Skip to content

GitLab

Closed
Created by Elizaveta Lutsenko@LizLutsenkoOwner

Allow to fill np.inf/-np.inf values if it occurs in InputData

Created by: J3FALL

It was discovered that if InputData contains inf/-inf values models can not be fitted correctly. image

The same issue with nan-values was fixed via _fill_remaining_gaps in operation layer: if data comes to operation/model with nan values, it will be filled by mean or most frequent value.

In order to deal with inf values, the feature should be improved via the following:

  1. Add new tests that check models/operations that get InputData with inf values
  2. Check ImputationImplementation and improve it: add np.inf -> np.nan convertation in features/target
  3. Replace SimpleImputer from sklearn in _fill_remaining_gaps to new version of ImputationImplementation
  4. Add logging for this operations
  5. Check that it works correctly

For any details and suggestions, you can discuss it with me, @Dreamlone or @nicl-nno.