Pandas 8-bit nullable integer type not supported

[andreostrovsky]

#7760 added support for Pandas nullable integer types. Yet it only added support for Int16, Int32, and Int64; not Int8. Is there a reason this was excluded?

Pandas has a commonly used function to reduce DataFrame size to_numeric(series, downcast='integer'), which reduces the specified series to the smallest nullable integer type that fits the data, including Int8.

Currently, users need to then explicitly convert Int8 columns to Int16 so as to avoid xgboost error DataFrame.dtypes for data must be int, float, bool or category. ...

Using:

• xgboost 1.7.1
• pandas 1.5.1

[trivialfis]

At the time I just tried to implement support for whatever was listed by the pandas' doc about nullable type.