Parameter optimization for forecast validation

Barclaeys
Barclaeys New Altair Community Member
edited November 2024 in Community Q&A
Hello, is it possible to perform a parameter optimization for a forecast validation model? Does the window size, step size or horizon make a difference for the accuracy of the e.g. Holt-Winters model?
Tagged:

Welcome!

It looks like you're new here. Sign in or register to get started.

Best Answer

  • lionelderkrikor
    lionelderkrikor New Altair Community Member
    Answer ✓
    Hi @Barclaeys,

    You can find Martin's TS extension by going on the MarketPlace (Extension -> MarketPlace) and search for "forecasting" : 

      

    Regards,

    Lionel

     

Answers

  • MartinLiebig
    MartinLiebig
    Altair Employee
    yes of course. Again i would recommend you to have a look at my Auto-TS extension: https://community.rapidminer.com/discussion/57109/new-extension-how-to-do-time-series-forecasting-with-just-1-operator . Internally it is doing exactly this for you.

    Note that you do not optimize on window_size, step_size nor horizon. Those are settings of your validation. you rather optimize alpha beta and gamma of HW.
    Think about it like this: A validation is a technique to measure the accuracy of your model. You determine the settings of this "measurement" and then try to find the right parameters of the model. Something like a calibration.

    Best,
    Martin

  • Barclaeys
    Barclaeys New Altair Community Member
    hi Martin, question where can I find your TS extension with new operator? Download a RM upgrade? A specific component that you built? Thanks
  • lionelderkrikor
    lionelderkrikor New Altair Community Member
    Answer ✓
    Hi @Barclaeys,

    You can find Martin's TS extension by going on the MarketPlace (Extension -> MarketPlace) and search for "forecasting" : 

      

    Regards,

    Lionel

     
  • Barclaeys
    Barclaeys New Altair Community Member
    Thank you Lionel - package installed! Will check it out now.

Welcome!

It looks like you're new here. Sign in or register to get started.

Welcome!

It looks like you're new here. Sign in or register to get started.