This message was deleted.

Hey. Yes, you can provide the step_size argument to cross_validation

Thanks for replying! I’m still not seeing how that would lead to a rolling window. To be clear, rolling window CV would mean, say I have monthly data and I always want to estimate my model on 120 months, then my first train window is months 1-120, my second train window is months 2-121, and so on (assuming I wanted to roll forward one month each time).

That's a step size of 1

Would I need to specify keep_last_n=120?

No,

h=120

But h is the forecast horizon? Actually I think I understand now. To expand my example, let’s say I want to forecast the next three monthly values. I would do h=3, freq=M, input_size=120, and step_size=1, since the docs says input_size is the maximum number of training observations. And what I’m after is a CV where I’m only training my model on 120 observations at any given time.