Flexible .compute_base_score() private method #54
Description
Could kill a few birds with one stone on all methods
- Performs initial resampling of the untrained learner on the task
- Option to use either full model or featureless learner for base model (relevant for "leave-in" methods in WVIM)
- Could detect if learner is trained, and if yes, predict and score on a given test dataset or the task (depending on API)
The last point would be a nice step towards #4
Also, it would reduce duplication among methods because some version of this is always the first step of the intiialization.