Hi,
Just curious is it ok to not set an input length in an embedding layer? If a sentence from training data is very long, the dimension of the embedding layer’s weights would be large, could it cause slow computation or memory issue?
Also only embedding_dim=64 is provided, where does the 523840 Param number come from? Thanks!
Hi @meiyou,
About memory consumption and computation adding very long sentences will impact in those parameters, but not impacting computation that much as you can expect.
Param # is vocabulary size * embedding dim. Have you changed vocabulary size as well?
Best,
thanks for your answer @german.mesa ! It clears my confusion. I used a different vocab size
Hi there,
I still have a question:
Why the maximum length is not provided in the output shape of the embedding layer?
I assume that using padded_batch method, the length should equal BATCH_SIZE, which is set to 64. Given that the embedding dimension is also 64, I expect the output shape to be something like this: (None, 64, 64), not (None, None, 64).
When input_length is not given for Embedding layer, the Keras gives None as ‘guessing’ .
Hopefully, it helps 
