Predicting time series

vivivivi Member Posts: 3 Contributor I
edited November 2018 in Help
Hello,

I've successfully trained and tested my time series forecasting model using neural networks. But now, I would like to use this model to predict some data points ahead. However, to predict for example t+3 my model requires the t+1 predicted point. Could you please help me on this topic?

Thanks!

Answers

  • wesselwessel Member Posts: 537 Maven
    Hello,

    You should create training examples with a +3 horizon.
    You can do this using the windowing operator.

    Best regards,

    Wessel
  • vivivivi Member Posts: 3 Contributor I
    Hi,

    I think the windowing can't solve my problem. My time series dataset ended at the last week of December, and I wanted to forecast the next 13 weeks. I guess the windowing process allows to only predict the next unknown value. AmI right? For example, II would like to forecast the 13th week using the 12th, 11th ... weeks. Do I have to use some kind of loop to forecast ahead?

    Thnaks!
  • wesselwessel Member Posts: 537 Maven
    Hey,

    Windowing can create training examples with an horizon of 13 weeks.

    E.g.

    1 2 15
    2 3 16
    3 4 17

    So you are using data from time slot 1 and 2, to predict time slot 15, data from time slot 2 and 3 for 16, 3 and 4 for 17, etc.
    This way you do not need a loop.

    Best regards,

    Wessel

Sign In or Register to comment.