At the moment conversion from pandas.Categorical columns fails with ArrowException: Invalid: only handle 1-dimensional arrays. As a better alternative, we should provide one of the following solutions:
- Convert the categorical column to a string (Pandas type object) column, then use the conversion routines for strings. Add some metadata to the Arrow column that it was initially a Pandas string column so that in the case of a roundtrip, it will be a categorical column again.
- Implement the conversion of the column to a dictionary-encoded Arrow column. This is the preferred solution but may be more complicated to implement as certain requirements have not yet been implemented.