A de-correlation model for feature exclusion

[FredrikKarlssonSpeech]

Is your feature request related to a problem? Please describe. Not all model frameworks handle highly correlated predictors well.

Describe the solution you'd like A de-correlation model similar to the standardizer would be great to have for these cases, that would remove superfluous predictors before model training.

Describe alternatives you've considered There seems to be none available in MLJ, except manually examining the correlations of predictors and removing identified not needed predictors using FreatureSelector.

[ablaom]

Correct. MLJ does not directly provide such a transformer. Ideally, this would be more than a Static transformer: You want to learn the redundant features by just looking at training data, and then apply the learned filter to arbitrary new data.

As this is feature selection, worth mentioning #70 and #426.