When multiple empty column names are given, there is a high risk that they end up referencing other the wrong dataframes. This is because the type alias created for the dataframes depends on the name of the column. If the name is an empty string, there is a high chance of overlapping
i tried your tool with a dict of lists with diff. types (bytes, int, float, str), each list contains about 50000 elements.
it took nearly 6 seconds to execute the follwowing command:
When type hinting new pandas accessors, these will not be properly type hinted. Consider the possibility of adding hardcoded lines and imports. Lines to be added to the class. Something like:
Where line_given_by_the_user can be: pint: PintAccessor and {extra_imports} being PintAccessor corresponding import
How this information is given is still to be decided
EDIT: Another alternative would be to scan all accessors within a pandas dataframe and include them in the class scope of the dataframe (including associated import)