilstm2_crf_model()
save_load_utils.load_all_weights(model, filename)
return model
if __name__ == '__main__':
model = build_embedding_bilstm2_crf_model()
Attention:
If executing the build model reports an error, it is most likely a keras version issue. With keras-contrib==2.0.8 and keras==2.0.8, the above code will not report an error.
Above this keras to solve the problem of loading lstm+crf model error is all I have shared with you, I hope to give you a reference, and I hope you support me more.