Train/Apply Model in "Sliding" Batches

Laszlo
Laszlo New Altair Community Member
edited November 5 in Community Q&A
Hi everyone.
I am trying to do the following:  Let's say I have 10000 examples in my data set.  I would like to take the
first 1000 to train my learner, then apply that model on the next 1000 examples for prediction.  Then "slide"
the training examples forward to 1001 to 2000, train, then apply to examples 2001 to 3000.  Repeat through the
whole data set.  In the end, I would have 9000 predictions.  The learner is being re-trained every 1000 data points.
I've been trying the "Loop Batches" operator, but that doesn't quite do what I want.  Sliding the window forward
manually is a bit tedious ( esp. if I expand this to several million examples!)

Any thoughts would be greatly appreciated.  ;)
Tagged:

Answers

  • wessel
    wessel New Altair Community Member
    Use the SlidingWindowValidation operator?
  • Laszlo
    Laszlo New Altair Community Member
    I don't see this operator.  Is it in the Series folder?  Maybe has a different name?
  • wessel
    wessel New Altair Community Member
    Series > Evaluation > Validation


    By the way, you can hit ctrl + I, to find any operator using text search.
  • Laszlo
    Laszlo New Altair Community Member
    Ah, thank you.  I needed to update to the latest Series Extension.

    I will experiment with it.