Skip to content

Fix cupy error in KernelDensity with epanechnikov #6388

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

Merged
rapids-bot merged 2 commits into from
Mar 5, 2025
Merged

Fix cupy error in KernelDensity with epanechnikov #6388

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
a647176
Fix cupy error in `KernelDensity` with `epanechnikov`
jcrist Mar 3, 2025
f73f574
Merge branch 'branch-25.04' into fix-epanechnikov-kernel-density
jcrist Mar 4, 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
14 changes: 11 additions & 3 deletions python/cuml/cuml/neighbors/kernel_density.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
#
# Copyright (c) 2022-2024, NVIDIA CORPORATION.
# Copyright (c) 2022-2025, NVIDIA CORPORATION.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
Expand Down Expand Up @@ -61,14 +61,22 @@ def tophat_log_kernel(x, h):


@cp.fuse()
def epanechnikov_log_kernel(x, h):
def _epanechnikov_log_kernel(x, h, h_squared):
# don't call log(0) otherwise we get NaNs
z = cp.maximum(1.0 - (x * x) / (h * h), 1e-30)
z = cp.maximum(1.0 - (x * x) / h_squared, 1e-30)
y = (x < h) * cp.log(z)
y += (x >= h) * np.finfo(y.dtype).min
return y


def epanechnikov_log_kernel(x, h):
# TODO: Due to https://github.com/cupy/cupy/issues/8536 cupy.fuse errors when trying
# to compile the elementwise operation in epanechnikov. Handling `h * h` on host
# (where `h` is a host scalar) seems to work around the bug completely. Once the upstream
# issue is fixed this can be reverted.
return _epanechnikov_log_kernel(x, h, h * h)


@cp.fuse()
def exponential_log_kernel(x, h):
return -x / h
Expand Down
12 changes: 3 additions & 9 deletions python/cuml/cuml/tests/test_kernel_density.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
#
# Copyright (c) 2022-2024, NVIDIA CORPORATION.
# Copyright (c) 2022-2025, NVIDIA CORPORATION.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
Expand All @@ -16,7 +16,6 @@

from cuml.testing.utils import as_type
from sklearn.model_selection import GridSearchCV
import os
import pytest
from hypothesis.extra.numpy import arrays
from hypothesis import given, settings, assume, strategies as st
Expand Down Expand Up @@ -82,10 +81,6 @@ def array_strategy(draw):
)


@pytest.mark.skipif(
os.environ.get("CUDA_VERSION") == "12.0.1",
reason="Flaky hypothesis CI failure: https://github.com/rapidsai/cuml/issues/6173",
)
@settings(deadline=None)
@given(
array_strategy(),
Expand All @@ -101,9 +96,8 @@ def test_kernel_density(arrays, kernel, metric, bandwidth):
# cosine is numerically unstable at high dimensions
# for both cuml and sklearn
assume(X.shape[1] <= 20)
kde = KernelDensity(kernel=kernel, metric=metric, bandwidth=bandwidth).fit(
X, sample_weight=sample_weight
)
kde = KernelDensity(kernel=kernel, metric=metric, bandwidth=bandwidth)
kde.fit(X, sample_weight=sample_weight)
cuml_prob = kde.score_samples(X)
cuml_prob_test = kde.score_samples(X_test)

Expand Down