Keras Embedding Layer and CNN: How do I get the trained vectors from the model

Question

I used a neural network which contains an embedding layer.

I converted my datset, which consists out of multiple files that contains sentences, with a Tokenizer to vectors and fed them to the network as training input. After the embedding layer I got a CNN. How can I know get the trained input vectors back for plotting purposes?

score 2 · Accepted Answer · answered Jul 09 '18 at 07:05

It is easy, try:

for layer in my_model.layers:
    weights = layer.get_weights()
    configs = layer.get_config()

You'll have to do some parsing in the configs to retrieve the embedding layer. You have the 'name' that you can use to identify and retrieve your layers

Keras Embedding Layer and CNN: How do I get the trained vectors from the model

1 Answers1