8000 [MRG + 1] FIX use high precision cumsum and check it is stable enough by jnothman · Pull Request #7331 · scikit-learn/scikit-learn · GitHub
[go: up one dir, main page]
Skip to content

[MRG + 1] FIX use high precision cumsum and check it is stable enough #7331

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
merged 6 commits into from
Sep 9, 2016
Merged

[MRG + 1] FIX use high precision cumsum and check it is stable enough #7331

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
6ce5f18
FIX use high precision cumsum and check it is stable enough
jnothman Sep 2, 2016
bd19e30
FIX corrected code
jnothman Sep 3, 2016
f58fdf7
Improved wording of error message
jnothman Sep 3, 2016
9ced9da
TST Add test with reduced rtol, atol
jnothman Sep 8, 2016
0cf09bc
Limit test to numpy >=1.9
jnothman Sep 8, 2016
eb3500a
TST correct numpy version check
jnothman Sep 8, 2016
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
5 changes: 3 additions & 2 deletions sklearn/metrics/ranking.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -27,6 +27,7 @@
from ..utils import check_consistent_length
from ..utils import column_or_1d, check_array
from ..utils.multiclass import type_of_target
from ..utils.extmath import stable_cumsum
from ..utils.fixes import isclose
from ..utils.fixes import bincount
from ..utils.fixes import array_equal
Expand Down Expand Up @@ -337,9 +338,9 @@ def _binary_clf_curve(y_true, y_score, pos_label=None, sample_weight=None):
threshold_idxs = np.r_[distinct_value_indices, y_true.size - 1]

# accumulate the true positives with decreasing threshold
tps = (y_true * weight).cumsum()[threshold_idxs]
tps = stable_cumsum(y_true * weight)[threshold_idxs]
if sample_weight is not None:
fps = weight.cumsum()[threshold_idxs] - tps
fps = stable_cumsum(weight)[threshold_idxs] - tps
else:
fps = 1 + threshold_idxs - tps
return fps, tps, y_score[threshold_idxs]
Expand Down
20 changes: 20 additions & 0 deletions sklearn/utils/extmath.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -851,3 +851,23 @@ def _deterministic_vector_sign_flip(u):
signs = np.sign(u[range(u.shape[0]), max_abs_rows])
u *= signs[:, np.newaxis]
return u


def stable_cumsum(arr, rtol=1e-05, atol=1e-08):
"""Use high precision for cumsum and check that final value matches sum

Parameters
----------
arr : array-like
To be cumulatively summed as flat
rtol : float
Relative tolerance, see ``np.allclose``
atol : float
Absolute tolerance, see ``np.allclose``
"""
out = np.cumsum(arr, dtype=np.float64)
expected = np.sum(arr, dtype=np.float64)
if not np.allclose(out[-1], expected, rtol=rtol, atol=atol):
raise RuntimeError('cumsum was found to be unstable: '
'its last element does not correspond to sum')
return out
15 changes: 15 additions & 0 deletions sklearn/utils/tests/test_extmath.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,7 +17,10 @@
from sklearn.utils.testing import assert_false
from sklearn.utils.testing import assert_greater
from sklearn.utils.testing import assert_raises
from sklearn.utils.testing import assert_raise_message
from sklearn.utils.testing import skip_if_32bit
from sklearn.utils.testing import SkipTest
from sklearn.utils.fixes import np_version

from sklearn.utils.extmath import density
from sklearn.utils.extmath import logsumexp
Expand All @@ -32,6 +35,7 @@
from sklearn.utils.extmath import _incremental_mean_and_var
from sklearn.utils.extmath import _deterministic_vector_sign_flip
from sklearn.utils.extmath import softmax
from sklearn.utils.extmath import stable_cumsum
from sklearn.datasets.samples_generator import make_low_rank_matrix


Expand Down Expand Up @@ -643,3 +647,14 @@ def test_softmax():
exp_X = np.exp(X)
sum_exp_X = np.sum(exp_X, axis=1).reshape((-1, 1))
assert_array_almost_equal(softmax(X), exp_X / sum_exp_X)


def test_stable_cumsum():
if np_version < (1, 9):
raise SkipTest("Sum is as unstable as cumsum for numpy < 1.9")
assert_array_equal(stable_cumsum([1, 2, 3]), np.cumsum([1, 2, 3]))
r = np.random.RandomState(0).rand(100000)
assert_raise_message(RuntimeError,
'cumsum was found to be unstable: its last element '
'does not correspond to sum',
stable_cumsum, r, rtol=0, atol=0)
0