Skip to content

REG: fix regression in df.corrwith on tied data when method is spearman #49032

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
wants to merge 12 commits into from
Closed

REG: fix regression in df.corrwith on tied data when method is spearman #49032

Show file tree
Hide file tree
Changes from 9 commits
Commits
Show all changes
12 commits
Select commit Hold shift + click to select a range
d228d7b
Update frame.py
GYHHAHA Oct 10, 2022
e230225
Update test_cov_corr.py
GYHHAHA Oct 10, 2022
059effd
Update v1.5.1.rst
GYHHAHA Oct 10, 2022
6a898f0
add blank line
GYHHAHA Oct 10, 2022
8fd679d
use algos.rank_1d
GYHHAHA Oct 10, 2022
3528e40
add more tests
GYHHAHA Oct 11, 2022
12e9232
Update frame.py
GYHHAHA Oct 11, 2022
8578581
convert nullable to np.array for pearson
GYHHAHA Oct 11, 2022
93879df
Update test_cov_corr.py
GYHHAHA Oct 11, 2022
8ae9403
fix import and pre-commit
GYHHAHA Oct 12, 2022
2f3c952
put params into tests
GYHHAHA Oct 14, 2022
cef9e9a
remove pd namespace
GYHHAHA Oct 14, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions doc/source/whatsnew/v1.5.1.rst
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -86,6 +86,7 @@ Fixed regressions
- Fixed regression in :meth:`DataFrame.apply` when passing non-zero ``axis`` via keyword argument (:issue:`48656`)
- Fixed regression in :meth:`Series.groupby` and :meth:`DataFrame.groupby` when the grouper is a nullable data type (e.g. :class:`Int64`) or a PyArrow-backed string array, contains null values, and ``dropna=False`` (:issue:`48794`)
- Fixed regression in :class:`ExcelWriter` where the ``book`` attribute could no longer be set; however setting this attribute is now deprecated and this ability will be removed in a future version of pandas (:issue:`48780`)
- Fixed regression in :meth:`DataFrame.corrwith` when computing correlation on tied data with ``method="spearman"`` (:issue:`48826`)

.. ---------------------------------------------------------------------------

Expand Down
31 changes: 19 additions & 12 deletions pandas/core/frame.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10593,23 +10593,30 @@ def corrwith(
corrs = {}
if numeric_only:
cols = self.select_dtypes(include=np.number).columns
ndf = self[cols].values.transpose()
else:
cols = self.columns
ndf = self.values.transpose()
k = other.values
k_mask = ~other.isna()
if isinstance(k, BaseMaskedArray):
k = k._data
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Why are we special Casing here?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nullable arrays are not supported by rank_1d.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You can use the mask too in that case

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Do you mean we still cast the type, but write rank step as the following?

libalgos.rank_1d(k, mask=nonnull_mask)

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I meant more like k_mask = k._mask

if method == "pearson":
for i, r in enumerate(ndf):
nonnull_mask = ~np.isnan(r) & ~np.isnan(k)
corrs[cols[i]] = np.corrcoef(r[nonnull_mask], k[nonnull_mask])[
0, 1
]
for col in cols:
val = self[col].values
nonnull_mask = ~self[col].isna() & k_mask
if isinstance(val, BaseMaskedArray):
val = val._data
corrs[col] = np.corrcoef(
val[nonnull_mask], k[nonnull_mask]
)[0, 1]
else:
for i, r in enumerate(ndf):
nonnull_mask = ~np.isnan(r) & ~np.isnan(k)
corrs[cols[i]] = np.corrcoef(
r[nonnull_mask].argsort().argsort(),
k[nonnull_mask].argsort().argsort(),
for col in cols:
val = self[col].values
nonnull_mask = ~self[col].isna() & k_mask
if isinstance(val, BaseMaskedArray):
val = val._data
corrs[col] = np.corrcoef(
libalgos.rank_1d(val[nonnull_mask]),
libalgos.rank_1d(k[nonnull_mask]),
)[0, 1]
return Series(corrs)
else:
Expand Down
88 changes: 87 additions & 1 deletion pandas/tests/frame/methods/test_cov_corr.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -355,7 +355,10 @@ def test_corrwith_mixed_dtypes(self, numeric_only):
expected = Series(data=corrs, index=["a", "b"])
tm.assert_series_equal(result, expected)
else:
with pytest.raises(TypeError, match="not supported for the input types"):
with pytest.raises(
TypeError,
match=r"unsupported operand type\(s\) for /: 'str' and 'int'",
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Now this is raised by np.corrcoef.

):
df.corrwith(s, numeric_only=numeric_only)

def test_corrwith_index_intersection(self):
Expand Down Expand Up @@ -406,3 +409,86 @@ def test_corrwith_kendall(self):
result = df.corrwith(df**2, method="kendall")
expected = Series(np.ones(len(result)))
tm.assert_series_equal(result, expected)

@pytest.mark.parametrize(
"numeric_only, ser, expected",
[
(
True,
Series([0, 1, 1, 0]),
Series([0.0] * 3 + [1.0] * 4, index=list("ABCDEFH")),
),
(
True,
Series([0.0, 1.0, 1.0, 0.0]),
Series([0.0] * 3 + [1.0] * 4, index=list("ABCDEFH")),
),
(
True,
Series([False, True, True, False]),
Series([0.0] * 3 + [1.0] * 4, index=list("ABCDEFH")),
),
(
False,
Series([0, 1, 1, 0]),
Series([0.0] * 3 + [1.0] * 5, index=list("ABCDEFGH")),
),
(
False,
Series([0.0, 1.0, 1.0, 0.0]),
Series([0.0] * 3 + [1.0] * 5, index=list("ABCDEFGH")),
),
(
False,
Series([False, True, True, False]),
Series([0.0] * 3 + [1.0] * 5, index=list("ABCDEFGH")),
),
(
True,
Series([0, pd.NA, 1, 0], dtype="Int64"),
Series([0.0] * 3 + [1.0] * 4, index=list("ABCDEFH")),
),
(
True,
Series([0.0, pd.NA, 1.0, 0.0], dtype="Float64"),
Series([0.0] * 3 + [1.0] * 4, index=list("ABCDEFH")),
),
(
True,
Series([False, pd.NA, True, False], dtype="boolean"),
Series([0.0] * 3 + [1.0] * 4, index=list("ABCDEFH")),
),
(
False,
Series([0, pd.NA, 1, 0], dtype="Int64"),
Series([0.0] * 3 + [1.0] * 5, index=list("ABCDEFGH")),
),
(
False,
Series([0, pd.NA, 1, 0], dtype="Int64"),
Series([0.0] * 3 + [1.0] * 5, index=list("ABCDEFGH")),
),
(
False,
Series([False, pd.NA, True, False], dtype="boolean"),
Series([0.0] * 3 + [1.0] * 5, index=list("ABCDEFGH")),
),
],
)
def test_corrwith_spearman_with_tied_data(self, ser, numeric_only, expected):
# GH#21925
df = DataFrame(
{
"A": [2, 5, 8, 9],
"B": [2, np.nan, 8, 9],
"C": [2, np.nan, 8, 9],
"D": [0, 1, 1, 0],
"E": [0, np.nan, 1, 0],
"F": [0, np.nan, 1, 0],
"G": [False, True, True, False],
"H": [False, pd.NA, True, False],
},
).astype({"C": "Int64", "F": "Float64", "H": "boolean"})
s = Series([0, 1, 1, 0])
result = df.corrwith(s, method="spearman", numeric_only=numeric_only)
tm.assert_series_equal(result, expected)