I want to make a LSTM RNN for timeseries prediction, but some of my predictors are monthly and others are daily. Any advice / examples on how to set up this network?

The frequency of the predictions is monthly.


  • 1
    $\begingroup$ What is the frequency of the predictions? $\endgroup$ – Aaron May 4 '16 at 21:08
  • $\begingroup$ Thanks for Q. Monthly. I have adjusted the original question. $\endgroup$ – BHP May 4 '16 at 21:39
  • $\begingroup$ Another descriptor is 'misaligned' time series. $\endgroup$ – user0 Aug 10 '17 at 18:04

You can maybe get some inspiration from the ideas presented in this article which proposes a representation of the time series such that it deals with asynchronous sampling: you encode what is the source (the id of the time series) and the duration (time to the last value considering all time series) of the current value, and you end up with a single time series of values (as pictured in Figure 1 of the article that I attached below).

enter image description here


You could use a hierarchical structure. One LSTM can create an embedding vector for the sequence of daily predictors for each month. Then this embedding is fed into a second LSTM along with the monthly predictor variables.

  • $\begingroup$ Is this possible with current libraries? @Aaron $\endgroup$ – user0 Aug 10 '17 at 18:06
  • 2
    $\begingroup$ Actually, check out the PhasedLSTM in the latest version of tensorflow $\endgroup$ – Aaron Aug 12 '17 at 7:25

Check the power demand experiments in this paper: https://www.elen.ucl.ac.be/Proceedings/esann/esannpdf/es2015-56.pdf. Different layers in a deep network can capture different time scales.

  • 3
    $\begingroup$ Could you make your answer a bit more explicit so that one would not need to read the linked paper to get the idea? $\endgroup$ – Richard Hardy Jul 10 '16 at 17:04

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.