BUG: fillna with DataFrame input should preserve dtype when possible #61742
Conversation
|
Since we now operate column-wise and use This also preserves extension dtypes like Let me know if this behavior change is fine, happy to update the test or tweak the logic based on what’s preferred! |
simonjayhawkins
added
Bug
Dtype Conversions
| # restore original dtype if fallback to object occurred | ||
| if lhs.dtype == rhs.dtype and filled.dtype == object: | ||
| try: | ||
| filled = filled.astype(lhs.dtype) |
There was a problem hiding this comment.
id expect this to be handled by Series.where. is it not?
| @@ -7145,7 +7145,24 @@ def fillna( | |||
| else: | |||
| new_data = self._mgr.fillna(value=value, limit=limit, inplace=inplace) | |||
| elif isinstance(value, ABCDataFrame) and self.ndim == 2: | |||
| new_data = self.where(self.notna(), value)._mgr | |||
| filled_columns = {} | |||
| for col in self.columns: | |||
There was a problem hiding this comment.
doing this column-by-column is going to mean a performance hit for non-object cases. i suspect we need to do this at the Block level in order to avoid that
|
This pull request is stale because it has been open for thirty days with no activity. Please update and respond to this comment if you're still interested in working on this. |
When filling a DataFrame with another DataFrame using
fillna, columns with matching dtypes were being unnecessarily cast toobjectdue to use ofnp.where.This PR updates the logic to use pandas’
Series.where, which is dtype-safe and respects extension and datetime types.