BUG: loc-indexing with a CategoricalIndex with non-string categories #29922
Conversation
topper-123
force-pushed
the
categorical_indexing_with_non_string_categories
branch
from
49eb424 to
5167024
Compare
|
Hello @topper-123! Thanks for updating this PR. We checked the lines you've touched for PEP 8 issues, and found: There are currently no PEP 8 issues detected in this Pull Request. Cheers! 🍻 Comment last updated at 2019-12-11 07:13:21 UTC |
topper-123
force-pushed
the
categorical_indexing_with_non_string_categories
branch
from
5167024 to
364f61e
Compare
pandas/core/indexes/category.py
Outdated
| try: | ||
| return self.categories._convert_scalar_indexer(key, kind=kind) | ||
| except TypeError: | ||
| self._invalid_indexer("label", key=key) |
There was a problem hiding this comment.
when is this TypeError hit?
There was a problem hiding this comment.
That was hit when you have pd.Series([1,2], index=pd.CategoricalIndex(["a", "b"])).loc[1], i.e. wrong indexing type. It's the same error type as in s = pd.Series([1,2], index=["a", "b"]).loc[1].
I've added a test for it.
jreback
added
Categorical
topper-123
force-pushed
the
categorical_indexing_with_non_string_categories
branch
2 times, most recently
from
4751ca7 to
f2274d4
Compare
pandas/core/indexes/category.py
Outdated
| if self.categories._defer_to_indexing: | ||
| return self.categories._convert_scalar_indexer(key, kind=kind) | ||
|
|
||
| if kind == "loc" or self.categories._defer_to_indexing: |
There was a problem hiding this comment.
we might not need the _defer_to_indexing any longer, as I think the only way to get here is kind=='loc'
|
Ill be abroad untill the weekend, will look at it then. I think it would cause a error with categories made from datetimeindex, but I’ll check. |
topper-123
force-pushed
the
categorical_indexing_with_non_string_categories
branch
3 times, most recently
from
e9f5832 to
746d6d4
Compare
|
I`ve updated the PR: I´ve removed _defer_to_indexing and added some tests. |
| result = df.loc[idx_values[0]] | ||
| expected = Series(["foo"], index=["A"], name=idx_values[0]) | ||
| tm.assert_series_equal(result, expected) | ||
| # list selection |
There was a problem hiding this comment.
can you put a blank line between cases
| np.array([1, 2, 3], dtype=dtype) | ||
| for dtype in [ | ||
| np.int8, | ||
| np.int16, |
There was a problem hiding this comment.
any way to use the fixtures (or definitions) in pandas/conftest.py for some of this
topper-123
force-pushed
the
categorical_indexing_with_non_string_categories
branch
from
746d6d4 to
7316414
Compare
|
I've updated and use the dtypes from pandas/conftest.py instead. |
| [1.5, 2.5, 3.5], | ||
| [-1.5, -2.5, -3.5], | ||
| # numpy int/uint | ||
| *[np.array([1, 2, 3], dtype=dtype) for dtype in conftest.ALL_INT_DTYPES], |
There was a problem hiding this comment.
great! nice comprehensive test
|
thanks @topper-123 very nice! |
topper-123
deleted the
categorical_indexing_with_non_string_categories
branch
black pandasgit diff upstream/master -u -- "*.py" | flake8 --diffBug when indexing with
.locand the index is a CategoricalIndex with integer or float categories.