Handle stacked ensembling for partial dependence fast mode
- Dominant language
- Python
- Stars
- 850
- Forks
- 96
- PR merge metrics
- No merged PRs in 30d
Description
Once #3779 is implemented, fast mode will not work in pipelines that use stacked ensembling.
This was not done due to the added complexity of handling it, but we should be able to handle stacked ensembling for partial dependence fast mode by recreating each individual inputted pipeline, getting their original X_t, transform the single column, and generate predictions, thus recreating the dataframe that gets passed into the stacked ensembler estimator.
This should be implemented by adding a `handle_partial_dependence_fast_mode` method in `StackedEnsembleBase` to handle the ensembler and change `_can_be_used_for_fast_partial_dependence` to True.
Note: I expect this to mean we have to change the parameters and return values of `_handle_partial_dependence_fast_mode`, because handling this component means we need to `_get_cloned_feature_pipelines` and `_transform_single_feature` for each sub-pipeline. We should make sure the implementation is still functional for future components that may need to use this method.
Contributor guide
Assessment
This issue has not been assessed yet.