Performance Issue with 'hasattr' Method

[owerbat]

• Linux
• Python 3.7
• Modin 0.8.0

Currently, hasattr(df, 'sparse') takes much longer if df is instance of modin.pandas.DataFrame than it is pandas.DataFrame.

[devin-petersohn]

Thanks @owerbat, this is probably something we can speed up a bit.

[gshimansky]

This happens because hasattr internally calls __getattribute__ and __getattr__ which in turn call the function itself. And this function DataFrame.sparse is currently implemented as a fallback to Pandas which is really inefficient. There is an issue to implement sparse accessor properly #1584, it should allow access to sub-properties.

[YarShev]

I think we can fix the issue for the upcoming release adding corresponding class containing sparse methods which will be defaulting to pandas.