Thanks to visit codestin.com
Credit goes to github.com

Skip to content

[MRG+1] avoid integer overflow by using floats for matthews_corrcoef #9693

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 11, 2017
Merged

[MRG+1] avoid integer overflow by using floats for matthews_corrcoef #9693

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
f951bbf
Fix bug#9622: avoid integer overflow by using floats for matthews_cor…
sam-s Sep 5, 2017
675e0e7
matthews_corrcoef: cosmetic change requested by jnothman
sam-s Sep 6, 2017
eebb584
Add test_matthews_corrcoef_overflow for Bug#9622
sam-s Sep 6, 2017
8963d40
test_matthews_corrcoef_overflow: clean-up and make deterministic
sam-s Sep 6, 2017
80284db
matthews_corrcoef: pass dtype=np.float64 to sum & trace instead of us…
sam-s Sep 7, 2017
ac01384
test_matthews_corrcoef_overflow: add simple deterministic tests
sam-s Sep 7, 2017
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
8 changes: 4 additions & 4 deletions sklearn/metrics/classification.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -167,7 +167,7 @@ def accuracy_score(y_true, y_pred, normalize=True, sample_weight=None):
2

In the multilabel case with binary label indicators:

>>> accuracy_score(np.array([[0, 1], [1, 1]]), np.ones((2, 2)))
0.5
"""
Expand Down Expand Up @@ -528,9 +528,9 @@ def matthews_corrcoef(y_true, y_pred, sample_weight=None):
y_pred = lb.transform(y_pred)

C = confusion_matrix(y_true, y_pred, sample_weight=sample_weight)
t_sum = C.sum(axis=1)
p_sum = C.sum(axis=0)
n_correct = np.trace(C)
t_sum = C.sum(axis=1, dtype=np.float64)
p_sum = C.sum(axis=0, dtype=np.float64)
n_correct = np.trace(C, dtype=np.float64)
n_samples = p_sum.sum()
cov_ytyp = n_correct * n_samples - np.dot(t_sum, p_sum)
cov_ypyp = n_samples ** 2 - np.dot(p_sum, p_sum)
Expand Down
35 changes: 35 additions & 0 deletions sklearn/metrics/tests/test_classification.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -483,6 +483,41 @@ def test_matthews_corrcoef_multiclass():
assert_almost_equal(mcc, 0.)


def test_matthews_corrcoef_overflow():
# https://github.com/scikit-learn/scikit-learn/issues/9622
rng = np.random.RandomState(20170906)

def mcc_safe(y_true, y_pred):
conf_matrix = confusion_matrix(y_true, y_pred)
true_pos = conf_matrix[1, 1]
false_pos = conf_matrix[1, 0]
false_neg = conf_matrix[0, 1]
n_points = len(y_true)
pos_rate = (true_pos + false_neg) / n_points
activity = (true_pos + false_pos) / n_points
mcc_numerator = true_pos / n_points - pos_rate * activity
mcc_denominator = activity * pos_rate * (1 - activity) * (1 - pos_rate)
return mcc_numerator / np.sqrt(mcc_denominator)

def random_ys(n_points): # binary
x_true = rng.random_sample(n_points)
x_pred = x_true + 0.2 * (rng.random_sample(n_points) - 0.5)
y_true = (x_true > 0.5)
y_pred = (x_pred > 0.5)
return y_true, y_pred

for n_points in [100, 10000, 1000000]:
arr = np.repeat([0., 1.], n_points) # binary
assert_almost_equal(matthews_corrcoef(arr, arr), 1.0)
arr = np.repeat([0., 1., 2.], n_points) # multiclass
assert_almost_equal(matthews_corrcoef(arr, arr), 1.0)

y_true, y_pred = random_ys(n_points)
assert_almost_equal(matthews_corrcoef(y_true, y_true), 1.0)
assert_almost_equal(matthews_corrcoef(y_true, y_pred),
mcc_safe(y_true, y_pred))


def test_precision_recall_f1_score_multiclass():
# Test Precision Recall and F1 Score for multiclass classification task
y_true, y_pred, _ = make_prediction(binary=False)
Expand Down