This issue is to track the work we need to do before we can merge sample-props branch into main:

• Based on this prototype: [WIP] sample props (proposal 4) #16079
• Merge Base sample-prop implementation and docs (alternative to #21284) #22083 into sample-props. This PR only touches BaseEstimator and hence consumers. It does NOT touch meta-estimators, scorers or cv splitters.
• Work on splitters and merge that into sample-props: FEAT add metadata routing to splitters #22765
• feature flag: SLEP006: introduction of metadata routing through a feature flag #26045: FEAT add SLEP006 with a feature flag #26103
• Add new terms to the glossary: router, consumer, metadata: DOC add metadata routing terms to glossary #26685
• Work on scorers and merge that into sample-props (note that this involves an ongoing discussion on whether we'd like to mutate a scorer or not): FEAT SLEP6: scorers #22757
• Get backward compatibility mechanisms in place for meta-estimators which already route given metadata: FEAT multioutput routes metadata  #22986
• Work on meta-estimators: this involves writing get_metadata_routing in easy cases, and a whole lot more in cases where we'd like to keep backward compatibility in parsing input args suchs as estimator__param in Pipeline
• Once all the above is merged into sample-props, do a few tests with third party estimators to see if they'd work out of the box. Note that consumer estimators should work out of the box as long as they inherit from BaseEstimator and their fit accepts metadata as explicit arguments rather than **kwargs.
  • Merged w/o completion of meta-estimators, they got done progressively.
• Check whether a library such as cuML could vendor _metadata_requests.py and work with scikit-learn meta-estimators w/o depending on the library.
  • dropped the idea of supporting vendoring, they'd need to inherit from BaseEstimator
• Refactor tests: Refactor metadata routing classes used in tests #23918
• Merge sample-props into main: FEAT SLEP006: metadata routing infrastructure #24027

Enhancements:

• ENH make sure warn_on errors on invalid child #24023
• FEA SLEP006: debugging tool for sample props #18936
• RFC Revisiting meta-routing developer API for defining method mappings #28467

Cleanup

• SLEP006: make sure generated set_{method}_request methods are all legit #26505
  • MNT mark non-standard inverse_transform X args as non-metadata (SLEP6) #26503

Open issues:

• SLEP006 - metadata handling: fit, transform, fit_transform #22987
  • ENH make fit_transform and fit_predict composite methods (SLEP6) #26506
• SLEP006 - equal requests? {fit, partial_fit}, {predict, predict_proba, predict_log_proba, decision_function} #22988: ENH allow all main top level methods to have a corresponding set metadata request #23342
• RFC SLEP006: allow users to enable a "strict" mode in metadata routing #23920
• RFC SLEP006: verbose vs non-verbose declaration in meta-estimator #23928
  • global flags: SLEP006: globally setting request values #26050
  • auto-routing: SLEP006: default routing #26179
• SLEP006: mypy can't find set_{method}_request methods #23933
  • MNT make type checkers happy with set_{method}_request methods #26911
• RFC: Breaking Changes for Version 2 #25776: MNT SLEP006: Remove backward compatibility from meta-estimators #25852 - resolution is to use the feature flag instead of breaking change.
• Routing metadata to the response_method used by a scorer #27977

Our plan is to hopefully have this feature in 1.1, which we should be releasing in late April/early May.

cc @jnothman @thomasjpfan @lorentzenchr

Misc:

• MNT SLEP006: change release version to 2.0 #25851
• MNT SLEP6: raise NotImplementedError for meta-estimators not supporting metadata routing #27389

From #24027

Once we're finished with merging PRs to sample-props branch, we should merge this.

I've opened this for us to have an idea of what the whole project is touching.

We probably should rebase and merge this instead of squash and merge.

This closes and fixes the following issues and PRs:

• Routing issues:
  Enables Add TimeSeriesCV and HomogeneousTimeSeriesCV #6322,

• AdaBoost:
  Fixes Make Pipeline compatible with AdaBoost #2630,
  Fixes fit_params for Adaboost estimator #21706

• ClassifierChain:
  Fixes ClassifierChain does not support GroupKFold #11429,

• FeatureUnion:
  Fixes fit_params in conjunction with FeatureUnion #7136

• LogisticRegressionCV:
  Fixes LogisticRegressionCV not compatible with LeaveOneGroupOut #8950,

• Multioutput*
  Fixes MultiOutputRegressor: Support for more fit parameters #15953

• OneVsRestClassifier
  Fixes OneVsRestClassifier's fit method does not accept kwargs #10882

• RFE/RFECV:
  Fixes RFE/RFECV doesn't work with sample weights #7308

• *SearchCV:
  Fixes Nested CV of LeaveOneGroupOut fails in permutation_test_score #8127,
  Fixes grid_search: feeding parameters to scorer functions #8158,

• cross_val*:
  Fixes Weighted scoring in cross validation (Closes #4632) #13432,
  Fixes Should cross-validation scoring take sample-weights into account? #4632,
  Fixes GroupKFold fails in nested cross-validation (similar to #2879) #7646,
  Fixes model_selection.cross_validate doesn't pass groups argument to estimator #20349

• Stacking*
  Fixes Support fit_params in stacking #18028,

• Voting*
  Fixes Enable mixed ensembles with estimators that do & don't accept the sample_weight fit_param #20167