Object dtype for empty describe #27184
Conversation
| @@ -595,7 +606,8 @@ def test_describe_empty_categorical_column(self): | |||
| df = pd.DataFrame({"empty_col": Categorical([])}) | |||
| result = df.describe() | |||
| expected = DataFrame({'empty_col': [0, 0, None, None]}, | |||
| index=['count', 'unique', 'top', 'freq']) | |||
| index=['count', 'unique', 'top', 'freq'], | |||
| dtype='object') | |||
There was a problem hiding this comment.
This was added in 0.25.0. Returning object dtype isn't an (additional) API change from 0.24
There was a problem hiding this comment.
should these be np.nan? (and not None); I am not sure we actually check this in the comparator.
There was a problem hiding this comment.
It wasn't clear to me whether I should use None or NaN. Arguments for NaN instead of None
- it's our most consistent missing value marker
- We often translate None to NaN implicitly
but I don't know what's best.
There was a problem hiding this comment.
right I think we should np.nan; however I don't remember if we actually test this in assert_series_equal; maybe the check_exact arg does this but I don't remember.
There was a problem hiding this comment.
nice catch
In [4]: a = pd.Series([None], dtype=object)
In [5]: b = pd.Series([np.nan], dtype=object)
In [6]: pd.util.testing.assert_series_equal(a, b)
| # https://github.com/pandas-dev/pandas/issues/27183 | ||
| s = pd.Series([None, None], dtype=object) | ||
| result = s.describe() | ||
| expected = pd.Series([0, 0, np.nan, np.nan], dtype=object, |
There was a problem hiding this comment.
actually you use np.nan here
|
thanks @TomAugspurger |
Closes #27183