Skip to content

Add sample_weight support for QuantileTransformer when fit on dense data #31147

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

Open
wants to merge 5 commits into
base: main
Choose a base branch
from
+57 −4
Open

Add sample_weight support for QuantileTransformer when fit on dense data #31147

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
841ed26
add sample_weight parameter in fit function
Apr 3, 2025
eb78c16
add support for sample_weight in QuantileTransformer
Apr 4, 2025
7d51a7d
Merge branch 'main' into quantile-transformer-sample-weight
kaekkr Apr 4, 2025
0b0f0e6
remove custom weighted_percentile function, use _averaged_weighted_pe…
Apr 10, 2025
06960cf
Merge branch 'main' into quantile-transformer-sample-weight
kaekkr Apr 10, 2025
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
33 changes: 29 additions & 4 deletions sklearn/preprocessing/_data.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,6 +10,7 @@
from scipy.special import boxcox, inv_boxcox

from sklearn.utils import metadata_routing
from sklearn.utils.stats import _averaged_weighted_percentile

from ..base import (
BaseEstimator,
Expand Down Expand Up @@ -2726,7 +2727,7 @@ def __init__(
self.random_state = random_state
self.copy = copy

def _dense_fit(self, X, random_state):
def _dense_fit(self, X, random_state, sample_weight=None):
"""Compute percentiles for dense matrices.

Parameters
Expand All @@ -2748,8 +2749,30 @@ def _dense_fit(self, X, random_state):
X = resample(
X, replace=False, n_samples=self.subsample, random_state=random_state
)
if sample_weight is not None:
sample_weight = sample_weight[: X.shape[0]]

self.quantiles_ = np.zeros((len(references), n_features))

for i in range(n_features):
Copy link
Member

Choose a reason for hiding this comment

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

Please rename i to feature_idx to make the code easier to follow.

col = X[:, i]
mask = ~np.isnan(col)
col_clean = col[mask]

if col_clean.size == 0:
self.quantiles_[:, i] = np.nan
continue

if sample_weight is not None:
weights_clean = sample_weight[mask]
self.quantiles_[:, i] = _averaged_weighted_percentile(
col_clean, sample_weight=weights_clean, quantile=references / 100.0
)
Copy link
Member

Choose a reason for hiding this comment

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

Please check the source code of _averaged_weighted_percentile and the docstring of _weighted_percentile to learn how to use it properly.

In particular:

  • it accepts percentile ranks, not quantile ranks;
  • it can compute for for a single rank at a time, hence you might need to call this in a list comprehension for all possible ranks passed in the references array.

else:
self.quantiles_[:, i] = np.nanquantile(
col_clean, references / 100.0, method="averaged_inverted_cdf"
)

self.quantiles_ = np.nanpercentile(X, references, axis=0)
# Due to floating-point precision error in `np.nanpercentile`,
# make sure that quantiles are monotonically increasing.
# Upstream issue in numpy:
Expand Down Expand Up @@ -2802,7 +2825,7 @@ def _sparse_fit(self, X, random_state):
self.quantiles_ = np.maximum.accumulate(self.quantiles_)

@_fit_context(prefer_skip_nested_validation=True)
def fit(self, X, y=None):
def fit(self, X, y=None, sample_weight=None):
"""Compute the quantiles used for transforming.

Parameters
Expand Down Expand Up @@ -2844,9 +2867,11 @@ def fit(self, X, y=None):
# Create the quantiles of reference
self.references_ = np.linspace(0, 1, self.n_quantiles_, endpoint=True)
if sparse.issparse(X):
if sample_weight is not None:
raise ValueError("sample_weight is not supported for sparse input.")
Copy link
Member

Choose a reason for hiding this comment

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

Suggested change
raise ValueError("sample_weight is not supported for sparse input.")
raise NotImplementedError(
"sample_weight is not supported for sparse input."
)

self._sparse_fit(X, rng)
else:
self._dense_fit(X, rng)
self._dense_fit(X, rng, sample_weight=sample_weight)

return self

Expand Down
22 changes: 22 additions & 0 deletions sklearn/preprocessing/tests/test_data.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1528,6 +1528,28 @@ def test_quantile_transformer_sorted_quantiles(array_type):
assert all(np.diff(quantiles) >= 0)


def test_quantile_transformer_with_sample_weight():
rng = np.random.RandomState(42)
X = rng.randn(100, 1)
sample_weight = np.linspace(1, 2, X.shape[0]) # increasing weights

qt_weighted = QuantileTransformer(n_quantiles=10, random_state=0)
qt_unweighted = QuantileTransformer(n_quantiles=10, random_state=0)

Xt_weighted = qt_weighted.fit_transform(X, sample_weight=sample_weight)
Xt_unweighted = qt_unweighted.fit_transform(X)

# Should not be equal
with pytest.raises(AssertionError):
np.testing.assert_allclose(Xt_weighted, Xt_unweighted)

# Inverse transform round-trip check
X_roundtrip = qt_weighted.inverse_transform(Xt_weighted)
np.testing.assert_allclose(
X[~np.isnan(X)], X_roundtrip[~np.isnan(X)], rtol=1e-2, atol=1e-2
)
Copy link
Member

Choose a reason for hiding this comment

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

I think this test is not necessary since there is already a common test check_sample_weight_equivalence_on_dense_data that should be executed when running the following command:

pytest -v -k "QuantileTransformer and check_sample_weight_equivalence" sklearn/tests/test_common.py



def test_robust_scaler_invalid_range():
for range_ in [
(-1, 90),
Expand Down
6 changes: 6 additions & 0 deletions sklearn/utils/_test_common/instance_generator.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -162,6 +162,7 @@
StandardScaler,
TargetEncoder,
)
from sklearn.preprocessing._data import QuantileTransformer
from sklearn.random_projection import (
GaussianRandomProjection,
SparseRandomProjection,
Expand Down Expand Up @@ -1255,6 +1256,11 @@ def _yield_instances_for_check(check, estimator_orig):
" tests are added for TunedThresholdClassifierCV specifically."
),
},
QuantileTransformer: {
"check_sample_weight_equivalence_on_sparse_data": (
"QuantileTransformer does not yet support sample_weight on sparse data."
),
},
}

# TODO: remove when scipy min version >= 1.11
Expand Down
Loading