We recently added the ability to run estimator_checks on instances.
We should add tests for the meta-estimators running estimator_checks on common settings, including

• pipeline
• feature union
• GridSearchCV(Estimator())
• GridSearchCV(Pipeline(...))
• RandomizedSearchCV(...)

And then possibly additional tests for anything that is currently not tested by the common tests. I think they raise SkipTest warnings, but I'm not sure? There's at least SparseEncode, RFECV, SelectFromModel, and probably more, that are not tested. Maybe CalibratedClassifierCV, VotingClassifier?

If someone wants to work on this, they can start with the five above and check out which other estimators are not tested and post a list here...