I have been reading up a bit on LSTM's and their use for time series and its been interesting but difficult at the same time. One thing I have had difficulties with understanding is the approach to adding additional features to what is already a list of time series features. Assuming you have your dataset up like this:


Now lets say you know you have a feature that does affect the output but its not necessarily a time series feature, lets say its the weather outside. Is this something you can just add and the LSTM will be able to distinguish what is the time series aspect and what isnt?


1 Answer 1


I don't think that people have discovered a really good way to combine non-sequence information with time series information in an LSTM. There are a few things that you can do though. One is to just feed in your non-sequence information at every time step even though it is constant over time. See this paper by Mikolov and Zweig where they feed in topic information at every timestep of a language model.

  • $\begingroup$ Link to the paper is dead. I am also quite interested in how other people are approaching this issue. $\endgroup$
    – photox
    Feb 21, 2017 at 22:41
  • $\begingroup$ I like the idea in the paper, I'm just not sure how to implement it in keras. I also don't see how they are connecting the augmented data with both the hidden layer and the output layer. Maybe keras is too high level to do that with, and you'd need TF or Theano directly. $\endgroup$
    – photox
    Feb 22, 2017 at 17:37

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.