search

chickenbestlover / RNN-Time-series-Anomaly-Detection

RNN based Time-series Anomaly detector model implemented in Pytorch.
Apache License 2.0
1.17k stars 316 forks source link

Question about the code of the model #32

Closed dagap closed 4 years ago

dagap commented 4 years ago

I see the following lines in the model construction and I wonder if you would be kind enough to comment on them:

if self.res_connection:
            decoded = decoded + input
        if return_hiddens:
            return decoded,hidden,output

        return decoded, hidden

res_connection: Is this basically a skip/residual connection?

return_hiddens: Should this be better renamed to return_outputs? I am guessing this is optionally returning the weighted output of each hidden layer

Also, the decoder is simply a linear layer. From what I see from looking at RNN autoencoder like tutorials, the decoder is usually also a RNN. Could you comment on the architecture?