skrub-data
Deprecating the `specific_transformers` parameter of the `TableVectorizer`
The TableVectorizer has the specific_transformers parameter, which allows to pass custom transformers to specific columns in the dataframe.
Now that the Data Ops have been released, and we have the ApplyOnCols transformer we should probably deprecate this parameter as it is more complex and less robust than the alternatives.
From the skrub meeting:
Yes, we should do this. We have enough tools (ApplyOnCols and Data Ops) to achieve the same result with a better interface. There should be an example that clearly explains how to do this with one of the alternatives when we deprecate this function.
If possible, deprecation should make it in 0.6.2 and release 0.7.0 should deprecate this.
We talked about this again during the sprint at PyData and the conclusion was that the actual functionality of specific_transformers is harder to replace than expected. We should instead be clear in the user guide that the "better way" of doing things is either by using data ops, or by concatenating ApplyToCols.