ValueError: Layer weight shape (30522, 768) not compatible with provided weight shape ()

Question

I got word-embedding using BERT and need to feed it as an embedding layer in the Keras model, and the error I got is

ValueError: Layer weight shape (30522, 768) not compatible with provided weight shape ()

the model is

embedding = Embedding(30522, 768, mask_zero=True)(sentence)
model.layers[1].set_weights([embedding_matrix])

Answer 1

You are passing to set_weights a list of list:

embedding_matrix = [np.random.uniform(0,1, (30522, 768))]

sentence = Input((20,))
embedding = Embedding(30522, 768, mask_zero=True)(sentence)
model = Model(sentence, embedding)

model.layers[1].set_weights([embedding_matrix])

while you should simply pass a list of arrays:

embedding_matrix = np.random.uniform(0,1, (30522, 768))

sentence = Input((20,))
embedding = Embedding(30522, 768, mask_zero=True)(sentence)
model = Model(sentence, embedding)

model.layers[1].set_weights([embedding_matrix])