Expanding Window Splitter

Is your feature request related to a problem? Please describe.

The sliding window splitter already exists, but I would also like to request the expanding window splitter as described here.

Describe the solution you’d like The image in the link above describes how the splits can be achieved.

Describe alternatives you’ve considered Currently manually only using CutoffSplitter, but what if I want to use this in ForecastingGridSearchCV. Use of expanding window with ForecastingGridSearchCV is currently unavailable as far as I could tell.

[aiwalter]

I am just following this discussion and have some thoughts. I think the misunderstanding is that there is a hyperpaeameter tuning part and a model training part. The sliding window is part of the tuning and training is usually done after with all training data and the found best params.

I think the model can only learn from data that is has seen. When using a sliding window, old data is not shown to all models (at least in tuning mode, you might show all after tuning to really train it for inference). So if there is an important pattern in the old data that might appear in future again, a sliding window tuning can be a problem because you might select hyperparameters from a model that has not seen this old data.

This can be solved with using expanding window. The problem of an expanding window is however that most likely in empiricism, the models with the bigger training data will have the better performance (“the bigger the better”). So in expanding window there is always a bias to select hyperparams from a model that has seen more data. So both window splitter methods have pros and cons.

Anyway I think it would be good to have an ExpandingWindowSplitter, so people would just have more choice based on what they prefer?

[mloning]

I’ll close this one in favour of #552 to keep discussion in one place