I have worked on VAR ( vector Auto regression), which forecasts multiple output values(continuous) when there is linear relationship between all variables. I want to use Neural nets to do so,sothat I can include non linear relation among variables also.

Is there any specific type of NN available?

If not how can I implement it using NN?


1 Answer 1


The type of artificial neural nets (ANN) you are looking for is ANN with cycles which is widely known as recurrent neural networks(RNN). As you have mentioned Vector Auto regression (VAR) I would recommend you to check out this paper which compares VAR to RNN on two different tasks pertaining to two different real world datasets.

One specific type of RNN is Long short term memory (LSTM). It is particularly good at learning over long sequences. Although RNN is also perfectly capable of that theoretically, but with little empirical success. So, in recent years LSTM has gained huge popularity and has been very successful over sequence learning tasks. I would recommend this book which provides a fair bit of understanding of ANN with cycles and the tasks they are being applied to.

For implementation of RNN there are many APIs available. Keras provides a high level API for the same.

For modelling multivariate time series with LSTM check out this tutorial.

You can also have a look at this question that I answered on cross validated.

  • $\begingroup$ I know different tools and API( not the deep understanding). My concern is -having multiple output in output layer. How would common loss function look like.( As now NN has to minimize multiple loss functions.. Will this be possible also? $\endgroup$ Commented Jun 18, 2018 at 6:15
  • $\begingroup$ @ArpitSisodia Edit you question accordingly. It doesn't exactly reflect what you are asking. $\endgroup$
    – naive
    Commented Jun 18, 2018 at 6:46
  • $\begingroup$ sure @Naive. anyway R2N2 is useful, I need to read this paper. $\endgroup$ Commented Jun 18, 2018 at 7:45

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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