error at keras/examples/pretrained_word_embeddings.py

[DaPenggg]

HI ,ALL
when I run keras/examples/pretrained_word_embeddings.py, it is ok. But,if i save this model like:
model.save('my_model.h5')
then I create a new py script and load model:
model = keras.models.load_model('my_model.h5')
a error raise: Invalid type for initial value: <class 'dict'>(expected python scalar,list or tuple of values or numpy )

the solution is that:
embedding_layer = Embedding(num_words,
EMBEDDING_DIM,
embeddings_initializer=Constant(embedding_matrix.tolist()),
input_length=MAX_SEQUENCE_LENGTH,
trainable=False)

modify Constant(embedding_matrix) to be Constant(embedding_matrix.tolist())
after that ,the model can be saved and reload.

[Paul0M]

good

[yanboliang]

@DaPenggg I guess this is caused by we don't handle numpy array properly in serialization, I sent #10727 to fix it, let me know it can solve your issue. Thanks.

[DaPenggg]

@yanboliang OK. after release new code , i will check it .