BUG GH22858 When creating empty dataframe, only cast int to float if index given #22963
Conversation
|
Hello @JustinZhengBC! Thanks for submitting the PR.
|
|
Looks pretty good. Can you add a whatsnew note for v0.24? |
WillAyd
added
Dtype Conversions
|
I updated the whatsnew for v0.24. Also, after reverting to master, I found that while passing "int64" as the dtype caused the bug, merely passing int did not. I have updated the test to reflect this. |
| @@ -806,6 +806,9 @@ def test_constructor_corner(self): | |||
| df = DataFrame(index=lrange(10), columns=['a', 'b'], dtype=object) | |||
| assert df.values.dtype == np.object_ | |||
|
|
|||
| df = DataFrame(columns=['a', 'b'], dtype="int64") | |||
There was a problem hiding this comment.
Can you add the GH issue number as a comment to reference?
doc/source/whatsnew/v0.24.0.txt
Outdated
| @@ -194,6 +194,7 @@ Other Enhancements | |||
| - :meth:`Index.to_frame` now supports overriding column name(s) (:issue:`22580`). | |||
| - New attribute :attr:`__git_version__` will return git commit sha of current build (:issue:`21295`). | |||
| - Compatibility with Matplotlib 3.0 (:issue:`22790`). | |||
| - A newly constructed empty :class:`DataFrame` of integers will now only be cast to ``float64`` if an index is specified (:issue:`22858`) | |||
There was a problem hiding this comment.
Hmm I think the wording DataFrame of integers is somewhat misleading as there aren't actually any integers here; the type is just passed explicitly during construction and not subsequently coerced.
Codecov Report
@@ Coverage Diff @@
## master #22963 +/- ##
==========================================
+ Coverage 92.18% 92.18% +<.01%
==========================================
Files 169 169
Lines 50833 50823 -10
==========================================
- Hits 46862 46853 -9
+ Misses 3971 3970 -1
Continue to review full report at Codecov.
|
doc/source/whatsnew/v0.24.0.txt
Outdated
| @@ -194,6 +194,7 @@ Other Enhancements | |||
| - :meth:`Index.to_frame` now supports overriding column name(s) (:issue:`22580`). | |||
| - New attribute :attr:`__git_version__` will return git commit sha of current build (:issue:`21295`). | |||
| - Compatibility with Matplotlib 3.0 (:issue:`22790`). | |||
| - A newly constructed empty :class:`DataFrame` with integer as the ``dtype`` will now only be cast to ``float64`` if ``index`` is specified (:issue:`22858`) | |||
There was a problem hiding this comment.
move to api breaking changes
pandas/core/dtypes/cast.py
Outdated
| if is_integer_dtype(dtype) and isna(value): | ||
| # GH 22858: only cast to float if an index | ||
| # (passed here as length) is specified | ||
| if is_integer_dtype(dtype) and isna(value) and length: |
| dtype=int) | ||
| assert df.values.dtype == np.dtype('float64') | ||
| @pytest.mark.parametrize("data, index, columns, dtype, ans", [ | ||
| (None, lrange(10), ['a', 'b'], object, np.object_), |
There was a problem hiding this comment.
can u rename ans -> expected
doc/source/whatsnew/v0.24.0.txt
Outdated
|
|
||
| .. _whatsnew_0240.api_breaking: | ||
| - A newly constructed empty :class:`DataFrame` with integer as the ``dtype`` will now only be cast to ``float64`` if ``index`` is specified (:issue:`22858`) |
There was a problem hiding this comment.
put in the section itself, just below this (it’s a backward incompatible change)
|
@WillAyd merge when satisfied |
|
Very nice change - thanks @JustinZhengBC ! |
git diff upstream/master -u -- "*.py" | flake8 --diffPreviously, when creating a dataframe with no data of dtype int, the dtype would be changed to float. This is necessary when a predefined number of rows is included as the index parameter, so that they can be filled with nan. However, when no index is passed, this cast is unexpected. This PR changes it so dtype is only altered when necessary.