search

alkaline-ml / pmdarima

A statistical library designed to fill the void in Python's time series analysis capabilities, including the equivalent of R's auto.arima function.
https://www.alkaline-ml.com/pmdarima
MIT License
1.58k stars 232 forks source link

Stepwise selection of exogenous features #553

Open dilwong opened 1 year ago

dilwong commented 1 year ago

A potentially useful feature could be for pmdarima.arima.auto_arima to stepwise determine (using the selected information_criterion) which exogenous features should be included in the model (e.g. add/remove columns in X to optimize the AIC).

tgsmith61591 commented 1 year ago

This is a cool idea. I think practically the best way to handle this would be to create an exog transformer that performs feature selection (i.e., similar to sklearn VarianceThreshold) that could be included in a Pipeline object. Then, I bet we could bootstrap the sklearn RandomSearchCV or GridSearchCV classes to do this.

If you'd like to take a swing at this, please feel free. Otherwise it may be some time before we get around to this enhancement.