LSTM time series multiple step forecasting

[mani...@googlemail.com]

Hallo guys,

currently I working on following problem and need your help.
I try to build a model that has the capability to forecast a time series multiple steps into the future.
Already, I read all post regarding LSTM's here in the group and a lot of tutorials, etc. from other sources,
but is still unclear to me how I can solve this task.

http://karpathy.github.io/2015/05/21/rnn-effectiveness/
After I read this blog my intuition is that I have to create a model that look
something like the above picture.

But how is it possible to implement this in Keras?
Is it necessary to use TimeDistributed layer?

Another question is how is it possible for a LSTM layer to handle the following example situation:
  We have a well training LSTM model that look like my intuition. With this model we try to predict
  which car at crossroads has priority at traffic, without traffic lights. Thus traffic signs are the significant
  indicator for the handling of those situations, when they are available.

  The problematic scenario I imagine is that we try to forecast now the next 20 time steps with last 100
  but the traffic sign that is really crucial to forecast the future in the right way is not in the current
  sliding window of the input data. How to overcome this problem is this related to the stateful/stateless
  LSTM's?

best regards

Manuel Volk 

[sebastia...@gmail.com]

Hi Manuel, 

Have you succeeded to find a solution to your problem? 

I would be very interested! 

Best regards, 

Sebastian Haglund 

[Faezeh Hajiaghajani]

This is my question too. If anyone has succeeded with developing a working model for multi step ahead prediction with LSTMs. 

Thanks,

Faezeh

--
You received this message because you are subscribed to the Google Groups "Keras-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to keras-users+unsubscribe@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/keras-users/7694fe55-aca8-4bfd-94eb-a5b64cfa2897%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.