iwyoo / lstm-autoencoder Goto Github PK

In this implementation you are generating z.codes by the encoder but you never use them again. However, as far as I remember the idea behind the autoencoders is to decode the hidden representation of the encoder part by the decoder part to reconstruct the input. What I see in this code is that for decoding you only take the final state of the encoder part and reconstruct the input by starting from a zero initialization. However, I am not sure whether there is something I am missing to understand this code.

Hi,I am interested in this project.But I got two questions:

1. How can I get the encode weight after training?
2. How can I get the encode result but not to decode? Because I want to use the encoding result to do something.
   I will really appreciate it if you can answer the questions. Thank you !!

What's the difference between decode_without_input=True and False?

I am trying to predict clear signal from a noisy signal. The data is time-series data. I am having difficulties in understanding how to give me input to the model. Can you explain your method for input in test.py? I am unclear about the dimensions and what parameters like elem_num, step_num, etc. signifies?

How can I save the hidden state?

the means of the batch_num, step_num, elem_num?