10000 ENH Add probability calibration based on isotonic regr. and Platt's s… · scikit-learn/scikit-learn@ecfc93d · GitHub
[go: up one dir, main page]
Skip to content

Commit ecfc93d

Browse files
Jan Hendrik Metzenogrisel
Jan Hendrik Metzen
authored and
ogrisel
committed
ENH Add probability calibration based on isotonic regr. and Platt's sigmoid fit + calibration-curve
CalibratedClassifierCV allows to calibrate the predicted probabilities of base classifiers based on a cross-validation scheme and either Platt's sigmoid fit or isotonic regression. This can be used to compensate for an under-confident or over-confident classifier. It allows also to turn the decision scores of a non-probabilistic classifier into valid probabilities. The function calibration_curve allows to evaluate how well calibrated the probabilties returned by a classifier are. Ideally, the curve should be close to diagonal.
1 parent 1a88eb0 commit ecfc93d

File tree

2 files changed

+526
-1
lines changed

2 files changed

+526
-1
lines changed

‎sk 7D7F learn/__init__.py

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -56,7 +56,7 @@
5656
from .base import clone
5757
__check_build # avoid flakes unused variable error
5858

59-
__all__ = ['cluster', 'covariance', 'cross_decomposition',
59+
__all__ = ['calibration', 'cluster', 'covariance', 'cross_decomposition',
6060
'cross_validation', 'datasets', 'decomposition', 'dummy',
6161
'ensemble', 'externals', 'feature_extraction',
6262
'feature_selection', 'gaussian_process', 'grid_search', 'hmm',

0 commit comments

Comments
 (0)
0