specific_transformers accepting 2 formats in TableVectorizer

[jeromedockes]

in TableVectorizer, for specific_transformers we can pass a list of (name, transformer, columns) tuples OR a list of (transformer, columns) tuples (but not a list mixing both). this adds complexity in the docstring and parameter validation code, and makes it different from scikit-learn's ColumnTransformer which only accepts the first format.

WDYT about allowing only the first format (ie the same as ColumnTransformer)?

[GaelVaroquaux]

WDYT about allowing only the first format (ie the same as ColumnTransformer)?

No. The ColumnTransformer's API is really clunky to use

[jeromedockes]

No. The ColumnTransformer's API is really clunky to use

only the first format then?

[GaelVaroquaux]

using only the first format can make model selection hard because you don't control the name of the corresponding transformer.