I am a newbie to machine learning. I have an LSTM model that predicts the next output n+1

time 1, params 1, output 1

time 2, params 2, output 2

time 3, params 3, output 3

. .

time n, params n, , output n

time n+1 --> predicts output n+1

 Here the times are all in minutes, so I can predict the next output in the series which is going to be the next minute. My question is that what if I want to predict the next 5 minutes. One solution was to throw out all the data except in steps of 5 minutes so the next step is automatically would be 5 minutes. This is clearly a waste of all the data that I have gathered. Can you please recommend what I can do about the prediction on different time scales? 


What you could do is just try and bypass the rest of the network after the LSTM if it isn't the 5'th minute. Depending on your framework this can be easy or a painstakingly task compared to the alternative. The alternative is just running and throwing away the output that isn't the next 5'th minute. While the last may seem inefficient it's rather easy to implement and takes just a bit more execution time. If execution time isn't a problem it's the easiest to get started with, if it doesn't work for your task you can always change it.

  • $\begingroup$ Thanks for the answer. Can you please elaborate on your first solution? Also, the problem I see with the second solution is that after time n until time n+5, I have no parameter information to keep continuing the model. I turn of data gathering for the next five minutes. I want to predict. Does that make sense? $\endgroup$ – user101464 Feb 12 '20 at 20:28

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.