Add embedding layer to Seq2Seq model

[azinnai]

Hi, I'm trying to add an embedding layer at the begging of the encoder but I cannot figured out an effective way to do this.

Does someone have an idea?

Thanks

[fredtcaroli]

Define the embedding layer yourself and then call the model on it

inp = Input(...)
emb = Embedding(...)(inp)
output = Seq2Seq(...)(emb)
model = Model(inp, output)

[zhongluwang]

@fredtcaroli if input_Y is shape(batchsize timestep),output is shape(batchsize timestep * outputdim),then how to calculate the loss?