Performance regressions in GroupBy.fillna

[TomAugspurger]

https://pandas.pydata.org/speed/pandas/#groupby.FillNA.time_df_ffill?python=3.8&Cython=0.29.21&commits=03e58585

From 03e58585

cc @GYHHAHA

[GYHHAHA]

take

[GYHHAHA]

Is there a faster way to get the df with dropping the NaN group?
I try to use the indice operation to replace calling filter, but still slow.
Any suggestion? Thanks! @TomAugspurger

[TomAugspurger]

I’m not sure. If the bug we fixed was for correctness and the performance loss is unavoidable, then we can close this issue.

[jreback]

could be fixed by #43578

cc @mzeitlin11 @jbrockmendel

[jbrockmendel]

closed by #43578