evalml icon indicating copy to clipboard operation
evalml copied to clipboard

[SPIKE] Using Winsorization to cap outliers

Open ParthivNaresh opened this issue 3 years ago • 1 comments

I'd like to be able to perform winsorization on my input data i.e. to be able to bound outliers in my numerical features to a certain value or within a certain range, as well as specify the value that these outliers should take.

A few use cases:

  • To be able to specify a value range outside of which points would be considered outliers
  • Provide the option to replace these outlier values with a static value
  • Provide the option to cap these outliers, for example within 3 standard deviations.

ParthivNaresh avatar Mar 22 '21 17:03 ParthivNaresh

Agreed! It would help to have a specific dataset for which this approach is beneficial.

This issue tracks adding a component for this; we should test it in automl separately.

dsherry avatar Mar 25 '21 15:03 dsherry

Issue has been cancelled.

exalate-issue-sync[bot] avatar Jan 18 '23 20:01 exalate-issue-sync[bot]

Issue has been cancelled.

exalate-issue-sync[bot] avatar Jan 18 '23 20:01 exalate-issue-sync[bot]