I would expect a common use of FunctionTransformer is to apply some function to a Pandas DataFrame, ideally using its own methods or accessors. As noted in #10648, it can be easy for users to miss that they need to set validate=False to pass through a DataFrame without converting it to a NumPy array. I think it would be more user-friendly to have validate='array-or-frame' by default, which would pass through DataFrames to the function, but otherwise convert its input to a 2d array. For strict backwards compatibility, the default should be changed through a deprecation cycle, warning whenever using the default validation means a DataFrame is currently converted to an array.

Do others agree?