[ENH] heterogeneous object/estimator should allow (name, estimator, other, ...) tuples
Closed this issue · 1 comments
Currently, it is assumed that the collection of objects in heterogeneous estimators/objects is a parameter in the form of (name, estimator).
However, there are use cases where additional metadata is provided per estimator, e.g., sktime column ensembles, where additionally it is said whichcolumn the named estimator is applied to.
Therefore, we should think about whether, and if yes how, to extend the convention to (name, estimator, metadata, etc).
sktime currently has a relaxed version (added after factor out to skbase) with minimal changes where the logic takes 0th and 1st element from the tuple, but does not assume a length. E.g., name, est = tpl[0], tpl[1] instead of name,est = tpl and so on.
Also see discussion in #122