Add files using upload-large-folder tool
Browse filesThis view is limited to 50 files because it contains too many changes. See raw diff
- .gitattributes +5 -0
- videollama2/lib/python3.10/site-packages/sklearn/cluster/_k_means_minibatch.cpython-310-x86_64-linux-gnu.so +3 -0
- videollama2/lib/python3.10/site-packages/sklearn/linear_model/_sag_fast.cpython-310-x86_64-linux-gnu.so +3 -0
- videollama2/lib/python3.10/site-packages/sklearn/metrics/_base.py +251 -0
- videollama2/lib/python3.10/site-packages/sklearn/metrics/_pairwise_distances_reduction/__init__.py +101 -0
- videollama2/lib/python3.10/site-packages/sklearn/metrics/_pairwise_distances_reduction/_dispatcher.py +449 -0
- videollama2/lib/python3.10/site-packages/sklearn/metrics/_ranking.py +1826 -0
- videollama2/lib/python3.10/site-packages/sklearn/tree/_utils.cpython-310-x86_64-linux-gnu.so +3 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/_available_if.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/_encode.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/_mask.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/_mocking.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/_set_output.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/_show_versions.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/_tags.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/_testing.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/deprecation.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/estimator_checks.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/extmath.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/fixes.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/graph.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/multiclass.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/optimize.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/stats.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/validation.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/_estimator_html_repr.py +428 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/_fast_dict.pxd +22 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/_isfinite.cpython-310-x86_64-linux-gnu.so +3 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/_random.pxd +39 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/_seq_dataset.pxd +116 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/_sorting.pxd +9 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/_typedefs.cpython-310-x86_64-linux-gnu.so +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/_vector_sentinel.cpython-310-x86_64-linux-gnu.so +3 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/deprecation.py +115 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/fixes.py +216 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/tests/__pycache__/__init__.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/tests/__pycache__/conftest.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/tests/__pycache__/test_array_api.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/tests/__pycache__/test_arrayfuncs.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/tests/__pycache__/test_class_weight.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/tests/__pycache__/test_cython_blas.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/tests/__pycache__/test_cython_templating.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/tests/__pycache__/test_deprecation.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/tests/__pycache__/test_encode.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/tests/__pycache__/test_estimator_checks.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/tests/__pycache__/test_estimator_html_repr.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/tests/__pycache__/test_fast_dict.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/tests/__pycache__/test_mocking.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/tests/__pycache__/test_murmurhash.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/tests/__pycache__/test_optimize.cpython-310.pyc +0 -0
.gitattributes
CHANGED
|
@@ -1926,3 +1926,8 @@ videollama2/lib/python3.10/site-packages/sklearn/cluster/_k_means_common.cpython
|
|
| 1926 |
videollama2/lib/python3.10/site-packages/sklearn/cluster/_hierarchical_fast.cpython-310-x86_64-linux-gnu.so filter=lfs diff=lfs merge=lfs -text
|
| 1927 |
parrot/lib/python3.10/site-packages/opencv_python.libs/libssl-28bef1ac.so.1.1 filter=lfs diff=lfs merge=lfs -text
|
| 1928 |
parrot/lib/python3.10/site-packages/opencv_python.libs/libgfortran-91cc3cb1.so.3.0.0 filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1926 |
videollama2/lib/python3.10/site-packages/sklearn/cluster/_hierarchical_fast.cpython-310-x86_64-linux-gnu.so filter=lfs diff=lfs merge=lfs -text
|
| 1927 |
parrot/lib/python3.10/site-packages/opencv_python.libs/libssl-28bef1ac.so.1.1 filter=lfs diff=lfs merge=lfs -text
|
| 1928 |
parrot/lib/python3.10/site-packages/opencv_python.libs/libgfortran-91cc3cb1.so.3.0.0 filter=lfs diff=lfs merge=lfs -text
|
| 1929 |
+
videollama2/lib/python3.10/site-packages/sklearn/cluster/_k_means_minibatch.cpython-310-x86_64-linux-gnu.so filter=lfs diff=lfs merge=lfs -text
|
| 1930 |
+
videollama2/lib/python3.10/site-packages/sklearn/linear_model/_sag_fast.cpython-310-x86_64-linux-gnu.so filter=lfs diff=lfs merge=lfs -text
|
| 1931 |
+
videollama2/lib/python3.10/site-packages/sklearn/utils/_isfinite.cpython-310-x86_64-linux-gnu.so filter=lfs diff=lfs merge=lfs -text
|
| 1932 |
+
videollama2/lib/python3.10/site-packages/sklearn/utils/_vector_sentinel.cpython-310-x86_64-linux-gnu.so filter=lfs diff=lfs merge=lfs -text
|
| 1933 |
+
videollama2/lib/python3.10/site-packages/sklearn/tree/_utils.cpython-310-x86_64-linux-gnu.so filter=lfs diff=lfs merge=lfs -text
|
videollama2/lib/python3.10/site-packages/sklearn/cluster/_k_means_minibatch.cpython-310-x86_64-linux-gnu.so
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:02427d73dd89676ee2cc95481e05e33b647a6aafeda54130d205c4560e762d5b
|
| 3 |
+
size 274121
|
videollama2/lib/python3.10/site-packages/sklearn/linear_model/_sag_fast.cpython-310-x86_64-linux-gnu.so
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:b0206e05f51082e3e1af0024f8321918f56819b5df690dbc159e81b081ccd9fd
|
| 3 |
+
size 170705
|
videollama2/lib/python3.10/site-packages/sklearn/metrics/_base.py
ADDED
|
@@ -0,0 +1,251 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""
|
| 2 |
+
Common code for all metrics.
|
| 3 |
+
|
| 4 |
+
"""
|
| 5 |
+
# Authors: Alexandre Gramfort <alexandre.gramfort@inria.fr>
|
| 6 |
+
# Mathieu Blondel <mathieu@mblondel.org>
|
| 7 |
+
# Olivier Grisel <olivier.grisel@ensta.org>
|
| 8 |
+
# Arnaud Joly <a.joly@ulg.ac.be>
|
| 9 |
+
# Jochen Wersdorfer <jochen@wersdoerfer.de>
|
| 10 |
+
# Lars Buitinck
|
| 11 |
+
# Joel Nothman <joel.nothman@gmail.com>
|
| 12 |
+
# Noel Dawe <noel@dawe.me>
|
| 13 |
+
# License: BSD 3 clause
|
| 14 |
+
|
| 15 |
+
from itertools import combinations
|
| 16 |
+
|
| 17 |
+
import numpy as np
|
| 18 |
+
|
| 19 |
+
from ..utils import check_array, check_consistent_length
|
| 20 |
+
from ..utils.multiclass import type_of_target
|
| 21 |
+
|
| 22 |
+
|
| 23 |
+
def _average_binary_score(binary_metric, y_true, y_score, average, sample_weight=None):
|
| 24 |
+
"""Average a binary metric for multilabel classification.
|
| 25 |
+
|
| 26 |
+
Parameters
|
| 27 |
+
----------
|
| 28 |
+
y_true : array, shape = [n_samples] or [n_samples, n_classes]
|
| 29 |
+
True binary labels in binary label indicators.
|
| 30 |
+
|
| 31 |
+
y_score : array, shape = [n_samples] or [n_samples, n_classes]
|
| 32 |
+
Target scores, can either be probability estimates of the positive
|
| 33 |
+
class, confidence values, or binary decisions.
|
| 34 |
+
|
| 35 |
+
average : {None, 'micro', 'macro', 'samples', 'weighted'}, default='macro'
|
| 36 |
+
If ``None``, the scores for each class are returned. Otherwise,
|
| 37 |
+
this determines the type of averaging performed on the data:
|
| 38 |
+
|
| 39 |
+
``'micro'``:
|
| 40 |
+
Calculate metrics globally by considering each element of the label
|
| 41 |
+
indicator matrix as a label.
|
| 42 |
+
``'macro'``:
|
| 43 |
+
Calculate metrics for each label, and find their unweighted
|
| 44 |
+
mean. This does not take label imbalance into account.
|
| 45 |
+
``'weighted'``:
|
| 46 |
+
Calculate metrics for each label, and find their average, weighted
|
| 47 |
+
by support (the number of true instances for each label).
|
| 48 |
+
``'samples'``:
|
| 49 |
+
Calculate metrics for each instance, and find their average.
|
| 50 |
+
|
| 51 |
+
Will be ignored when ``y_true`` is binary.
|
| 52 |
+
|
| 53 |
+
sample_weight : array-like of shape (n_samples,), default=None
|
| 54 |
+
Sample weights.
|
| 55 |
+
|
| 56 |
+
binary_metric : callable, returns shape [n_classes]
|
| 57 |
+
The binary metric function to use.
|
| 58 |
+
|
| 59 |
+
Returns
|
| 60 |
+
-------
|
| 61 |
+
score : float or array of shape [n_classes]
|
| 62 |
+
If not ``None``, average the score, else return the score for each
|
| 63 |
+
classes.
|
| 64 |
+
|
| 65 |
+
"""
|
| 66 |
+
average_options = (None, "micro", "macro", "weighted", "samples")
|
| 67 |
+
if average not in average_options:
|
| 68 |
+
raise ValueError("average has to be one of {0}".format(average_options))
|
| 69 |
+
|
| 70 |
+
y_type = type_of_target(y_true)
|
| 71 |
+
if y_type not in ("binary", "multilabel-indicator"):
|
| 72 |
+
raise ValueError("{0} format is not supported".format(y_type))
|
| 73 |
+
|
| 74 |
+
if y_type == "binary":
|
| 75 |
+
return binary_metric(y_true, y_score, sample_weight=sample_weight)
|
| 76 |
+
|
| 77 |
+
check_consistent_length(y_true, y_score, sample_weight)
|
| 78 |
+
y_true = check_array(y_true)
|
| 79 |
+
y_score = check_array(y_score)
|
| 80 |
+
|
| 81 |
+
not_average_axis = 1
|
| 82 |
+
score_weight = sample_weight
|
| 83 |
+
average_weight = None
|
| 84 |
+
|
| 85 |
+
if average == "micro":
|
| 86 |
+
if score_weight is not None:
|
| 87 |
+
score_weight = np.repeat(score_weight, y_true.shape[1])
|
| 88 |
+
y_true = y_true.ravel()
|
| 89 |
+
y_score = y_score.ravel()
|
| 90 |
+
|
| 91 |
+
elif average == "weighted":
|
| 92 |
+
if score_weight is not None:
|
| 93 |
+
average_weight = np.sum(
|
| 94 |
+
np.multiply(y_true, np.reshape(score_weight, (-1, 1))), axis=0
|
| 95 |
+
)
|
| 96 |
+
else:
|
| 97 |
+
average_weight = np.sum(y_true, axis=0)
|
| 98 |
+
if np.isclose(average_weight.sum(), 0.0):
|
| 99 |
+
return 0
|
| 100 |
+
|
| 101 |
+
elif average == "samples":
|
| 102 |
+
# swap average_weight <-> score_weight
|
| 103 |
+
average_weight = score_weight
|
| 104 |
+
score_weight = None
|
| 105 |
+
not_average_axis = 0
|
| 106 |
+
|
| 107 |
+
if y_true.ndim == 1:
|
| 108 |
+
y_true = y_true.reshape((-1, 1))
|
| 109 |
+
|
| 110 |
+
if y_score.ndim == 1:
|
| 111 |
+
y_score = y_score.reshape((-1, 1))
|
| 112 |
+
|
| 113 |
+
n_classes = y_score.shape[not_average_axis]
|
| 114 |
+
score = np.zeros((n_classes,))
|
| 115 |
+
for c in range(n_classes):
|
| 116 |
+
y_true_c = y_true.take([c], axis=not_average_axis).ravel()
|
| 117 |
+
y_score_c = y_score.take([c], axis=not_average_axis).ravel()
|
| 118 |
+
score[c] = binary_metric(y_true_c, y_score_c, sample_weight=score_weight)
|
| 119 |
+
|
| 120 |
+
# Average the results
|
| 121 |
+
if average is not None:
|
| 122 |
+
if average_weight is not None:
|
| 123 |
+
# Scores with 0 weights are forced to be 0, preventing the average
|
| 124 |
+
# score from being affected by 0-weighted NaN elements.
|
| 125 |
+
average_weight = np.asarray(average_weight)
|
| 126 |
+
score[average_weight == 0] = 0
|
| 127 |
+
return np.average(score, weights=average_weight)
|
| 128 |
+
else:
|
| 129 |
+
return score
|
| 130 |
+
|
| 131 |
+
|
| 132 |
+
def _average_multiclass_ovo_score(binary_metric, y_true, y_score, average="macro"):
|
| 133 |
+
"""Average one-versus-one scores for multiclass classification.
|
| 134 |
+
|
| 135 |
+
Uses the binary metric for one-vs-one multiclass classification,
|
| 136 |
+
where the score is computed according to the Hand & Till (2001) algorithm.
|
| 137 |
+
|
| 138 |
+
Parameters
|
| 139 |
+
----------
|
| 140 |
+
binary_metric : callable
|
| 141 |
+
The binary metric function to use that accepts the following as input:
|
| 142 |
+
y_true_target : array, shape = [n_samples_target]
|
| 143 |
+
Some sub-array of y_true for a pair of classes designated
|
| 144 |
+
positive and negative in the one-vs-one scheme.
|
| 145 |
+
y_score_target : array, shape = [n_samples_target]
|
| 146 |
+
Scores corresponding to the probability estimates
|
| 147 |
+
of a sample belonging to the designated positive class label
|
| 148 |
+
|
| 149 |
+
y_true : array-like of shape (n_samples,)
|
| 150 |
+
True multiclass labels.
|
| 151 |
+
|
| 152 |
+
y_score : array-like of shape (n_samples, n_classes)
|
| 153 |
+
Target scores corresponding to probability estimates of a sample
|
| 154 |
+
belonging to a particular class.
|
| 155 |
+
|
| 156 |
+
average : {'macro', 'weighted'}, default='macro'
|
| 157 |
+
Determines the type of averaging performed on the pairwise binary
|
| 158 |
+
metric scores:
|
| 159 |
+
``'macro'``:
|
| 160 |
+
Calculate metrics for each label, and find their unweighted
|
| 161 |
+
mean. This does not take label imbalance into account. Classes
|
| 162 |
+
are assumed to be uniformly distributed.
|
| 163 |
+
``'weighted'``:
|
| 164 |
+
Calculate metrics for each label, taking into account the
|
| 165 |
+
prevalence of the classes.
|
| 166 |
+
|
| 167 |
+
Returns
|
| 168 |
+
-------
|
| 169 |
+
score : float
|
| 170 |
+
Average of the pairwise binary metric scores.
|
| 171 |
+
"""
|
| 172 |
+
check_consistent_length(y_true, y_score)
|
| 173 |
+
|
| 174 |
+
y_true_unique = np.unique(y_true)
|
| 175 |
+
n_classes = y_true_unique.shape[0]
|
| 176 |
+
n_pairs = n_classes * (n_classes - 1) // 2
|
| 177 |
+
pair_scores = np.empty(n_pairs)
|
| 178 |
+
|
| 179 |
+
is_weighted = average == "weighted"
|
| 180 |
+
prevalence = np.empty(n_pairs) if is_weighted else None
|
| 181 |
+
|
| 182 |
+
# Compute scores treating a as positive class and b as negative class,
|
| 183 |
+
# then b as positive class and a as negative class
|
| 184 |
+
for ix, (a, b) in enumerate(combinations(y_true_unique, 2)):
|
| 185 |
+
a_mask = y_true == a
|
| 186 |
+
b_mask = y_true == b
|
| 187 |
+
ab_mask = np.logical_or(a_mask, b_mask)
|
| 188 |
+
|
| 189 |
+
if is_weighted:
|
| 190 |
+
prevalence[ix] = np.average(ab_mask)
|
| 191 |
+
|
| 192 |
+
a_true = a_mask[ab_mask]
|
| 193 |
+
b_true = b_mask[ab_mask]
|
| 194 |
+
|
| 195 |
+
a_true_score = binary_metric(a_true, y_score[ab_mask, a])
|
| 196 |
+
b_true_score = binary_metric(b_true, y_score[ab_mask, b])
|
| 197 |
+
pair_scores[ix] = (a_true_score + b_true_score) / 2
|
| 198 |
+
|
| 199 |
+
return np.average(pair_scores, weights=prevalence)
|
| 200 |
+
|
| 201 |
+
|
| 202 |
+
def _check_pos_label_consistency(pos_label, y_true):
|
| 203 |
+
"""Check if `pos_label` need to be specified or not.
|
| 204 |
+
|
| 205 |
+
In binary classification, we fix `pos_label=1` if the labels are in the set
|
| 206 |
+
{-1, 1} or {0, 1}. Otherwise, we raise an error asking to specify the
|
| 207 |
+
`pos_label` parameters.
|
| 208 |
+
|
| 209 |
+
Parameters
|
| 210 |
+
----------
|
| 211 |
+
pos_label : int, str or None
|
| 212 |
+
The positive label.
|
| 213 |
+
y_true : ndarray of shape (n_samples,)
|
| 214 |
+
The target vector.
|
| 215 |
+
|
| 216 |
+
Returns
|
| 217 |
+
-------
|
| 218 |
+
pos_label : int
|
| 219 |
+
If `pos_label` can be inferred, it will be returned.
|
| 220 |
+
|
| 221 |
+
Raises
|
| 222 |
+
------
|
| 223 |
+
ValueError
|
| 224 |
+
In the case that `y_true` does not have label in {-1, 1} or {0, 1},
|
| 225 |
+
it will raise a `ValueError`.
|
| 226 |
+
"""
|
| 227 |
+
# ensure binary classification if pos_label is not specified
|
| 228 |
+
# classes.dtype.kind in ('O', 'U', 'S') is required to avoid
|
| 229 |
+
# triggering a FutureWarning by calling np.array_equal(a, b)
|
| 230 |
+
# when elements in the two arrays are not comparable.
|
| 231 |
+
classes = np.unique(y_true)
|
| 232 |
+
if pos_label is None and (
|
| 233 |
+
classes.dtype.kind in "OUS"
|
| 234 |
+
or not (
|
| 235 |
+
np.array_equal(classes, [0, 1])
|
| 236 |
+
or np.array_equal(classes, [-1, 1])
|
| 237 |
+
or np.array_equal(classes, [0])
|
| 238 |
+
or np.array_equal(classes, [-1])
|
| 239 |
+
or np.array_equal(classes, [1])
|
| 240 |
+
)
|
| 241 |
+
):
|
| 242 |
+
classes_repr = ", ".join(repr(c) for c in classes)
|
| 243 |
+
raise ValueError(
|
| 244 |
+
f"y_true takes value in {{{classes_repr}}} and pos_label is not "
|
| 245 |
+
"specified: either make y_true take value in {0, 1} or "
|
| 246 |
+
"{-1, 1} or pass pos_label explicitly."
|
| 247 |
+
)
|
| 248 |
+
elif pos_label is None:
|
| 249 |
+
pos_label = 1
|
| 250 |
+
|
| 251 |
+
return pos_label
|
videollama2/lib/python3.10/site-packages/sklearn/metrics/_pairwise_distances_reduction/__init__.py
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Pairwise Distances Reductions
|
| 2 |
+
# =============================
|
| 3 |
+
#
|
| 4 |
+
# Author: Julien Jerphanion <git@jjerphan.xyz>
|
| 5 |
+
#
|
| 6 |
+
# Overview
|
| 7 |
+
# --------
|
| 8 |
+
#
|
| 9 |
+
# This module provides routines to compute pairwise distances between a set
|
| 10 |
+
# of row vectors of X and another set of row vectors of Y and apply a
|
| 11 |
+
# reduction on top. The canonical example is the brute-force computation
|
| 12 |
+
# of the top k nearest neighbors by leveraging the arg-k-min reduction.
|
| 13 |
+
#
|
| 14 |
+
# The reduction takes a matrix of pairwise distances between rows of X and Y
|
| 15 |
+
# as input and outputs an aggregate data-structure for each row of X. The
|
| 16 |
+
# aggregate values are typically smaller than the number of rows in Y, hence
|
| 17 |
+
# the term reduction.
|
| 18 |
+
#
|
| 19 |
+
# For computational reasons, the reduction are performed on the fly on chunks
|
| 20 |
+
# of rows of X and Y so as to keep intermediate data-structures in CPU cache
|
| 21 |
+
# and avoid unnecessary round trips of large distance arrays with the RAM
|
| 22 |
+
# that would otherwise severely degrade the speed by making the overall
|
| 23 |
+
# processing memory-bound.
|
| 24 |
+
#
|
| 25 |
+
# Finally, the routines follow a generic parallelization template to process
|
| 26 |
+
# chunks of data with OpenMP loops (via Cython prange), either on rows of X
|
| 27 |
+
# or rows of Y depending on their respective sizes.
|
| 28 |
+
#
|
| 29 |
+
#
|
| 30 |
+
# Dispatching to specialized implementations
|
| 31 |
+
# ------------------------------------------
|
| 32 |
+
#
|
| 33 |
+
# Dispatchers are meant to be used in the Python code. Under the hood, a
|
| 34 |
+
# dispatcher must only define the logic to choose at runtime to the correct
|
| 35 |
+
# dtype-specialized :class:`BaseDistancesReductionDispatcher` implementation based
|
| 36 |
+
# on the dtype of X and of Y.
|
| 37 |
+
#
|
| 38 |
+
#
|
| 39 |
+
# High-level diagram
|
| 40 |
+
# ------------------
|
| 41 |
+
#
|
| 42 |
+
# Legend:
|
| 43 |
+
#
|
| 44 |
+
# A ---⊳ B: A inherits from B
|
| 45 |
+
# A ---x B: A dispatches to B
|
| 46 |
+
#
|
| 47 |
+
#
|
| 48 |
+
# (base dispatcher)
|
| 49 |
+
# BaseDistancesReductionDispatcher
|
| 50 |
+
# ∆
|
| 51 |
+
# |
|
| 52 |
+
# |
|
| 53 |
+
# +-----------------------+----------------------+
|
| 54 |
+
# | |
|
| 55 |
+
# (dispatcher) (dispatcher)
|
| 56 |
+
# ArgKmin RadiusNeighbors
|
| 57 |
+
# | |
|
| 58 |
+
# | |
|
| 59 |
+
# | (float{32,64} implem.) |
|
| 60 |
+
# | BaseDistancesReduction{32,64} |
|
| 61 |
+
# | ∆ |
|
| 62 |
+
# | | |
|
| 63 |
+
# | | |
|
| 64 |
+
# | +-----------------+-----------------+ |
|
| 65 |
+
# | | | |
|
| 66 |
+
# | | | |
|
| 67 |
+
# x | | x
|
| 68 |
+
# ArgKmin{32,64} RadiusNeighbors{32,64}
|
| 69 |
+
# | ∆ ∆ |
|
| 70 |
+
# | | | |
|
| 71 |
+
# ======================= Specializations =============================
|
| 72 |
+
# | | | |
|
| 73 |
+
# | | | |
|
| 74 |
+
# x | | x
|
| 75 |
+
# EuclideanArgKmin{32,64} EuclideanRadiusNeighbors{32,64}
|
| 76 |
+
#
|
| 77 |
+
# For instance :class:`ArgKmin` dispatches to:
|
| 78 |
+
# - :class:`ArgKmin64` if X and Y are two `float64` array-likes
|
| 79 |
+
# - :class:`ArgKmin32` if X and Y are two `float32` array-likes
|
| 80 |
+
#
|
| 81 |
+
# In addition, if the metric parameter is set to "euclidean" or "sqeuclidean",
|
| 82 |
+
# then `ArgKmin{32,64}` further dispatches to `EuclideanArgKmin{32,64}`. For
|
| 83 |
+
# example, :class:`ArgKmin64` would dispatch to :class:`EuclideanArgKmin64`, a
|
| 84 |
+
# specialized subclass that optimally handles the Euclidean distance case
|
| 85 |
+
# using Generalized Matrix Multiplication over `float64` data (see the
|
| 86 |
+
# docstring of :class:`GEMMTermComputer64` for details).
|
| 87 |
+
|
| 88 |
+
|
| 89 |
+
from ._dispatcher import (
|
| 90 |
+
BaseDistancesReductionDispatcher,
|
| 91 |
+
ArgKmin,
|
| 92 |
+
RadiusNeighbors,
|
| 93 |
+
sqeuclidean_row_norms,
|
| 94 |
+
)
|
| 95 |
+
|
| 96 |
+
__all__ = [
|
| 97 |
+
"BaseDistancesReductionDispatcher",
|
| 98 |
+
"ArgKmin",
|
| 99 |
+
"RadiusNeighbors",
|
| 100 |
+
"sqeuclidean_row_norms",
|
| 101 |
+
]
|
videollama2/lib/python3.10/site-packages/sklearn/metrics/_pairwise_distances_reduction/_dispatcher.py
ADDED
|
@@ -0,0 +1,449 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from abc import abstractmethod
|
| 2 |
+
|
| 3 |
+
import numpy as np
|
| 4 |
+
|
| 5 |
+
from typing import List
|
| 6 |
+
|
| 7 |
+
from scipy.sparse import isspmatrix_csr
|
| 8 |
+
|
| 9 |
+
from .._dist_metrics import BOOL_METRICS, METRIC_MAPPING
|
| 10 |
+
|
| 11 |
+
from ._base import _sqeuclidean_row_norms32, _sqeuclidean_row_norms64
|
| 12 |
+
from ._argkmin import (
|
| 13 |
+
ArgKmin64,
|
| 14 |
+
ArgKmin32,
|
| 15 |
+
)
|
| 16 |
+
from ._radius_neighbors import (
|
| 17 |
+
RadiusNeighbors64,
|
| 18 |
+
RadiusNeighbors32,
|
| 19 |
+
)
|
| 20 |
+
|
| 21 |
+
from ... import get_config
|
| 22 |
+
|
| 23 |
+
|
| 24 |
+
def sqeuclidean_row_norms(X, num_threads):
|
| 25 |
+
"""Compute the squared euclidean norm of the rows of X in parallel.
|
| 26 |
+
|
| 27 |
+
Parameters
|
| 28 |
+
----------
|
| 29 |
+
X : ndarray or CSR matrix of shape (n_samples, n_features)
|
| 30 |
+
Input data. Must be c-contiguous.
|
| 31 |
+
|
| 32 |
+
num_threads : int
|
| 33 |
+
The number of OpenMP threads to use.
|
| 34 |
+
|
| 35 |
+
Returns
|
| 36 |
+
-------
|
| 37 |
+
sqeuclidean_row_norms : ndarray of shape (n_samples,)
|
| 38 |
+
Arrays containing the squared euclidean norm of each row of X.
|
| 39 |
+
"""
|
| 40 |
+
if X.dtype == np.float64:
|
| 41 |
+
return np.asarray(_sqeuclidean_row_norms64(X, num_threads))
|
| 42 |
+
if X.dtype == np.float32:
|
| 43 |
+
return np.asarray(_sqeuclidean_row_norms32(X, num_threads))
|
| 44 |
+
|
| 45 |
+
raise ValueError(
|
| 46 |
+
"Only float64 or float32 datasets are supported at this time, "
|
| 47 |
+
f"got: X.dtype={X.dtype}."
|
| 48 |
+
)
|
| 49 |
+
|
| 50 |
+
|
| 51 |
+
class BaseDistancesReductionDispatcher:
|
| 52 |
+
"""Abstract base dispatcher for pairwise distance computation & reduction.
|
| 53 |
+
|
| 54 |
+
Each dispatcher extending the base :class:`BaseDistancesReductionDispatcher`
|
| 55 |
+
dispatcher must implement the :meth:`compute` classmethod.
|
| 56 |
+
"""
|
| 57 |
+
|
| 58 |
+
@classmethod
|
| 59 |
+
def valid_metrics(cls) -> List[str]:
|
| 60 |
+
excluded = {
|
| 61 |
+
# PyFunc cannot be supported because it necessitates interacting with
|
| 62 |
+
# the CPython interpreter to call user defined functions.
|
| 63 |
+
"pyfunc",
|
| 64 |
+
"mahalanobis", # is numerically unstable
|
| 65 |
+
# In order to support discrete distance metrics, we need to have a
|
| 66 |
+
# stable simultaneous sort which preserves the order of the indices
|
| 67 |
+
# because there generally is a lot of occurrences for a given values
|
| 68 |
+
# of distances in this case.
|
| 69 |
+
# TODO: implement a stable simultaneous_sort.
|
| 70 |
+
"hamming",
|
| 71 |
+
*BOOL_METRICS,
|
| 72 |
+
}
|
| 73 |
+
return sorted(({"sqeuclidean"} | set(METRIC_MAPPING.keys())) - excluded)
|
| 74 |
+
|
| 75 |
+
@classmethod
|
| 76 |
+
def is_usable_for(cls, X, Y, metric) -> bool:
|
| 77 |
+
"""Return True if the dispatcher can be used for the
|
| 78 |
+
given parameters.
|
| 79 |
+
|
| 80 |
+
Parameters
|
| 81 |
+
----------
|
| 82 |
+
X : {ndarray, sparse matrix} of shape (n_samples_X, n_features)
|
| 83 |
+
Input data.
|
| 84 |
+
|
| 85 |
+
Y : {ndarray, sparse matrix} of shape (n_samples_Y, n_features)
|
| 86 |
+
Input data.
|
| 87 |
+
|
| 88 |
+
metric : str, default='euclidean'
|
| 89 |
+
The distance metric to use.
|
| 90 |
+
For a list of available metrics, see the documentation of
|
| 91 |
+
:class:`~sklearn.metrics.DistanceMetric`.
|
| 92 |
+
|
| 93 |
+
Returns
|
| 94 |
+
-------
|
| 95 |
+
True if the dispatcher can be used, else False.
|
| 96 |
+
"""
|
| 97 |
+
|
| 98 |
+
def is_numpy_c_ordered(X):
|
| 99 |
+
return hasattr(X, "flags") and X.flags.c_contiguous
|
| 100 |
+
|
| 101 |
+
def is_valid_sparse_matrix(X):
|
| 102 |
+
return (
|
| 103 |
+
isspmatrix_csr(X)
|
| 104 |
+
and
|
| 105 |
+
# TODO: support CSR matrices without non-zeros elements
|
| 106 |
+
X.nnz > 0
|
| 107 |
+
and
|
| 108 |
+
# TODO: support CSR matrices with int64 indices and indptr
|
| 109 |
+
# See: https://github.com/scikit-learn/scikit-learn/issues/23653
|
| 110 |
+
X.indices.dtype == X.indptr.dtype == np.int32
|
| 111 |
+
)
|
| 112 |
+
|
| 113 |
+
is_usable = (
|
| 114 |
+
get_config().get("enable_cython_pairwise_dist", True)
|
| 115 |
+
and (is_numpy_c_ordered(X) or is_valid_sparse_matrix(X))
|
| 116 |
+
and (is_numpy_c_ordered(Y) or is_valid_sparse_matrix(Y))
|
| 117 |
+
and X.dtype == Y.dtype
|
| 118 |
+
and X.dtype in (np.float32, np.float64)
|
| 119 |
+
and metric in cls.valid_metrics()
|
| 120 |
+
)
|
| 121 |
+
|
| 122 |
+
# The other joblib-based back-end might be more efficient on fused sparse-dense
|
| 123 |
+
# datasets' pairs on metric="(sq)euclidean" for some configurations because it
|
| 124 |
+
# uses the Squared Euclidean matrix decomposition, i.e.:
|
| 125 |
+
#
|
| 126 |
+
# ||X_c_i - Y_c_j||² = ||X_c_i||² - 2 X_c_i.Y_c_j^T + ||Y_c_j||²
|
| 127 |
+
#
|
| 128 |
+
# calling efficient sparse-dense routines for matrix and vectors multiplication
|
| 129 |
+
# implemented in SciPy we do not use yet here.
|
| 130 |
+
# See: https://github.com/scikit-learn/scikit-learn/pull/23585#issuecomment-1247996669 # noqa
|
| 131 |
+
# TODO: implement specialisation for (sq)euclidean on fused sparse-dense
|
| 132 |
+
# using sparse-dense routines for matrix-vector multiplications.
|
| 133 |
+
# Currently, only dense-dense and sparse-sparse are optimized for
|
| 134 |
+
# the Euclidean case.
|
| 135 |
+
fused_sparse_dense_euclidean_case_guard = not (
|
| 136 |
+
(is_valid_sparse_matrix(X) ^ is_valid_sparse_matrix(Y)) # "^" is XOR
|
| 137 |
+
and isinstance(metric, str)
|
| 138 |
+
and "euclidean" in metric
|
| 139 |
+
)
|
| 140 |
+
|
| 141 |
+
return is_usable and fused_sparse_dense_euclidean_case_guard
|
| 142 |
+
|
| 143 |
+
@classmethod
|
| 144 |
+
@abstractmethod
|
| 145 |
+
def compute(
|
| 146 |
+
cls,
|
| 147 |
+
X,
|
| 148 |
+
Y,
|
| 149 |
+
**kwargs,
|
| 150 |
+
):
|
| 151 |
+
"""Compute the reduction.
|
| 152 |
+
|
| 153 |
+
Parameters
|
| 154 |
+
----------
|
| 155 |
+
X : ndarray or CSR matrix of shape (n_samples_X, n_features)
|
| 156 |
+
Input data.
|
| 157 |
+
|
| 158 |
+
Y : ndarray or CSR matrix of shape (n_samples_Y, n_features)
|
| 159 |
+
Input data.
|
| 160 |
+
|
| 161 |
+
**kwargs : additional parameters for the reduction
|
| 162 |
+
|
| 163 |
+
Notes
|
| 164 |
+
-----
|
| 165 |
+
This method is an abstract class method: it has to be implemented
|
| 166 |
+
for all subclasses.
|
| 167 |
+
"""
|
| 168 |
+
|
| 169 |
+
|
| 170 |
+
class ArgKmin(BaseDistancesReductionDispatcher):
|
| 171 |
+
"""Compute the argkmin of row vectors of X on the ones of Y.
|
| 172 |
+
|
| 173 |
+
For each row vector of X, computes the indices of k first the rows
|
| 174 |
+
vectors of Y with the smallest distances.
|
| 175 |
+
|
| 176 |
+
ArgKmin is typically used to perform
|
| 177 |
+
bruteforce k-nearest neighbors queries.
|
| 178 |
+
|
| 179 |
+
This class is not meant to be instanciated, one should only use
|
| 180 |
+
its :meth:`compute` classmethod which handles allocation and
|
| 181 |
+
deallocation consistently.
|
| 182 |
+
"""
|
| 183 |
+
|
| 184 |
+
@classmethod
|
| 185 |
+
def compute(
|
| 186 |
+
cls,
|
| 187 |
+
X,
|
| 188 |
+
Y,
|
| 189 |
+
k,
|
| 190 |
+
metric="euclidean",
|
| 191 |
+
chunk_size=None,
|
| 192 |
+
metric_kwargs=None,
|
| 193 |
+
strategy=None,
|
| 194 |
+
return_distance=False,
|
| 195 |
+
):
|
| 196 |
+
"""Compute the argkmin reduction.
|
| 197 |
+
|
| 198 |
+
Parameters
|
| 199 |
+
----------
|
| 200 |
+
X : ndarray or CSR matrix of shape (n_samples_X, n_features)
|
| 201 |
+
Input data.
|
| 202 |
+
|
| 203 |
+
Y : ndarray or CSR matrix of shape (n_samples_Y, n_features)
|
| 204 |
+
Input data.
|
| 205 |
+
|
| 206 |
+
k : int
|
| 207 |
+
The k for the argkmin reduction.
|
| 208 |
+
|
| 209 |
+
metric : str, default='euclidean'
|
| 210 |
+
The distance metric to use for argkmin.
|
| 211 |
+
For a list of available metrics, see the documentation of
|
| 212 |
+
:class:`~sklearn.metrics.DistanceMetric`.
|
| 213 |
+
|
| 214 |
+
chunk_size : int, default=None,
|
| 215 |
+
The number of vectors per chunk. If None (default) looks-up in
|
| 216 |
+
scikit-learn configuration for `pairwise_dist_chunk_size`,
|
| 217 |
+
and use 256 if it is not set.
|
| 218 |
+
|
| 219 |
+
metric_kwargs : dict, default=None
|
| 220 |
+
Keyword arguments to pass to specified metric function.
|
| 221 |
+
|
| 222 |
+
strategy : str, {'auto', 'parallel_on_X', 'parallel_on_Y'}, default=None
|
| 223 |
+
The chunking strategy defining which dataset parallelization are made on.
|
| 224 |
+
|
| 225 |
+
For both strategies the computations happens with two nested loops,
|
| 226 |
+
respectively on chunks of X and chunks of Y.
|
| 227 |
+
Strategies differs on which loop (outer or inner) is made to run
|
| 228 |
+
in parallel with the Cython `prange` construct:
|
| 229 |
+
|
| 230 |
+
- 'parallel_on_X' dispatches chunks of X uniformly on threads.
|
| 231 |
+
Each thread then iterates on all the chunks of Y. This strategy is
|
| 232 |
+
embarrassingly parallel and comes with no datastructures
|
| 233 |
+
synchronisation.
|
| 234 |
+
|
| 235 |
+
- 'parallel_on_Y' dispatches chunks of Y uniformly on threads.
|
| 236 |
+
Each thread processes all the chunks of X in turn. This strategy is
|
| 237 |
+
a sequence of embarrassingly parallel subtasks (the inner loop on Y
|
| 238 |
+
chunks) with intermediate datastructures synchronisation at each
|
| 239 |
+
iteration of the sequential outer loop on X chunks.
|
| 240 |
+
|
| 241 |
+
- 'auto' relies on a simple heuristic to choose between
|
| 242 |
+
'parallel_on_X' and 'parallel_on_Y': when `X.shape[0]` is large enough,
|
| 243 |
+
'parallel_on_X' is usually the most efficient strategy.
|
| 244 |
+
When `X.shape[0]` is small but `Y.shape[0]` is large, 'parallel_on_Y'
|
| 245 |
+
brings more opportunity for parallelism and is therefore more efficient
|
| 246 |
+
|
| 247 |
+
- None (default) looks-up in scikit-learn configuration for
|
| 248 |
+
`pairwise_dist_parallel_strategy`, and use 'auto' if it is not set.
|
| 249 |
+
|
| 250 |
+
return_distance : boolean, default=False
|
| 251 |
+
Return distances between each X vector and its
|
| 252 |
+
argkmin if set to True.
|
| 253 |
+
|
| 254 |
+
Returns
|
| 255 |
+
-------
|
| 256 |
+
If return_distance=False:
|
| 257 |
+
- argkmin_indices : ndarray of shape (n_samples_X, k)
|
| 258 |
+
Indices of the argkmin for each vector in X.
|
| 259 |
+
|
| 260 |
+
If return_distance=True:
|
| 261 |
+
- argkmin_distances : ndarray of shape (n_samples_X, k)
|
| 262 |
+
Distances to the argkmin for each vector in X.
|
| 263 |
+
- argkmin_indices : ndarray of shape (n_samples_X, k)
|
| 264 |
+
Indices of the argkmin for each vector in X.
|
| 265 |
+
|
| 266 |
+
Notes
|
| 267 |
+
-----
|
| 268 |
+
This classmethod inspects the arguments values to dispatch to the
|
| 269 |
+
dtype-specialized implementation of :class:`ArgKmin`.
|
| 270 |
+
|
| 271 |
+
This allows decoupling the API entirely from the implementation details
|
| 272 |
+
whilst maintaining RAII: all temporarily allocated datastructures necessary
|
| 273 |
+
for the concrete implementation are therefore freed when this classmethod
|
| 274 |
+
returns.
|
| 275 |
+
"""
|
| 276 |
+
if X.dtype == Y.dtype == np.float64:
|
| 277 |
+
return ArgKmin64.compute(
|
| 278 |
+
X=X,
|
| 279 |
+
Y=Y,
|
| 280 |
+
k=k,
|
| 281 |
+
metric=metric,
|
| 282 |
+
chunk_size=chunk_size,
|
| 283 |
+
metric_kwargs=metric_kwargs,
|
| 284 |
+
strategy=strategy,
|
| 285 |
+
return_distance=return_distance,
|
| 286 |
+
)
|
| 287 |
+
|
| 288 |
+
if X.dtype == Y.dtype == np.float32:
|
| 289 |
+
return ArgKmin32.compute(
|
| 290 |
+
X=X,
|
| 291 |
+
Y=Y,
|
| 292 |
+
k=k,
|
| 293 |
+
metric=metric,
|
| 294 |
+
chunk_size=chunk_size,
|
| 295 |
+
metric_kwargs=metric_kwargs,
|
| 296 |
+
strategy=strategy,
|
| 297 |
+
return_distance=return_distance,
|
| 298 |
+
)
|
| 299 |
+
|
| 300 |
+
raise ValueError(
|
| 301 |
+
"Only float64 or float32 datasets pairs are supported at this time, "
|
| 302 |
+
f"got: X.dtype={X.dtype} and Y.dtype={Y.dtype}."
|
| 303 |
+
)
|
| 304 |
+
|
| 305 |
+
|
| 306 |
+
class RadiusNeighbors(BaseDistancesReductionDispatcher):
|
| 307 |
+
"""Compute radius-based neighbors for two sets of vectors.
|
| 308 |
+
|
| 309 |
+
For each row-vector X[i] of the queries X, find all the indices j of
|
| 310 |
+
row-vectors in Y such that:
|
| 311 |
+
|
| 312 |
+
dist(X[i], Y[j]) <= radius
|
| 313 |
+
|
| 314 |
+
The distance function `dist` depends on the values of the `metric`
|
| 315 |
+
and `metric_kwargs` parameters.
|
| 316 |
+
|
| 317 |
+
This class is not meant to be instanciated, one should only use
|
| 318 |
+
its :meth:`compute` classmethod which handles allocation and
|
| 319 |
+
deallocation consistently.
|
| 320 |
+
"""
|
| 321 |
+
|
| 322 |
+
@classmethod
|
| 323 |
+
def compute(
|
| 324 |
+
cls,
|
| 325 |
+
X,
|
| 326 |
+
Y,
|
| 327 |
+
radius,
|
| 328 |
+
metric="euclidean",
|
| 329 |
+
chunk_size=None,
|
| 330 |
+
metric_kwargs=None,
|
| 331 |
+
strategy=None,
|
| 332 |
+
return_distance=False,
|
| 333 |
+
sort_results=False,
|
| 334 |
+
):
|
| 335 |
+
"""Return the results of the reduction for the given arguments.
|
| 336 |
+
|
| 337 |
+
Parameters
|
| 338 |
+
----------
|
| 339 |
+
X : ndarray or CSR matrix of shape (n_samples_X, n_features)
|
| 340 |
+
Input data.
|
| 341 |
+
|
| 342 |
+
Y : ndarray or CSR matrix of shape (n_samples_Y, n_features)
|
| 343 |
+
Input data.
|
| 344 |
+
|
| 345 |
+
radius : float
|
| 346 |
+
The radius defining the neighborhood.
|
| 347 |
+
|
| 348 |
+
metric : str, default='euclidean'
|
| 349 |
+
The distance metric to use.
|
| 350 |
+
For a list of available metrics, see the documentation of
|
| 351 |
+
:class:`~sklearn.metrics.DistanceMetric`.
|
| 352 |
+
|
| 353 |
+
chunk_size : int, default=None,
|
| 354 |
+
The number of vectors per chunk. If None (default) looks-up in
|
| 355 |
+
scikit-learn configuration for `pairwise_dist_chunk_size`,
|
| 356 |
+
and use 256 if it is not set.
|
| 357 |
+
|
| 358 |
+
metric_kwargs : dict, default=None
|
| 359 |
+
Keyword arguments to pass to specified metric function.
|
| 360 |
+
|
| 361 |
+
strategy : str, {'auto', 'parallel_on_X', 'parallel_on_Y'}, default=None
|
| 362 |
+
The chunking strategy defining which dataset parallelization are made on.
|
| 363 |
+
|
| 364 |
+
For both strategies the computations happens with two nested loops,
|
| 365 |
+
respectively on chunks of X and chunks of Y.
|
| 366 |
+
Strategies differs on which loop (outer or inner) is made to run
|
| 367 |
+
in parallel with the Cython `prange` construct:
|
| 368 |
+
|
| 369 |
+
- 'parallel_on_X' dispatches chunks of X uniformly on threads.
|
| 370 |
+
Each thread then iterates on all the chunks of Y. This strategy is
|
| 371 |
+
embarrassingly parallel and comes with no datastructures
|
| 372 |
+
synchronisation.
|
| 373 |
+
|
| 374 |
+
- 'parallel_on_Y' dispatches chunks of Y uniformly on threads.
|
| 375 |
+
Each thread processes all the chunks of X in turn. This strategy is
|
| 376 |
+
a sequence of embarrassingly parallel subtasks (the inner loop on Y
|
| 377 |
+
chunks) with intermediate datastructures synchronisation at each
|
| 378 |
+
iteration of the sequential outer loop on X chunks.
|
| 379 |
+
|
| 380 |
+
- 'auto' relies on a simple heuristic to choose between
|
| 381 |
+
'parallel_on_X' and 'parallel_on_Y': when `X.shape[0]` is large enough,
|
| 382 |
+
'parallel_on_X' is usually the most efficient strategy.
|
| 383 |
+
When `X.shape[0]` is small but `Y.shape[0]` is large, 'parallel_on_Y'
|
| 384 |
+
brings more opportunity for parallelism and is therefore more efficient
|
| 385 |
+
despite the synchronization step at each iteration of the outer loop
|
| 386 |
+
on chunks of `X`.
|
| 387 |
+
|
| 388 |
+
- None (default) looks-up in scikit-learn configuration for
|
| 389 |
+
`pairwise_dist_parallel_strategy`, and use 'auto' if it is not set.
|
| 390 |
+
|
| 391 |
+
return_distance : boolean, default=False
|
| 392 |
+
Return distances between each X vector and its neighbors if set to True.
|
| 393 |
+
|
| 394 |
+
sort_results : boolean, default=False
|
| 395 |
+
Sort results with respect to distances between each X vector and its
|
| 396 |
+
neighbors if set to True.
|
| 397 |
+
|
| 398 |
+
Returns
|
| 399 |
+
-------
|
| 400 |
+
If return_distance=False:
|
| 401 |
+
- neighbors_indices : ndarray of n_samples_X ndarray
|
| 402 |
+
Indices of the neighbors for each vector in X.
|
| 403 |
+
|
| 404 |
+
If return_distance=True:
|
| 405 |
+
- neighbors_indices : ndarray of n_samples_X ndarray
|
| 406 |
+
Indices of the neighbors for each vector in X.
|
| 407 |
+
- neighbors_distances : ndarray of n_samples_X ndarray
|
| 408 |
+
Distances to the neighbors for each vector in X.
|
| 409 |
+
|
| 410 |
+
Notes
|
| 411 |
+
-----
|
| 412 |
+
This classmethod inspects the arguments values to dispatch to the
|
| 413 |
+
dtype-specialized implementation of :class:`RadiusNeighbors`.
|
| 414 |
+
|
| 415 |
+
This allows decoupling the API entirely from the implementation details
|
| 416 |
+
whilst maintaining RAII: all temporarily allocated datastructures necessary
|
| 417 |
+
for the concrete implementation are therefore freed when this classmethod
|
| 418 |
+
returns.
|
| 419 |
+
"""
|
| 420 |
+
if X.dtype == Y.dtype == np.float64:
|
| 421 |
+
return RadiusNeighbors64.compute(
|
| 422 |
+
X=X,
|
| 423 |
+
Y=Y,
|
| 424 |
+
radius=radius,
|
| 425 |
+
metric=metric,
|
| 426 |
+
chunk_size=chunk_size,
|
| 427 |
+
metric_kwargs=metric_kwargs,
|
| 428 |
+
strategy=strategy,
|
| 429 |
+
sort_results=sort_results,
|
| 430 |
+
return_distance=return_distance,
|
| 431 |
+
)
|
| 432 |
+
|
| 433 |
+
if X.dtype == Y.dtype == np.float32:
|
| 434 |
+
return RadiusNeighbors32.compute(
|
| 435 |
+
X=X,
|
| 436 |
+
Y=Y,
|
| 437 |
+
radius=radius,
|
| 438 |
+
metric=metric,
|
| 439 |
+
chunk_size=chunk_size,
|
| 440 |
+
metric_kwargs=metric_kwargs,
|
| 441 |
+
strategy=strategy,
|
| 442 |
+
sort_results=sort_results,
|
| 443 |
+
return_distance=return_distance,
|
| 444 |
+
)
|
| 445 |
+
|
| 446 |
+
raise ValueError(
|
| 447 |
+
"Only float64 or float32 datasets pairs are supported at this time, "
|
| 448 |
+
f"got: X.dtype={X.dtype} and Y.dtype={Y.dtype}."
|
| 449 |
+
)
|
videollama2/lib/python3.10/site-packages/sklearn/metrics/_ranking.py
ADDED
|
@@ -0,0 +1,1826 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""Metrics to assess performance on classification task given scores.
|
| 2 |
+
|
| 3 |
+
Functions named as ``*_score`` return a scalar value to maximize: the higher
|
| 4 |
+
the better.
|
| 5 |
+
|
| 6 |
+
Function named as ``*_error`` or ``*_loss`` return a scalar value to minimize:
|
| 7 |
+
the lower the better.
|
| 8 |
+
"""
|
| 9 |
+
|
| 10 |
+
# Authors: Alexandre Gramfort <alexandre.gramfort@inria.fr>
|
| 11 |
+
# Mathieu Blondel <mathieu@mblondel.org>
|
| 12 |
+
# Olivier Grisel <olivier.grisel@ensta.org>
|
| 13 |
+
# Arnaud Joly <a.joly@ulg.ac.be>
|
| 14 |
+
# Jochen Wersdorfer <jochen@wersdoerfer.de>
|
| 15 |
+
# Lars Buitinck
|
| 16 |
+
# Joel Nothman <joel.nothman@gmail.com>
|
| 17 |
+
# Noel Dawe <noel@dawe.me>
|
| 18 |
+
# Michal Karbownik <michakarbownik@gmail.com>
|
| 19 |
+
# License: BSD 3 clause
|
| 20 |
+
|
| 21 |
+
|
| 22 |
+
import warnings
|
| 23 |
+
from functools import partial
|
| 24 |
+
|
| 25 |
+
import numpy as np
|
| 26 |
+
from scipy.sparse import csr_matrix, issparse
|
| 27 |
+
from scipy.stats import rankdata
|
| 28 |
+
|
| 29 |
+
from ..utils import assert_all_finite
|
| 30 |
+
from ..utils import check_consistent_length
|
| 31 |
+
from ..utils.validation import _check_sample_weight
|
| 32 |
+
from ..utils import column_or_1d, check_array
|
| 33 |
+
from ..utils.multiclass import type_of_target
|
| 34 |
+
from ..utils.extmath import stable_cumsum
|
| 35 |
+
from ..utils.sparsefuncs import count_nonzero
|
| 36 |
+
from ..exceptions import UndefinedMetricWarning
|
| 37 |
+
from ..preprocessing import label_binarize
|
| 38 |
+
from ..utils._encode import _encode, _unique
|
| 39 |
+
|
| 40 |
+
from ._base import (
|
| 41 |
+
_average_binary_score,
|
| 42 |
+
_average_multiclass_ovo_score,
|
| 43 |
+
_check_pos_label_consistency,
|
| 44 |
+
)
|
| 45 |
+
|
| 46 |
+
|
| 47 |
+
def auc(x, y):
|
| 48 |
+
"""Compute Area Under the Curve (AUC) using the trapezoidal rule.
|
| 49 |
+
|
| 50 |
+
This is a general function, given points on a curve. For computing the
|
| 51 |
+
area under the ROC-curve, see :func:`roc_auc_score`. For an alternative
|
| 52 |
+
way to summarize a precision-recall curve, see
|
| 53 |
+
:func:`average_precision_score`.
|
| 54 |
+
|
| 55 |
+
Parameters
|
| 56 |
+
----------
|
| 57 |
+
x : ndarray of shape (n,)
|
| 58 |
+
X coordinates. These must be either monotonic increasing or monotonic
|
| 59 |
+
decreasing.
|
| 60 |
+
y : ndarray of shape, (n,)
|
| 61 |
+
Y coordinates.
|
| 62 |
+
|
| 63 |
+
Returns
|
| 64 |
+
-------
|
| 65 |
+
auc : float
|
| 66 |
+
Area Under the Curve.
|
| 67 |
+
|
| 68 |
+
See Also
|
| 69 |
+
--------
|
| 70 |
+
roc_auc_score : Compute the area under the ROC curve.
|
| 71 |
+
average_precision_score : Compute average precision from prediction scores.
|
| 72 |
+
precision_recall_curve : Compute precision-recall pairs for different
|
| 73 |
+
probability thresholds.
|
| 74 |
+
|
| 75 |
+
Examples
|
| 76 |
+
--------
|
| 77 |
+
>>> import numpy as np
|
| 78 |
+
>>> from sklearn import metrics
|
| 79 |
+
>>> y = np.array([1, 1, 2, 2])
|
| 80 |
+
>>> pred = np.array([0.1, 0.4, 0.35, 0.8])
|
| 81 |
+
>>> fpr, tpr, thresholds = metrics.roc_curve(y, pred, pos_label=2)
|
| 82 |
+
>>> metrics.auc(fpr, tpr)
|
| 83 |
+
0.75
|
| 84 |
+
"""
|
| 85 |
+
check_consistent_length(x, y)
|
| 86 |
+
x = column_or_1d(x)
|
| 87 |
+
y = column_or_1d(y)
|
| 88 |
+
|
| 89 |
+
if x.shape[0] < 2:
|
| 90 |
+
raise ValueError(
|
| 91 |
+
"At least 2 points are needed to compute area under curve, but x.shape = %s"
|
| 92 |
+
% x.shape
|
| 93 |
+
)
|
| 94 |
+
|
| 95 |
+
direction = 1
|
| 96 |
+
dx = np.diff(x)
|
| 97 |
+
if np.any(dx < 0):
|
| 98 |
+
if np.all(dx <= 0):
|
| 99 |
+
direction = -1
|
| 100 |
+
else:
|
| 101 |
+
raise ValueError("x is neither increasing nor decreasing : {}.".format(x))
|
| 102 |
+
|
| 103 |
+
area = direction * np.trapz(y, x)
|
| 104 |
+
if isinstance(area, np.memmap):
|
| 105 |
+
# Reductions such as .sum used internally in np.trapz do not return a
|
| 106 |
+
# scalar by default for numpy.memmap instances contrary to
|
| 107 |
+
# regular numpy.ndarray instances.
|
| 108 |
+
area = area.dtype.type(area)
|
| 109 |
+
return area
|
| 110 |
+
|
| 111 |
+
|
| 112 |
+
def average_precision_score(
|
| 113 |
+
y_true, y_score, *, average="macro", pos_label=1, sample_weight=None
|
| 114 |
+
):
|
| 115 |
+
"""Compute average precision (AP) from prediction scores.
|
| 116 |
+
|
| 117 |
+
AP summarizes a precision-recall curve as the weighted mean of precisions
|
| 118 |
+
achieved at each threshold, with the increase in recall from the previous
|
| 119 |
+
threshold used as the weight:
|
| 120 |
+
|
| 121 |
+
.. math::
|
| 122 |
+
\\text{AP} = \\sum_n (R_n - R_{n-1}) P_n
|
| 123 |
+
|
| 124 |
+
where :math:`P_n` and :math:`R_n` are the precision and recall at the nth
|
| 125 |
+
threshold [1]_. This implementation is not interpolated and is different
|
| 126 |
+
from computing the area under the precision-recall curve with the
|
| 127 |
+
trapezoidal rule, which uses linear interpolation and can be too
|
| 128 |
+
optimistic.
|
| 129 |
+
|
| 130 |
+
Note: this implementation is restricted to the binary classification task
|
| 131 |
+
or multilabel classification task.
|
| 132 |
+
|
| 133 |
+
Read more in the :ref:`User Guide <precision_recall_f_measure_metrics>`.
|
| 134 |
+
|
| 135 |
+
Parameters
|
| 136 |
+
----------
|
| 137 |
+
y_true : ndarray of shape (n_samples,) or (n_samples, n_classes)
|
| 138 |
+
True binary labels or binary label indicators.
|
| 139 |
+
|
| 140 |
+
y_score : ndarray of shape (n_samples,) or (n_samples, n_classes)
|
| 141 |
+
Target scores, can either be probability estimates of the positive
|
| 142 |
+
class, confidence values, or non-thresholded measure of decisions
|
| 143 |
+
(as returned by :term:`decision_function` on some classifiers).
|
| 144 |
+
|
| 145 |
+
average : {'micro', 'samples', 'weighted', 'macro'} or None, \
|
| 146 |
+
default='macro'
|
| 147 |
+
If ``None``, the scores for each class are returned. Otherwise,
|
| 148 |
+
this determines the type of averaging performed on the data:
|
| 149 |
+
|
| 150 |
+
``'micro'``:
|
| 151 |
+
Calculate metrics globally by considering each element of the label
|
| 152 |
+
indicator matrix as a label.
|
| 153 |
+
``'macro'``:
|
| 154 |
+
Calculate metrics for each label, and find their unweighted
|
| 155 |
+
mean. This does not take label imbalance into account.
|
| 156 |
+
``'weighted'``:
|
| 157 |
+
Calculate metrics for each label, and find their average, weighted
|
| 158 |
+
by support (the number of true instances for each label).
|
| 159 |
+
``'samples'``:
|
| 160 |
+
Calculate metrics for each instance, and find their average.
|
| 161 |
+
|
| 162 |
+
Will be ignored when ``y_true`` is binary.
|
| 163 |
+
|
| 164 |
+
pos_label : int or str, default=1
|
| 165 |
+
The label of the positive class. Only applied to binary ``y_true``.
|
| 166 |
+
For multilabel-indicator ``y_true``, ``pos_label`` is fixed to 1.
|
| 167 |
+
|
| 168 |
+
sample_weight : array-like of shape (n_samples,), default=None
|
| 169 |
+
Sample weights.
|
| 170 |
+
|
| 171 |
+
Returns
|
| 172 |
+
-------
|
| 173 |
+
average_precision : float
|
| 174 |
+
Average precision score.
|
| 175 |
+
|
| 176 |
+
See Also
|
| 177 |
+
--------
|
| 178 |
+
roc_auc_score : Compute the area under the ROC curve.
|
| 179 |
+
precision_recall_curve : Compute precision-recall pairs for different
|
| 180 |
+
probability thresholds.
|
| 181 |
+
|
| 182 |
+
Notes
|
| 183 |
+
-----
|
| 184 |
+
.. versionchanged:: 0.19
|
| 185 |
+
Instead of linearly interpolating between operating points, precisions
|
| 186 |
+
are weighted by the change in recall since the last operating point.
|
| 187 |
+
|
| 188 |
+
References
|
| 189 |
+
----------
|
| 190 |
+
.. [1] `Wikipedia entry for the Average precision
|
| 191 |
+
<https://en.wikipedia.org/w/index.php?title=Information_retrieval&
|
| 192 |
+
oldid=793358396#Average_precision>`_
|
| 193 |
+
|
| 194 |
+
Examples
|
| 195 |
+
--------
|
| 196 |
+
>>> import numpy as np
|
| 197 |
+
>>> from sklearn.metrics import average_precision_score
|
| 198 |
+
>>> y_true = np.array([0, 0, 1, 1])
|
| 199 |
+
>>> y_scores = np.array([0.1, 0.4, 0.35, 0.8])
|
| 200 |
+
>>> average_precision_score(y_true, y_scores)
|
| 201 |
+
0.83...
|
| 202 |
+
"""
|
| 203 |
+
|
| 204 |
+
def _binary_uninterpolated_average_precision(
|
| 205 |
+
y_true, y_score, pos_label=1, sample_weight=None
|
| 206 |
+
):
|
| 207 |
+
precision, recall, _ = precision_recall_curve(
|
| 208 |
+
y_true, y_score, pos_label=pos_label, sample_weight=sample_weight
|
| 209 |
+
)
|
| 210 |
+
# Return the step function integral
|
| 211 |
+
# The following works because the last entry of precision is
|
| 212 |
+
# guaranteed to be 1, as returned by precision_recall_curve
|
| 213 |
+
return -np.sum(np.diff(recall) * np.array(precision)[:-1])
|
| 214 |
+
|
| 215 |
+
y_type = type_of_target(y_true, input_name="y_true")
|
| 216 |
+
if y_type == "multilabel-indicator" and pos_label != 1:
|
| 217 |
+
raise ValueError(
|
| 218 |
+
"Parameter pos_label is fixed to 1 for "
|
| 219 |
+
"multilabel-indicator y_true. Do not set "
|
| 220 |
+
"pos_label or set pos_label to 1."
|
| 221 |
+
)
|
| 222 |
+
elif y_type == "binary":
|
| 223 |
+
# Convert to Python primitive type to avoid NumPy type / Python str
|
| 224 |
+
# comparison. See https://github.com/numpy/numpy/issues/6784
|
| 225 |
+
present_labels = np.unique(y_true).tolist()
|
| 226 |
+
if len(present_labels) == 2 and pos_label not in present_labels:
|
| 227 |
+
raise ValueError(
|
| 228 |
+
f"pos_label={pos_label} is not a valid label. It should be "
|
| 229 |
+
f"one of {present_labels}"
|
| 230 |
+
)
|
| 231 |
+
average_precision = partial(
|
| 232 |
+
_binary_uninterpolated_average_precision, pos_label=pos_label
|
| 233 |
+
)
|
| 234 |
+
return _average_binary_score(
|
| 235 |
+
average_precision, y_true, y_score, average, sample_weight=sample_weight
|
| 236 |
+
)
|
| 237 |
+
|
| 238 |
+
|
| 239 |
+
def det_curve(y_true, y_score, pos_label=None, sample_weight=None):
|
| 240 |
+
"""Compute error rates for different probability thresholds.
|
| 241 |
+
|
| 242 |
+
.. note::
|
| 243 |
+
This metric is used for evaluation of ranking and error tradeoffs of
|
| 244 |
+
a binary classification task.
|
| 245 |
+
|
| 246 |
+
Read more in the :ref:`User Guide <det_curve>`.
|
| 247 |
+
|
| 248 |
+
.. versionadded:: 0.24
|
| 249 |
+
|
| 250 |
+
Parameters
|
| 251 |
+
----------
|
| 252 |
+
y_true : ndarray of shape (n_samples,)
|
| 253 |
+
True binary labels. If labels are not either {-1, 1} or {0, 1}, then
|
| 254 |
+
pos_label should be explicitly given.
|
| 255 |
+
|
| 256 |
+
y_score : ndarray of shape of (n_samples,)
|
| 257 |
+
Target scores, can either be probability estimates of the positive
|
| 258 |
+
class, confidence values, or non-thresholded measure of decisions
|
| 259 |
+
(as returned by "decision_function" on some classifiers).
|
| 260 |
+
|
| 261 |
+
pos_label : int or str, default=None
|
| 262 |
+
The label of the positive class.
|
| 263 |
+
When ``pos_label=None``, if `y_true` is in {-1, 1} or {0, 1},
|
| 264 |
+
``pos_label`` is set to 1, otherwise an error will be raised.
|
| 265 |
+
|
| 266 |
+
sample_weight : array-like of shape (n_samples,), default=None
|
| 267 |
+
Sample weights.
|
| 268 |
+
|
| 269 |
+
Returns
|
| 270 |
+
-------
|
| 271 |
+
fpr : ndarray of shape (n_thresholds,)
|
| 272 |
+
False positive rate (FPR) such that element i is the false positive
|
| 273 |
+
rate of predictions with score >= thresholds[i]. This is occasionally
|
| 274 |
+
referred to as false acceptance propability or fall-out.
|
| 275 |
+
|
| 276 |
+
fnr : ndarray of shape (n_thresholds,)
|
| 277 |
+
False negative rate (FNR) such that element i is the false negative
|
| 278 |
+
rate of predictions with score >= thresholds[i]. This is occasionally
|
| 279 |
+
referred to as false rejection or miss rate.
|
| 280 |
+
|
| 281 |
+
thresholds : ndarray of shape (n_thresholds,)
|
| 282 |
+
Decreasing score values.
|
| 283 |
+
|
| 284 |
+
See Also
|
| 285 |
+
--------
|
| 286 |
+
DetCurveDisplay.from_estimator : Plot DET curve given an estimator and
|
| 287 |
+
some data.
|
| 288 |
+
DetCurveDisplay.from_predictions : Plot DET curve given the true and
|
| 289 |
+
predicted labels.
|
| 290 |
+
DetCurveDisplay : DET curve visualization.
|
| 291 |
+
roc_curve : Compute Receiver operating characteristic (ROC) curve.
|
| 292 |
+
precision_recall_curve : Compute precision-recall curve.
|
| 293 |
+
|
| 294 |
+
Examples
|
| 295 |
+
--------
|
| 296 |
+
>>> import numpy as np
|
| 297 |
+
>>> from sklearn.metrics import det_curve
|
| 298 |
+
>>> y_true = np.array([0, 0, 1, 1])
|
| 299 |
+
>>> y_scores = np.array([0.1, 0.4, 0.35, 0.8])
|
| 300 |
+
>>> fpr, fnr, thresholds = det_curve(y_true, y_scores)
|
| 301 |
+
>>> fpr
|
| 302 |
+
array([0.5, 0.5, 0. ])
|
| 303 |
+
>>> fnr
|
| 304 |
+
array([0. , 0.5, 0.5])
|
| 305 |
+
>>> thresholds
|
| 306 |
+
array([0.35, 0.4 , 0.8 ])
|
| 307 |
+
"""
|
| 308 |
+
fps, tps, thresholds = _binary_clf_curve(
|
| 309 |
+
y_true, y_score, pos_label=pos_label, sample_weight=sample_weight
|
| 310 |
+
)
|
| 311 |
+
|
| 312 |
+
if len(np.unique(y_true)) != 2:
|
| 313 |
+
raise ValueError(
|
| 314 |
+
"Only one class present in y_true. Detection error "
|
| 315 |
+
"tradeoff curve is not defined in that case."
|
| 316 |
+
)
|
| 317 |
+
|
| 318 |
+
fns = tps[-1] - tps
|
| 319 |
+
p_count = tps[-1]
|
| 320 |
+
n_count = fps[-1]
|
| 321 |
+
|
| 322 |
+
# start with false positives zero
|
| 323 |
+
first_ind = (
|
| 324 |
+
fps.searchsorted(fps[0], side="right") - 1
|
| 325 |
+
if fps.searchsorted(fps[0], side="right") > 0
|
| 326 |
+
else None
|
| 327 |
+
)
|
| 328 |
+
# stop with false negatives zero
|
| 329 |
+
last_ind = tps.searchsorted(tps[-1]) + 1
|
| 330 |
+
sl = slice(first_ind, last_ind)
|
| 331 |
+
|
| 332 |
+
# reverse the output such that list of false positives is decreasing
|
| 333 |
+
return (fps[sl][::-1] / n_count, fns[sl][::-1] / p_count, thresholds[sl][::-1])
|
| 334 |
+
|
| 335 |
+
|
| 336 |
+
def _binary_roc_auc_score(y_true, y_score, sample_weight=None, max_fpr=None):
|
| 337 |
+
"""Binary roc auc score."""
|
| 338 |
+
if len(np.unique(y_true)) != 2:
|
| 339 |
+
raise ValueError(
|
| 340 |
+
"Only one class present in y_true. ROC AUC score "
|
| 341 |
+
"is not defined in that case."
|
| 342 |
+
)
|
| 343 |
+
|
| 344 |
+
fpr, tpr, _ = roc_curve(y_true, y_score, sample_weight=sample_weight)
|
| 345 |
+
if max_fpr is None or max_fpr == 1:
|
| 346 |
+
return auc(fpr, tpr)
|
| 347 |
+
if max_fpr <= 0 or max_fpr > 1:
|
| 348 |
+
raise ValueError("Expected max_fpr in range (0, 1], got: %r" % max_fpr)
|
| 349 |
+
|
| 350 |
+
# Add a single point at max_fpr by linear interpolation
|
| 351 |
+
stop = np.searchsorted(fpr, max_fpr, "right")
|
| 352 |
+
x_interp = [fpr[stop - 1], fpr[stop]]
|
| 353 |
+
y_interp = [tpr[stop - 1], tpr[stop]]
|
| 354 |
+
tpr = np.append(tpr[:stop], np.interp(max_fpr, x_interp, y_interp))
|
| 355 |
+
fpr = np.append(fpr[:stop], max_fpr)
|
| 356 |
+
partial_auc = auc(fpr, tpr)
|
| 357 |
+
|
| 358 |
+
# McClish correction: standardize result to be 0.5 if non-discriminant
|
| 359 |
+
# and 1 if maximal
|
| 360 |
+
min_area = 0.5 * max_fpr**2
|
| 361 |
+
max_area = max_fpr
|
| 362 |
+
return 0.5 * (1 + (partial_auc - min_area) / (max_area - min_area))
|
| 363 |
+
|
| 364 |
+
|
| 365 |
+
def roc_auc_score(
|
| 366 |
+
y_true,
|
| 367 |
+
y_score,
|
| 368 |
+
*,
|
| 369 |
+
average="macro",
|
| 370 |
+
sample_weight=None,
|
| 371 |
+
max_fpr=None,
|
| 372 |
+
multi_class="raise",
|
| 373 |
+
labels=None,
|
| 374 |
+
):
|
| 375 |
+
"""Compute Area Under the Receiver Operating Characteristic Curve (ROC AUC) \
|
| 376 |
+
from prediction scores.
|
| 377 |
+
|
| 378 |
+
Note: this implementation can be used with binary, multiclass and
|
| 379 |
+
multilabel classification, but some restrictions apply (see Parameters).
|
| 380 |
+
|
| 381 |
+
Read more in the :ref:`User Guide <roc_metrics>`.
|
| 382 |
+
|
| 383 |
+
Parameters
|
| 384 |
+
----------
|
| 385 |
+
y_true : array-like of shape (n_samples,) or (n_samples, n_classes)
|
| 386 |
+
True labels or binary label indicators. The binary and multiclass cases
|
| 387 |
+
expect labels with shape (n_samples,) while the multilabel case expects
|
| 388 |
+
binary label indicators with shape (n_samples, n_classes).
|
| 389 |
+
|
| 390 |
+
y_score : array-like of shape (n_samples,) or (n_samples, n_classes)
|
| 391 |
+
Target scores.
|
| 392 |
+
|
| 393 |
+
* In the binary case, it corresponds to an array of shape
|
| 394 |
+
`(n_samples,)`. Both probability estimates and non-thresholded
|
| 395 |
+
decision values can be provided. The probability estimates correspond
|
| 396 |
+
to the **probability of the class with the greater label**,
|
| 397 |
+
i.e. `estimator.classes_[1]` and thus
|
| 398 |
+
`estimator.predict_proba(X, y)[:, 1]`. The decision values
|
| 399 |
+
corresponds to the output of `estimator.decision_function(X, y)`.
|
| 400 |
+
See more information in the :ref:`User guide <roc_auc_binary>`;
|
| 401 |
+
* In the multiclass case, it corresponds to an array of shape
|
| 402 |
+
`(n_samples, n_classes)` of probability estimates provided by the
|
| 403 |
+
`predict_proba` method. The probability estimates **must**
|
| 404 |
+
sum to 1 across the possible classes. In addition, the order of the
|
| 405 |
+
class scores must correspond to the order of ``labels``,
|
| 406 |
+
if provided, or else to the numerical or lexicographical order of
|
| 407 |
+
the labels in ``y_true``. See more information in the
|
| 408 |
+
:ref:`User guide <roc_auc_multiclass>`;
|
| 409 |
+
* In the multilabel case, it corresponds to an array of shape
|
| 410 |
+
`(n_samples, n_classes)`. Probability estimates are provided by the
|
| 411 |
+
`predict_proba` method and the non-thresholded decision values by
|
| 412 |
+
the `decision_function` method. The probability estimates correspond
|
| 413 |
+
to the **probability of the class with the greater label for each
|
| 414 |
+
output** of the classifier. See more information in the
|
| 415 |
+
:ref:`User guide <roc_auc_multilabel>`.
|
| 416 |
+
|
| 417 |
+
average : {'micro', 'macro', 'samples', 'weighted'} or None, \
|
| 418 |
+
default='macro'
|
| 419 |
+
If ``None``, the scores for each class are returned.
|
| 420 |
+
Otherwise, this determines the type of averaging performed on the data.
|
| 421 |
+
Note: multiclass ROC AUC currently only handles the 'macro' and
|
| 422 |
+
'weighted' averages. For multiclass targets, `average=None` is only
|
| 423 |
+
implemented for `multi_class='ovr'` and `average='micro'` is only
|
| 424 |
+
implemented for `multi_class='ovr'`.
|
| 425 |
+
|
| 426 |
+
``'micro'``:
|
| 427 |
+
Calculate metrics globally by considering each element of the label
|
| 428 |
+
indicator matrix as a label.
|
| 429 |
+
``'macro'``:
|
| 430 |
+
Calculate metrics for each label, and find their unweighted
|
| 431 |
+
mean. This does not take label imbalance into account.
|
| 432 |
+
``'weighted'``:
|
| 433 |
+
Calculate metrics for each label, and find their average, weighted
|
| 434 |
+
by support (the number of true instances for each label).
|
| 435 |
+
``'samples'``:
|
| 436 |
+
Calculate metrics for each instance, and find their average.
|
| 437 |
+
|
| 438 |
+
Will be ignored when ``y_true`` is binary.
|
| 439 |
+
|
| 440 |
+
sample_weight : array-like of shape (n_samples,), default=None
|
| 441 |
+
Sample weights.
|
| 442 |
+
|
| 443 |
+
max_fpr : float > 0 and <= 1, default=None
|
| 444 |
+
If not ``None``, the standardized partial AUC [2]_ over the range
|
| 445 |
+
[0, max_fpr] is returned. For the multiclass case, ``max_fpr``,
|
| 446 |
+
should be either equal to ``None`` or ``1.0`` as AUC ROC partial
|
| 447 |
+
computation currently is not supported for multiclass.
|
| 448 |
+
|
| 449 |
+
multi_class : {'raise', 'ovr', 'ovo'}, default='raise'
|
| 450 |
+
Only used for multiclass targets. Determines the type of configuration
|
| 451 |
+
to use. The default value raises an error, so either
|
| 452 |
+
``'ovr'`` or ``'ovo'`` must be passed explicitly.
|
| 453 |
+
|
| 454 |
+
``'ovr'``:
|
| 455 |
+
Stands for One-vs-rest. Computes the AUC of each class
|
| 456 |
+
against the rest [3]_ [4]_. This
|
| 457 |
+
treats the multiclass case in the same way as the multilabel case.
|
| 458 |
+
Sensitive to class imbalance even when ``average == 'macro'``,
|
| 459 |
+
because class imbalance affects the composition of each of the
|
| 460 |
+
'rest' groupings.
|
| 461 |
+
``'ovo'``:
|
| 462 |
+
Stands for One-vs-one. Computes the average AUC of all
|
| 463 |
+
possible pairwise combinations of classes [5]_.
|
| 464 |
+
Insensitive to class imbalance when
|
| 465 |
+
``average == 'macro'``.
|
| 466 |
+
|
| 467 |
+
labels : array-like of shape (n_classes,), default=None
|
| 468 |
+
Only used for multiclass targets. List of labels that index the
|
| 469 |
+
classes in ``y_score``. If ``None``, the numerical or lexicographical
|
| 470 |
+
order of the labels in ``y_true`` is used.
|
| 471 |
+
|
| 472 |
+
Returns
|
| 473 |
+
-------
|
| 474 |
+
auc : float
|
| 475 |
+
Area Under the Curve score.
|
| 476 |
+
|
| 477 |
+
See Also
|
| 478 |
+
--------
|
| 479 |
+
average_precision_score : Area under the precision-recall curve.
|
| 480 |
+
roc_curve : Compute Receiver operating characteristic (ROC) curve.
|
| 481 |
+
RocCurveDisplay.from_estimator : Plot Receiver Operating Characteristic
|
| 482 |
+
(ROC) curve given an estimator and some data.
|
| 483 |
+
RocCurveDisplay.from_predictions : Plot Receiver Operating Characteristic
|
| 484 |
+
(ROC) curve given the true and predicted values.
|
| 485 |
+
|
| 486 |
+
References
|
| 487 |
+
----------
|
| 488 |
+
.. [1] `Wikipedia entry for the Receiver operating characteristic
|
| 489 |
+
<https://en.wikipedia.org/wiki/Receiver_operating_characteristic>`_
|
| 490 |
+
|
| 491 |
+
.. [2] `Analyzing a portion of the ROC curve. McClish, 1989
|
| 492 |
+
<https://www.ncbi.nlm.nih.gov/pubmed/2668680>`_
|
| 493 |
+
|
| 494 |
+
.. [3] Provost, F., Domingos, P. (2000). Well-trained PETs: Improving
|
| 495 |
+
probability estimation trees (Section 6.2), CeDER Working Paper
|
| 496 |
+
#IS-00-04, Stern School of Business, New York University.
|
| 497 |
+
|
| 498 |
+
.. [4] `Fawcett, T. (2006). An introduction to ROC analysis. Pattern
|
| 499 |
+
Recognition Letters, 27(8), 861-874.
|
| 500 |
+
<https://www.sciencedirect.com/science/article/pii/S016786550500303X>`_
|
| 501 |
+
|
| 502 |
+
.. [5] `Hand, D.J., Till, R.J. (2001). A Simple Generalisation of the Area
|
| 503 |
+
Under the ROC Curve for Multiple Class Classification Problems.
|
| 504 |
+
Machine Learning, 45(2), 171-186.
|
| 505 |
+
<http://link.springer.com/article/10.1023/A:1010920819831>`_
|
| 506 |
+
|
| 507 |
+
Examples
|
| 508 |
+
--------
|
| 509 |
+
Binary case:
|
| 510 |
+
|
| 511 |
+
>>> from sklearn.datasets import load_breast_cancer
|
| 512 |
+
>>> from sklearn.linear_model import LogisticRegression
|
| 513 |
+
>>> from sklearn.metrics import roc_auc_score
|
| 514 |
+
>>> X, y = load_breast_cancer(return_X_y=True)
|
| 515 |
+
>>> clf = LogisticRegression(solver="liblinear", random_state=0).fit(X, y)
|
| 516 |
+
>>> roc_auc_score(y, clf.predict_proba(X)[:, 1])
|
| 517 |
+
0.99...
|
| 518 |
+
>>> roc_auc_score(y, clf.decision_function(X))
|
| 519 |
+
0.99...
|
| 520 |
+
|
| 521 |
+
Multiclass case:
|
| 522 |
+
|
| 523 |
+
>>> from sklearn.datasets import load_iris
|
| 524 |
+
>>> X, y = load_iris(return_X_y=True)
|
| 525 |
+
>>> clf = LogisticRegression(solver="liblinear").fit(X, y)
|
| 526 |
+
>>> roc_auc_score(y, clf.predict_proba(X), multi_class='ovr')
|
| 527 |
+
0.99...
|
| 528 |
+
|
| 529 |
+
Multilabel case:
|
| 530 |
+
|
| 531 |
+
>>> import numpy as np
|
| 532 |
+
>>> from sklearn.datasets import make_multilabel_classification
|
| 533 |
+
>>> from sklearn.multioutput import MultiOutputClassifier
|
| 534 |
+
>>> X, y = make_multilabel_classification(random_state=0)
|
| 535 |
+
>>> clf = MultiOutputClassifier(clf).fit(X, y)
|
| 536 |
+
>>> # get a list of n_output containing probability arrays of shape
|
| 537 |
+
>>> # (n_samples, n_classes)
|
| 538 |
+
>>> y_pred = clf.predict_proba(X)
|
| 539 |
+
>>> # extract the positive columns for each output
|
| 540 |
+
>>> y_pred = np.transpose([pred[:, 1] for pred in y_pred])
|
| 541 |
+
>>> roc_auc_score(y, y_pred, average=None)
|
| 542 |
+
array([0.82..., 0.86..., 0.94..., 0.85... , 0.94...])
|
| 543 |
+
>>> from sklearn.linear_model import RidgeClassifierCV
|
| 544 |
+
>>> clf = RidgeClassifierCV().fit(X, y)
|
| 545 |
+
>>> roc_auc_score(y, clf.decision_function(X), average=None)
|
| 546 |
+
array([0.81..., 0.84... , 0.93..., 0.87..., 0.94...])
|
| 547 |
+
"""
|
| 548 |
+
|
| 549 |
+
y_type = type_of_target(y_true, input_name="y_true")
|
| 550 |
+
y_true = check_array(y_true, ensure_2d=False, dtype=None)
|
| 551 |
+
y_score = check_array(y_score, ensure_2d=False)
|
| 552 |
+
|
| 553 |
+
if y_type == "multiclass" or (
|
| 554 |
+
y_type == "binary" and y_score.ndim == 2 and y_score.shape[1] > 2
|
| 555 |
+
):
|
| 556 |
+
# do not support partial ROC computation for multiclass
|
| 557 |
+
if max_fpr is not None and max_fpr != 1.0:
|
| 558 |
+
raise ValueError(
|
| 559 |
+
"Partial AUC computation not available in "
|
| 560 |
+
"multiclass setting, 'max_fpr' must be"
|
| 561 |
+
" set to `None`, received `max_fpr={0}` "
|
| 562 |
+
"instead".format(max_fpr)
|
| 563 |
+
)
|
| 564 |
+
if multi_class == "raise":
|
| 565 |
+
raise ValueError("multi_class must be in ('ovo', 'ovr')")
|
| 566 |
+
return _multiclass_roc_auc_score(
|
| 567 |
+
y_true, y_score, labels, multi_class, average, sample_weight
|
| 568 |
+
)
|
| 569 |
+
elif y_type == "binary":
|
| 570 |
+
labels = np.unique(y_true)
|
| 571 |
+
y_true = label_binarize(y_true, classes=labels)[:, 0]
|
| 572 |
+
return _average_binary_score(
|
| 573 |
+
partial(_binary_roc_auc_score, max_fpr=max_fpr),
|
| 574 |
+
y_true,
|
| 575 |
+
y_score,
|
| 576 |
+
average,
|
| 577 |
+
sample_weight=sample_weight,
|
| 578 |
+
)
|
| 579 |
+
else: # multilabel-indicator
|
| 580 |
+
return _average_binary_score(
|
| 581 |
+
partial(_binary_roc_auc_score, max_fpr=max_fpr),
|
| 582 |
+
y_true,
|
| 583 |
+
y_score,
|
| 584 |
+
average,
|
| 585 |
+
sample_weight=sample_weight,
|
| 586 |
+
)
|
| 587 |
+
|
| 588 |
+
|
| 589 |
+
def _multiclass_roc_auc_score(
|
| 590 |
+
y_true, y_score, labels, multi_class, average, sample_weight
|
| 591 |
+
):
|
| 592 |
+
"""Multiclass roc auc score.
|
| 593 |
+
|
| 594 |
+
Parameters
|
| 595 |
+
----------
|
| 596 |
+
y_true : array-like of shape (n_samples,)
|
| 597 |
+
True multiclass labels.
|
| 598 |
+
|
| 599 |
+
y_score : array-like of shape (n_samples, n_classes)
|
| 600 |
+
Target scores corresponding to probability estimates of a sample
|
| 601 |
+
belonging to a particular class
|
| 602 |
+
|
| 603 |
+
labels : array-like of shape (n_classes,) or None
|
| 604 |
+
List of labels to index ``y_score`` used for multiclass. If ``None``,
|
| 605 |
+
the lexical order of ``y_true`` is used to index ``y_score``.
|
| 606 |
+
|
| 607 |
+
multi_class : {'ovr', 'ovo'}
|
| 608 |
+
Determines the type of multiclass configuration to use.
|
| 609 |
+
``'ovr'``:
|
| 610 |
+
Calculate metrics for the multiclass case using the one-vs-rest
|
| 611 |
+
approach.
|
| 612 |
+
``'ovo'``:
|
| 613 |
+
Calculate metrics for the multiclass case using the one-vs-one
|
| 614 |
+
approach.
|
| 615 |
+
|
| 616 |
+
average : {'micro', 'macro', 'weighted'}
|
| 617 |
+
Determines the type of averaging performed on the pairwise binary
|
| 618 |
+
metric scores
|
| 619 |
+
``'micro'``:
|
| 620 |
+
Calculate metrics for the binarized-raveled classes. Only supported
|
| 621 |
+
for `multi_class='ovr'`.
|
| 622 |
+
|
| 623 |
+
.. versionadded:: 1.2
|
| 624 |
+
|
| 625 |
+
``'macro'``:
|
| 626 |
+
Calculate metrics for each label, and find their unweighted
|
| 627 |
+
mean. This does not take label imbalance into account. Classes
|
| 628 |
+
are assumed to be uniformly distributed.
|
| 629 |
+
``'weighted'``:
|
| 630 |
+
Calculate metrics for each label, taking into account the
|
| 631 |
+
prevalence of the classes.
|
| 632 |
+
|
| 633 |
+
sample_weight : array-like of shape (n_samples,) or None
|
| 634 |
+
Sample weights.
|
| 635 |
+
|
| 636 |
+
"""
|
| 637 |
+
# validation of the input y_score
|
| 638 |
+
if not np.allclose(1, y_score.sum(axis=1)):
|
| 639 |
+
raise ValueError(
|
| 640 |
+
"Target scores need to be probabilities for multiclass "
|
| 641 |
+
"roc_auc, i.e. they should sum up to 1.0 over classes"
|
| 642 |
+
)
|
| 643 |
+
|
| 644 |
+
# validation for multiclass parameter specifications
|
| 645 |
+
average_options = ("macro", "weighted", None)
|
| 646 |
+
if multi_class == "ovr":
|
| 647 |
+
average_options = ("micro",) + average_options
|
| 648 |
+
if average not in average_options:
|
| 649 |
+
raise ValueError(
|
| 650 |
+
"average must be one of {0} for multiclass problems".format(average_options)
|
| 651 |
+
)
|
| 652 |
+
|
| 653 |
+
multiclass_options = ("ovo", "ovr")
|
| 654 |
+
if multi_class not in multiclass_options:
|
| 655 |
+
raise ValueError(
|
| 656 |
+
"multi_class='{0}' is not supported "
|
| 657 |
+
"for multiclass ROC AUC, multi_class must be "
|
| 658 |
+
"in {1}".format(multi_class, multiclass_options)
|
| 659 |
+
)
|
| 660 |
+
|
| 661 |
+
if average is None and multi_class == "ovo":
|
| 662 |
+
raise NotImplementedError(
|
| 663 |
+
"average=None is not implemented for multi_class='ovo'."
|
| 664 |
+
)
|
| 665 |
+
|
| 666 |
+
if labels is not None:
|
| 667 |
+
labels = column_or_1d(labels)
|
| 668 |
+
classes = _unique(labels)
|
| 669 |
+
if len(classes) != len(labels):
|
| 670 |
+
raise ValueError("Parameter 'labels' must be unique")
|
| 671 |
+
if not np.array_equal(classes, labels):
|
| 672 |
+
raise ValueError("Parameter 'labels' must be ordered")
|
| 673 |
+
if len(classes) != y_score.shape[1]:
|
| 674 |
+
raise ValueError(
|
| 675 |
+
"Number of given labels, {0}, not equal to the number "
|
| 676 |
+
"of columns in 'y_score', {1}".format(len(classes), y_score.shape[1])
|
| 677 |
+
)
|
| 678 |
+
if len(np.setdiff1d(y_true, classes)):
|
| 679 |
+
raise ValueError("'y_true' contains labels not in parameter 'labels'")
|
| 680 |
+
else:
|
| 681 |
+
classes = _unique(y_true)
|
| 682 |
+
if len(classes) != y_score.shape[1]:
|
| 683 |
+
raise ValueError(
|
| 684 |
+
"Number of classes in y_true not equal to the number of "
|
| 685 |
+
"columns in 'y_score'"
|
| 686 |
+
)
|
| 687 |
+
|
| 688 |
+
if multi_class == "ovo":
|
| 689 |
+
if sample_weight is not None:
|
| 690 |
+
raise ValueError(
|
| 691 |
+
"sample_weight is not supported "
|
| 692 |
+
"for multiclass one-vs-one ROC AUC, "
|
| 693 |
+
"'sample_weight' must be None in this case."
|
| 694 |
+
)
|
| 695 |
+
y_true_encoded = _encode(y_true, uniques=classes)
|
| 696 |
+
# Hand & Till (2001) implementation (ovo)
|
| 697 |
+
return _average_multiclass_ovo_score(
|
| 698 |
+
_binary_roc_auc_score, y_true_encoded, y_score, average=average
|
| 699 |
+
)
|
| 700 |
+
else:
|
| 701 |
+
# ovr is same as multi-label
|
| 702 |
+
y_true_multilabel = label_binarize(y_true, classes=classes)
|
| 703 |
+
return _average_binary_score(
|
| 704 |
+
_binary_roc_auc_score,
|
| 705 |
+
y_true_multilabel,
|
| 706 |
+
y_score,
|
| 707 |
+
average,
|
| 708 |
+
sample_weight=sample_weight,
|
| 709 |
+
)
|
| 710 |
+
|
| 711 |
+
|
| 712 |
+
def _binary_clf_curve(y_true, y_score, pos_label=None, sample_weight=None):
|
| 713 |
+
"""Calculate true and false positives per binary classification threshold.
|
| 714 |
+
|
| 715 |
+
Parameters
|
| 716 |
+
----------
|
| 717 |
+
y_true : ndarray of shape (n_samples,)
|
| 718 |
+
True targets of binary classification.
|
| 719 |
+
|
| 720 |
+
y_score : ndarray of shape (n_samples,)
|
| 721 |
+
Estimated probabilities or output of a decision function.
|
| 722 |
+
|
| 723 |
+
pos_label : int or str, default=None
|
| 724 |
+
The label of the positive class.
|
| 725 |
+
|
| 726 |
+
sample_weight : array-like of shape (n_samples,), default=None
|
| 727 |
+
Sample weights.
|
| 728 |
+
|
| 729 |
+
Returns
|
| 730 |
+
-------
|
| 731 |
+
fps : ndarray of shape (n_thresholds,)
|
| 732 |
+
A count of false positives, at index i being the number of negative
|
| 733 |
+
samples assigned a score >= thresholds[i]. The total number of
|
| 734 |
+
negative samples is equal to fps[-1] (thus true negatives are given by
|
| 735 |
+
fps[-1] - fps).
|
| 736 |
+
|
| 737 |
+
tps : ndarray of shape (n_thresholds,)
|
| 738 |
+
An increasing count of true positives, at index i being the number
|
| 739 |
+
of positive samples assigned a score >= thresholds[i]. The total
|
| 740 |
+
number of positive samples is equal to tps[-1] (thus false negatives
|
| 741 |
+
are given by tps[-1] - tps).
|
| 742 |
+
|
| 743 |
+
thresholds : ndarray of shape (n_thresholds,)
|
| 744 |
+
Decreasing score values.
|
| 745 |
+
"""
|
| 746 |
+
# Check to make sure y_true is valid
|
| 747 |
+
y_type = type_of_target(y_true, input_name="y_true")
|
| 748 |
+
if not (y_type == "binary" or (y_type == "multiclass" and pos_label is not None)):
|
| 749 |
+
raise ValueError("{0} format is not supported".format(y_type))
|
| 750 |
+
|
| 751 |
+
check_consistent_length(y_true, y_score, sample_weight)
|
| 752 |
+
y_true = column_or_1d(y_true)
|
| 753 |
+
y_score = column_or_1d(y_score)
|
| 754 |
+
assert_all_finite(y_true)
|
| 755 |
+
assert_all_finite(y_score)
|
| 756 |
+
|
| 757 |
+
# Filter out zero-weighted samples, as they should not impact the result
|
| 758 |
+
if sample_weight is not None:
|
| 759 |
+
sample_weight = column_or_1d(sample_weight)
|
| 760 |
+
sample_weight = _check_sample_weight(sample_weight, y_true)
|
| 761 |
+
nonzero_weight_mask = sample_weight != 0
|
| 762 |
+
y_true = y_true[nonzero_weight_mask]
|
| 763 |
+
y_score = y_score[nonzero_weight_mask]
|
| 764 |
+
sample_weight = sample_weight[nonzero_weight_mask]
|
| 765 |
+
|
| 766 |
+
pos_label = _check_pos_label_consistency(pos_label, y_true)
|
| 767 |
+
|
| 768 |
+
# make y_true a boolean vector
|
| 769 |
+
y_true = y_true == pos_label
|
| 770 |
+
|
| 771 |
+
# sort scores and corresponding truth values
|
| 772 |
+
desc_score_indices = np.argsort(y_score, kind="mergesort")[::-1]
|
| 773 |
+
y_score = y_score[desc_score_indices]
|
| 774 |
+
y_true = y_true[desc_score_indices]
|
| 775 |
+
if sample_weight is not None:
|
| 776 |
+
weight = sample_weight[desc_score_indices]
|
| 777 |
+
else:
|
| 778 |
+
weight = 1.0
|
| 779 |
+
|
| 780 |
+
# y_score typically has many tied values. Here we extract
|
| 781 |
+
# the indices associated with the distinct values. We also
|
| 782 |
+
# concatenate a value for the end of the curve.
|
| 783 |
+
distinct_value_indices = np.where(np.diff(y_score))[0]
|
| 784 |
+
threshold_idxs = np.r_[distinct_value_indices, y_true.size - 1]
|
| 785 |
+
|
| 786 |
+
# accumulate the true positives with decreasing threshold
|
| 787 |
+
tps = stable_cumsum(y_true * weight)[threshold_idxs]
|
| 788 |
+
if sample_weight is not None:
|
| 789 |
+
# express fps as a cumsum to ensure fps is increasing even in
|
| 790 |
+
# the presence of floating point errors
|
| 791 |
+
fps = stable_cumsum((1 - y_true) * weight)[threshold_idxs]
|
| 792 |
+
else:
|
| 793 |
+
fps = 1 + threshold_idxs - tps
|
| 794 |
+
return fps, tps, y_score[threshold_idxs]
|
| 795 |
+
|
| 796 |
+
|
| 797 |
+
def precision_recall_curve(y_true, probas_pred, *, pos_label=None, sample_weight=None):
|
| 798 |
+
"""Compute precision-recall pairs for different probability thresholds.
|
| 799 |
+
|
| 800 |
+
Note: this implementation is restricted to the binary classification task.
|
| 801 |
+
|
| 802 |
+
The precision is the ratio ``tp / (tp + fp)`` where ``tp`` is the number of
|
| 803 |
+
true positives and ``fp`` the number of false positives. The precision is
|
| 804 |
+
intuitively the ability of the classifier not to label as positive a sample
|
| 805 |
+
that is negative.
|
| 806 |
+
|
| 807 |
+
The recall is the ratio ``tp / (tp + fn)`` where ``tp`` is the number of
|
| 808 |
+
true positives and ``fn`` the number of false negatives. The recall is
|
| 809 |
+
intuitively the ability of the classifier to find all the positive samples.
|
| 810 |
+
|
| 811 |
+
The last precision and recall values are 1. and 0. respectively and do not
|
| 812 |
+
have a corresponding threshold. This ensures that the graph starts on the
|
| 813 |
+
y axis.
|
| 814 |
+
|
| 815 |
+
The first precision and recall values are precision=class balance and recall=1.0
|
| 816 |
+
which corresponds to a classifier that always predicts the positive class.
|
| 817 |
+
|
| 818 |
+
Read more in the :ref:`User Guide <precision_recall_f_measure_metrics>`.
|
| 819 |
+
|
| 820 |
+
Parameters
|
| 821 |
+
----------
|
| 822 |
+
y_true : ndarray of shape (n_samples,)
|
| 823 |
+
True binary labels. If labels are not either {-1, 1} or {0, 1}, then
|
| 824 |
+
pos_label should be explicitly given.
|
| 825 |
+
|
| 826 |
+
probas_pred : ndarray of shape (n_samples,)
|
| 827 |
+
Target scores, can either be probability estimates of the positive
|
| 828 |
+
class, or non-thresholded measure of decisions (as returned by
|
| 829 |
+
`decision_function` on some classifiers).
|
| 830 |
+
|
| 831 |
+
pos_label : int or str, default=None
|
| 832 |
+
The label of the positive class.
|
| 833 |
+
When ``pos_label=None``, if y_true is in {-1, 1} or {0, 1},
|
| 834 |
+
``pos_label`` is set to 1, otherwise an error will be raised.
|
| 835 |
+
|
| 836 |
+
sample_weight : array-like of shape (n_samples,), default=None
|
| 837 |
+
Sample weights.
|
| 838 |
+
|
| 839 |
+
Returns
|
| 840 |
+
-------
|
| 841 |
+
precision : ndarray of shape (n_thresholds + 1,)
|
| 842 |
+
Precision values such that element i is the precision of
|
| 843 |
+
predictions with score >= thresholds[i] and the last element is 1.
|
| 844 |
+
|
| 845 |
+
recall : ndarray of shape (n_thresholds + 1,)
|
| 846 |
+
Decreasing recall values such that element i is the recall of
|
| 847 |
+
predictions with score >= thresholds[i] and the last element is 0.
|
| 848 |
+
|
| 849 |
+
thresholds : ndarray of shape (n_thresholds,)
|
| 850 |
+
Increasing thresholds on the decision function used to compute
|
| 851 |
+
precision and recall where `n_thresholds = len(np.unique(probas_pred))`.
|
| 852 |
+
|
| 853 |
+
See Also
|
| 854 |
+
--------
|
| 855 |
+
PrecisionRecallDisplay.from_estimator : Plot Precision Recall Curve given
|
| 856 |
+
a binary classifier.
|
| 857 |
+
PrecisionRecallDisplay.from_predictions : Plot Precision Recall Curve
|
| 858 |
+
using predictions from a binary classifier.
|
| 859 |
+
average_precision_score : Compute average precision from prediction scores.
|
| 860 |
+
det_curve: Compute error rates for different probability thresholds.
|
| 861 |
+
roc_curve : Compute Receiver operating characteristic (ROC) curve.
|
| 862 |
+
|
| 863 |
+
Examples
|
| 864 |
+
--------
|
| 865 |
+
>>> import numpy as np
|
| 866 |
+
>>> from sklearn.metrics import precision_recall_curve
|
| 867 |
+
>>> y_true = np.array([0, 0, 1, 1])
|
| 868 |
+
>>> y_scores = np.array([0.1, 0.4, 0.35, 0.8])
|
| 869 |
+
>>> precision, recall, thresholds = precision_recall_curve(
|
| 870 |
+
... y_true, y_scores)
|
| 871 |
+
>>> precision
|
| 872 |
+
array([0.5 , 0.66666667, 0.5 , 1. , 1. ])
|
| 873 |
+
>>> recall
|
| 874 |
+
array([1. , 1. , 0.5, 0.5, 0. ])
|
| 875 |
+
>>> thresholds
|
| 876 |
+
array([0.1 , 0.35, 0.4 , 0.8 ])
|
| 877 |
+
"""
|
| 878 |
+
fps, tps, thresholds = _binary_clf_curve(
|
| 879 |
+
y_true, probas_pred, pos_label=pos_label, sample_weight=sample_weight
|
| 880 |
+
)
|
| 881 |
+
|
| 882 |
+
ps = tps + fps
|
| 883 |
+
# Initialize the result array with zeros to make sure that precision[ps == 0]
|
| 884 |
+
# does not contain uninitialized values.
|
| 885 |
+
precision = np.zeros_like(tps)
|
| 886 |
+
np.divide(tps, ps, out=precision, where=(ps != 0))
|
| 887 |
+
|
| 888 |
+
# When no positive label in y_true, recall is set to 1 for all thresholds
|
| 889 |
+
# tps[-1] == 0 <=> y_true == all negative labels
|
| 890 |
+
if tps[-1] == 0:
|
| 891 |
+
warnings.warn(
|
| 892 |
+
"No positive class found in y_true, "
|
| 893 |
+
"recall is set to one for all thresholds."
|
| 894 |
+
)
|
| 895 |
+
recall = np.ones_like(tps)
|
| 896 |
+
else:
|
| 897 |
+
recall = tps / tps[-1]
|
| 898 |
+
|
| 899 |
+
# reverse the outputs so recall is decreasing
|
| 900 |
+
sl = slice(None, None, -1)
|
| 901 |
+
return np.hstack((precision[sl], 1)), np.hstack((recall[sl], 0)), thresholds[sl]
|
| 902 |
+
|
| 903 |
+
|
| 904 |
+
def roc_curve(
|
| 905 |
+
y_true, y_score, *, pos_label=None, sample_weight=None, drop_intermediate=True
|
| 906 |
+
):
|
| 907 |
+
"""Compute Receiver operating characteristic (ROC).
|
| 908 |
+
|
| 909 |
+
Note: this implementation is restricted to the binary classification task.
|
| 910 |
+
|
| 911 |
+
Read more in the :ref:`User Guide <roc_metrics>`.
|
| 912 |
+
|
| 913 |
+
Parameters
|
| 914 |
+
----------
|
| 915 |
+
y_true : ndarray of shape (n_samples,)
|
| 916 |
+
True binary labels. If labels are not either {-1, 1} or {0, 1}, then
|
| 917 |
+
pos_label should be explicitly given.
|
| 918 |
+
|
| 919 |
+
y_score : ndarray of shape (n_samples,)
|
| 920 |
+
Target scores, can either be probability estimates of the positive
|
| 921 |
+
class, confidence values, or non-thresholded measure of decisions
|
| 922 |
+
(as returned by "decision_function" on some classifiers).
|
| 923 |
+
|
| 924 |
+
pos_label : int or str, default=None
|
| 925 |
+
The label of the positive class.
|
| 926 |
+
When ``pos_label=None``, if `y_true` is in {-1, 1} or {0, 1},
|
| 927 |
+
``pos_label`` is set to 1, otherwise an error will be raised.
|
| 928 |
+
|
| 929 |
+
sample_weight : array-like of shape (n_samples,), default=None
|
| 930 |
+
Sample weights.
|
| 931 |
+
|
| 932 |
+
drop_intermediate : bool, default=True
|
| 933 |
+
Whether to drop some suboptimal thresholds which would not appear
|
| 934 |
+
on a plotted ROC curve. This is useful in order to create lighter
|
| 935 |
+
ROC curves.
|
| 936 |
+
|
| 937 |
+
.. versionadded:: 0.17
|
| 938 |
+
parameter *drop_intermediate*.
|
| 939 |
+
|
| 940 |
+
Returns
|
| 941 |
+
-------
|
| 942 |
+
fpr : ndarray of shape (>2,)
|
| 943 |
+
Increasing false positive rates such that element i is the false
|
| 944 |
+
positive rate of predictions with score >= `thresholds[i]`.
|
| 945 |
+
|
| 946 |
+
tpr : ndarray of shape (>2,)
|
| 947 |
+
Increasing true positive rates such that element `i` is the true
|
| 948 |
+
positive rate of predictions with score >= `thresholds[i]`.
|
| 949 |
+
|
| 950 |
+
thresholds : ndarray of shape = (n_thresholds,)
|
| 951 |
+
Decreasing thresholds on the decision function used to compute
|
| 952 |
+
fpr and tpr. `thresholds[0]` represents no instances being predicted
|
| 953 |
+
and is arbitrarily set to `max(y_score) + 1`.
|
| 954 |
+
|
| 955 |
+
See Also
|
| 956 |
+
--------
|
| 957 |
+
RocCurveDisplay.from_estimator : Plot Receiver Operating Characteristic
|
| 958 |
+
(ROC) curve given an estimator and some data.
|
| 959 |
+
RocCurveDisplay.from_predictions : Plot Receiver Operating Characteristic
|
| 960 |
+
(ROC) curve given the true and predicted values.
|
| 961 |
+
det_curve: Compute error rates for different probability thresholds.
|
| 962 |
+
roc_auc_score : Compute the area under the ROC curve.
|
| 963 |
+
|
| 964 |
+
Notes
|
| 965 |
+
-----
|
| 966 |
+
Since the thresholds are sorted from low to high values, they
|
| 967 |
+
are reversed upon returning them to ensure they correspond to both ``fpr``
|
| 968 |
+
and ``tpr``, which are sorted in reversed order during their calculation.
|
| 969 |
+
|
| 970 |
+
References
|
| 971 |
+
----------
|
| 972 |
+
.. [1] `Wikipedia entry for the Receiver operating characteristic
|
| 973 |
+
<https://en.wikipedia.org/wiki/Receiver_operating_characteristic>`_
|
| 974 |
+
|
| 975 |
+
.. [2] Fawcett T. An introduction to ROC analysis[J]. Pattern Recognition
|
| 976 |
+
Letters, 2006, 27(8):861-874.
|
| 977 |
+
|
| 978 |
+
Examples
|
| 979 |
+
--------
|
| 980 |
+
>>> import numpy as np
|
| 981 |
+
>>> from sklearn import metrics
|
| 982 |
+
>>> y = np.array([1, 1, 2, 2])
|
| 983 |
+
>>> scores = np.array([0.1, 0.4, 0.35, 0.8])
|
| 984 |
+
>>> fpr, tpr, thresholds = metrics.roc_curve(y, scores, pos_label=2)
|
| 985 |
+
>>> fpr
|
| 986 |
+
array([0. , 0. , 0.5, 0.5, 1. ])
|
| 987 |
+
>>> tpr
|
| 988 |
+
array([0. , 0.5, 0.5, 1. , 1. ])
|
| 989 |
+
>>> thresholds
|
| 990 |
+
array([1.8 , 0.8 , 0.4 , 0.35, 0.1 ])
|
| 991 |
+
"""
|
| 992 |
+
fps, tps, thresholds = _binary_clf_curve(
|
| 993 |
+
y_true, y_score, pos_label=pos_label, sample_weight=sample_weight
|
| 994 |
+
)
|
| 995 |
+
|
| 996 |
+
# Attempt to drop thresholds corresponding to points in between and
|
| 997 |
+
# collinear with other points. These are always suboptimal and do not
|
| 998 |
+
# appear on a plotted ROC curve (and thus do not affect the AUC).
|
| 999 |
+
# Here np.diff(_, 2) is used as a "second derivative" to tell if there
|
| 1000 |
+
# is a corner at the point. Both fps and tps must be tested to handle
|
| 1001 |
+
# thresholds with multiple data points (which are combined in
|
| 1002 |
+
# _binary_clf_curve). This keeps all cases where the point should be kept,
|
| 1003 |
+
# but does not drop more complicated cases like fps = [1, 3, 7],
|
| 1004 |
+
# tps = [1, 2, 4]; there is no harm in keeping too many thresholds.
|
| 1005 |
+
if drop_intermediate and len(fps) > 2:
|
| 1006 |
+
optimal_idxs = np.where(
|
| 1007 |
+
np.r_[True, np.logical_or(np.diff(fps, 2), np.diff(tps, 2)), True]
|
| 1008 |
+
)[0]
|
| 1009 |
+
fps = fps[optimal_idxs]
|
| 1010 |
+
tps = tps[optimal_idxs]
|
| 1011 |
+
thresholds = thresholds[optimal_idxs]
|
| 1012 |
+
|
| 1013 |
+
# Add an extra threshold position
|
| 1014 |
+
# to make sure that the curve starts at (0, 0)
|
| 1015 |
+
tps = np.r_[0, tps]
|
| 1016 |
+
fps = np.r_[0, fps]
|
| 1017 |
+
thresholds = np.r_[thresholds[0] + 1, thresholds]
|
| 1018 |
+
|
| 1019 |
+
if fps[-1] <= 0:
|
| 1020 |
+
warnings.warn(
|
| 1021 |
+
"No negative samples in y_true, false positive value should be meaningless",
|
| 1022 |
+
UndefinedMetricWarning,
|
| 1023 |
+
)
|
| 1024 |
+
fpr = np.repeat(np.nan, fps.shape)
|
| 1025 |
+
else:
|
| 1026 |
+
fpr = fps / fps[-1]
|
| 1027 |
+
|
| 1028 |
+
if tps[-1] <= 0:
|
| 1029 |
+
warnings.warn(
|
| 1030 |
+
"No positive samples in y_true, true positive value should be meaningless",
|
| 1031 |
+
UndefinedMetricWarning,
|
| 1032 |
+
)
|
| 1033 |
+
tpr = np.repeat(np.nan, tps.shape)
|
| 1034 |
+
else:
|
| 1035 |
+
tpr = tps / tps[-1]
|
| 1036 |
+
|
| 1037 |
+
return fpr, tpr, thresholds
|
| 1038 |
+
|
| 1039 |
+
|
| 1040 |
+
def label_ranking_average_precision_score(y_true, y_score, *, sample_weight=None):
|
| 1041 |
+
"""Compute ranking-based average precision.
|
| 1042 |
+
|
| 1043 |
+
Label ranking average precision (LRAP) is the average over each ground
|
| 1044 |
+
truth label assigned to each sample, of the ratio of true vs. total
|
| 1045 |
+
labels with lower score.
|
| 1046 |
+
|
| 1047 |
+
This metric is used in multilabel ranking problem, where the goal
|
| 1048 |
+
is to give better rank to the labels associated to each sample.
|
| 1049 |
+
|
| 1050 |
+
The obtained score is always strictly greater than 0 and
|
| 1051 |
+
the best value is 1.
|
| 1052 |
+
|
| 1053 |
+
Read more in the :ref:`User Guide <label_ranking_average_precision>`.
|
| 1054 |
+
|
| 1055 |
+
Parameters
|
| 1056 |
+
----------
|
| 1057 |
+
y_true : {ndarray, sparse matrix} of shape (n_samples, n_labels)
|
| 1058 |
+
True binary labels in binary indicator format.
|
| 1059 |
+
|
| 1060 |
+
y_score : ndarray of shape (n_samples, n_labels)
|
| 1061 |
+
Target scores, can either be probability estimates of the positive
|
| 1062 |
+
class, confidence values, or non-thresholded measure of decisions
|
| 1063 |
+
(as returned by "decision_function" on some classifiers).
|
| 1064 |
+
|
| 1065 |
+
sample_weight : array-like of shape (n_samples,), default=None
|
| 1066 |
+
Sample weights.
|
| 1067 |
+
|
| 1068 |
+
.. versionadded:: 0.20
|
| 1069 |
+
|
| 1070 |
+
Returns
|
| 1071 |
+
-------
|
| 1072 |
+
score : float
|
| 1073 |
+
Ranking-based average precision score.
|
| 1074 |
+
|
| 1075 |
+
Examples
|
| 1076 |
+
--------
|
| 1077 |
+
>>> import numpy as np
|
| 1078 |
+
>>> from sklearn.metrics import label_ranking_average_precision_score
|
| 1079 |
+
>>> y_true = np.array([[1, 0, 0], [0, 0, 1]])
|
| 1080 |
+
>>> y_score = np.array([[0.75, 0.5, 1], [1, 0.2, 0.1]])
|
| 1081 |
+
>>> label_ranking_average_precision_score(y_true, y_score)
|
| 1082 |
+
0.416...
|
| 1083 |
+
"""
|
| 1084 |
+
check_consistent_length(y_true, y_score, sample_weight)
|
| 1085 |
+
y_true = check_array(y_true, ensure_2d=False, accept_sparse="csr")
|
| 1086 |
+
y_score = check_array(y_score, ensure_2d=False)
|
| 1087 |
+
|
| 1088 |
+
if y_true.shape != y_score.shape:
|
| 1089 |
+
raise ValueError("y_true and y_score have different shape")
|
| 1090 |
+
|
| 1091 |
+
# Handle badly formatted array and the degenerate case with one label
|
| 1092 |
+
y_type = type_of_target(y_true, input_name="y_true")
|
| 1093 |
+
if y_type != "multilabel-indicator" and not (
|
| 1094 |
+
y_type == "binary" and y_true.ndim == 2
|
| 1095 |
+
):
|
| 1096 |
+
raise ValueError("{0} format is not supported".format(y_type))
|
| 1097 |
+
|
| 1098 |
+
if not issparse(y_true):
|
| 1099 |
+
y_true = csr_matrix(y_true)
|
| 1100 |
+
|
| 1101 |
+
y_score = -y_score
|
| 1102 |
+
|
| 1103 |
+
n_samples, n_labels = y_true.shape
|
| 1104 |
+
|
| 1105 |
+
out = 0.0
|
| 1106 |
+
for i, (start, stop) in enumerate(zip(y_true.indptr, y_true.indptr[1:])):
|
| 1107 |
+
relevant = y_true.indices[start:stop]
|
| 1108 |
+
|
| 1109 |
+
if relevant.size == 0 or relevant.size == n_labels:
|
| 1110 |
+
# If all labels are relevant or unrelevant, the score is also
|
| 1111 |
+
# equal to 1. The label ranking has no meaning.
|
| 1112 |
+
aux = 1.0
|
| 1113 |
+
else:
|
| 1114 |
+
scores_i = y_score[i]
|
| 1115 |
+
rank = rankdata(scores_i, "max")[relevant]
|
| 1116 |
+
L = rankdata(scores_i[relevant], "max")
|
| 1117 |
+
aux = (L / rank).mean()
|
| 1118 |
+
|
| 1119 |
+
if sample_weight is not None:
|
| 1120 |
+
aux = aux * sample_weight[i]
|
| 1121 |
+
out += aux
|
| 1122 |
+
|
| 1123 |
+
if sample_weight is None:
|
| 1124 |
+
out /= n_samples
|
| 1125 |
+
else:
|
| 1126 |
+
out /= np.sum(sample_weight)
|
| 1127 |
+
|
| 1128 |
+
return out
|
| 1129 |
+
|
| 1130 |
+
|
| 1131 |
+
def coverage_error(y_true, y_score, *, sample_weight=None):
|
| 1132 |
+
"""Coverage error measure.
|
| 1133 |
+
|
| 1134 |
+
Compute how far we need to go through the ranked scores to cover all
|
| 1135 |
+
true labels. The best value is equal to the average number
|
| 1136 |
+
of labels in ``y_true`` per sample.
|
| 1137 |
+
|
| 1138 |
+
Ties in ``y_scores`` are broken by giving maximal rank that would have
|
| 1139 |
+
been assigned to all tied values.
|
| 1140 |
+
|
| 1141 |
+
Note: Our implementation's score is 1 greater than the one given in
|
| 1142 |
+
Tsoumakas et al., 2010. This extends it to handle the degenerate case
|
| 1143 |
+
in which an instance has 0 true labels.
|
| 1144 |
+
|
| 1145 |
+
Read more in the :ref:`User Guide <coverage_error>`.
|
| 1146 |
+
|
| 1147 |
+
Parameters
|
| 1148 |
+
----------
|
| 1149 |
+
y_true : ndarray of shape (n_samples, n_labels)
|
| 1150 |
+
True binary labels in binary indicator format.
|
| 1151 |
+
|
| 1152 |
+
y_score : ndarray of shape (n_samples, n_labels)
|
| 1153 |
+
Target scores, can either be probability estimates of the positive
|
| 1154 |
+
class, confidence values, or non-thresholded measure of decisions
|
| 1155 |
+
(as returned by "decision_function" on some classifiers).
|
| 1156 |
+
|
| 1157 |
+
sample_weight : array-like of shape (n_samples,), default=None
|
| 1158 |
+
Sample weights.
|
| 1159 |
+
|
| 1160 |
+
Returns
|
| 1161 |
+
-------
|
| 1162 |
+
coverage_error : float
|
| 1163 |
+
The coverage error.
|
| 1164 |
+
|
| 1165 |
+
References
|
| 1166 |
+
----------
|
| 1167 |
+
.. [1] Tsoumakas, G., Katakis, I., & Vlahavas, I. (2010).
|
| 1168 |
+
Mining multi-label data. In Data mining and knowledge discovery
|
| 1169 |
+
handbook (pp. 667-685). Springer US.
|
| 1170 |
+
"""
|
| 1171 |
+
y_true = check_array(y_true, ensure_2d=True)
|
| 1172 |
+
y_score = check_array(y_score, ensure_2d=True)
|
| 1173 |
+
check_consistent_length(y_true, y_score, sample_weight)
|
| 1174 |
+
|
| 1175 |
+
y_type = type_of_target(y_true, input_name="y_true")
|
| 1176 |
+
if y_type != "multilabel-indicator":
|
| 1177 |
+
raise ValueError("{0} format is not supported".format(y_type))
|
| 1178 |
+
|
| 1179 |
+
if y_true.shape != y_score.shape:
|
| 1180 |
+
raise ValueError("y_true and y_score have different shape")
|
| 1181 |
+
|
| 1182 |
+
y_score_mask = np.ma.masked_array(y_score, mask=np.logical_not(y_true))
|
| 1183 |
+
y_min_relevant = y_score_mask.min(axis=1).reshape((-1, 1))
|
| 1184 |
+
coverage = (y_score >= y_min_relevant).sum(axis=1)
|
| 1185 |
+
coverage = coverage.filled(0)
|
| 1186 |
+
|
| 1187 |
+
return np.average(coverage, weights=sample_weight)
|
| 1188 |
+
|
| 1189 |
+
|
| 1190 |
+
def label_ranking_loss(y_true, y_score, *, sample_weight=None):
|
| 1191 |
+
"""Compute Ranking loss measure.
|
| 1192 |
+
|
| 1193 |
+
Compute the average number of label pairs that are incorrectly ordered
|
| 1194 |
+
given y_score weighted by the size of the label set and the number of
|
| 1195 |
+
labels not in the label set.
|
| 1196 |
+
|
| 1197 |
+
This is similar to the error set size, but weighted by the number of
|
| 1198 |
+
relevant and irrelevant labels. The best performance is achieved with
|
| 1199 |
+
a ranking loss of zero.
|
| 1200 |
+
|
| 1201 |
+
Read more in the :ref:`User Guide <label_ranking_loss>`.
|
| 1202 |
+
|
| 1203 |
+
.. versionadded:: 0.17
|
| 1204 |
+
A function *label_ranking_loss*
|
| 1205 |
+
|
| 1206 |
+
Parameters
|
| 1207 |
+
----------
|
| 1208 |
+
y_true : {ndarray, sparse matrix} of shape (n_samples, n_labels)
|
| 1209 |
+
True binary labels in binary indicator format.
|
| 1210 |
+
|
| 1211 |
+
y_score : ndarray of shape (n_samples, n_labels)
|
| 1212 |
+
Target scores, can either be probability estimates of the positive
|
| 1213 |
+
class, confidence values, or non-thresholded measure of decisions
|
| 1214 |
+
(as returned by "decision_function" on some classifiers).
|
| 1215 |
+
|
| 1216 |
+
sample_weight : array-like of shape (n_samples,), default=None
|
| 1217 |
+
Sample weights.
|
| 1218 |
+
|
| 1219 |
+
Returns
|
| 1220 |
+
-------
|
| 1221 |
+
loss : float
|
| 1222 |
+
Average number of label pairs that are incorrectly ordered given
|
| 1223 |
+
y_score weighted by the size of the label set and the number of labels not
|
| 1224 |
+
in the label set.
|
| 1225 |
+
|
| 1226 |
+
References
|
| 1227 |
+
----------
|
| 1228 |
+
.. [1] Tsoumakas, G., Katakis, I., & Vlahavas, I. (2010).
|
| 1229 |
+
Mining multi-label data. In Data mining and knowledge discovery
|
| 1230 |
+
handbook (pp. 667-685). Springer US.
|
| 1231 |
+
"""
|
| 1232 |
+
y_true = check_array(y_true, ensure_2d=False, accept_sparse="csr")
|
| 1233 |
+
y_score = check_array(y_score, ensure_2d=False)
|
| 1234 |
+
check_consistent_length(y_true, y_score, sample_weight)
|
| 1235 |
+
|
| 1236 |
+
y_type = type_of_target(y_true, input_name="y_true")
|
| 1237 |
+
if y_type not in ("multilabel-indicator",):
|
| 1238 |
+
raise ValueError("{0} format is not supported".format(y_type))
|
| 1239 |
+
|
| 1240 |
+
if y_true.shape != y_score.shape:
|
| 1241 |
+
raise ValueError("y_true and y_score have different shape")
|
| 1242 |
+
|
| 1243 |
+
n_samples, n_labels = y_true.shape
|
| 1244 |
+
|
| 1245 |
+
y_true = csr_matrix(y_true)
|
| 1246 |
+
|
| 1247 |
+
loss = np.zeros(n_samples)
|
| 1248 |
+
for i, (start, stop) in enumerate(zip(y_true.indptr, y_true.indptr[1:])):
|
| 1249 |
+
# Sort and bin the label scores
|
| 1250 |
+
unique_scores, unique_inverse = np.unique(y_score[i], return_inverse=True)
|
| 1251 |
+
true_at_reversed_rank = np.bincount(
|
| 1252 |
+
unique_inverse[y_true.indices[start:stop]], minlength=len(unique_scores)
|
| 1253 |
+
)
|
| 1254 |
+
all_at_reversed_rank = np.bincount(unique_inverse, minlength=len(unique_scores))
|
| 1255 |
+
false_at_reversed_rank = all_at_reversed_rank - true_at_reversed_rank
|
| 1256 |
+
|
| 1257 |
+
# if the scores are ordered, it's possible to count the number of
|
| 1258 |
+
# incorrectly ordered paires in linear time by cumulatively counting
|
| 1259 |
+
# how many false labels of a given score have a score higher than the
|
| 1260 |
+
# accumulated true labels with lower score.
|
| 1261 |
+
loss[i] = np.dot(true_at_reversed_rank.cumsum(), false_at_reversed_rank)
|
| 1262 |
+
|
| 1263 |
+
n_positives = count_nonzero(y_true, axis=1)
|
| 1264 |
+
with np.errstate(divide="ignore", invalid="ignore"):
|
| 1265 |
+
loss /= (n_labels - n_positives) * n_positives
|
| 1266 |
+
|
| 1267 |
+
# When there is no positive or no negative labels, those values should
|
| 1268 |
+
# be consider as correct, i.e. the ranking doesn't matter.
|
| 1269 |
+
loss[np.logical_or(n_positives == 0, n_positives == n_labels)] = 0.0
|
| 1270 |
+
|
| 1271 |
+
return np.average(loss, weights=sample_weight)
|
| 1272 |
+
|
| 1273 |
+
|
| 1274 |
+
def _dcg_sample_scores(y_true, y_score, k=None, log_base=2, ignore_ties=False):
|
| 1275 |
+
"""Compute Discounted Cumulative Gain.
|
| 1276 |
+
|
| 1277 |
+
Sum the true scores ranked in the order induced by the predicted scores,
|
| 1278 |
+
after applying a logarithmic discount.
|
| 1279 |
+
|
| 1280 |
+
This ranking metric yields a high value if true labels are ranked high by
|
| 1281 |
+
``y_score``.
|
| 1282 |
+
|
| 1283 |
+
Parameters
|
| 1284 |
+
----------
|
| 1285 |
+
y_true : ndarray of shape (n_samples, n_labels)
|
| 1286 |
+
True targets of multilabel classification, or true scores of entities
|
| 1287 |
+
to be ranked.
|
| 1288 |
+
|
| 1289 |
+
y_score : ndarray of shape (n_samples, n_labels)
|
| 1290 |
+
Target scores, can either be probability estimates, confidence values,
|
| 1291 |
+
or non-thresholded measure of decisions (as returned by
|
| 1292 |
+
"decision_function" on some classifiers).
|
| 1293 |
+
|
| 1294 |
+
k : int, default=None
|
| 1295 |
+
Only consider the highest k scores in the ranking. If `None`, use all
|
| 1296 |
+
outputs.
|
| 1297 |
+
|
| 1298 |
+
log_base : float, default=2
|
| 1299 |
+
Base of the logarithm used for the discount. A low value means a
|
| 1300 |
+
sharper discount (top results are more important).
|
| 1301 |
+
|
| 1302 |
+
ignore_ties : bool, default=False
|
| 1303 |
+
Assume that there are no ties in y_score (which is likely to be the
|
| 1304 |
+
case if y_score is continuous) for efficiency gains.
|
| 1305 |
+
|
| 1306 |
+
Returns
|
| 1307 |
+
-------
|
| 1308 |
+
discounted_cumulative_gain : ndarray of shape (n_samples,)
|
| 1309 |
+
The DCG score for each sample.
|
| 1310 |
+
|
| 1311 |
+
See Also
|
| 1312 |
+
--------
|
| 1313 |
+
ndcg_score : The Discounted Cumulative Gain divided by the Ideal Discounted
|
| 1314 |
+
Cumulative Gain (the DCG obtained for a perfect ranking), in order to
|
| 1315 |
+
have a score between 0 and 1.
|
| 1316 |
+
"""
|
| 1317 |
+
discount = 1 / (np.log(np.arange(y_true.shape[1]) + 2) / np.log(log_base))
|
| 1318 |
+
if k is not None:
|
| 1319 |
+
discount[k:] = 0
|
| 1320 |
+
if ignore_ties:
|
| 1321 |
+
ranking = np.argsort(y_score)[:, ::-1]
|
| 1322 |
+
ranked = y_true[np.arange(ranking.shape[0])[:, np.newaxis], ranking]
|
| 1323 |
+
cumulative_gains = discount.dot(ranked.T)
|
| 1324 |
+
else:
|
| 1325 |
+
discount_cumsum = np.cumsum(discount)
|
| 1326 |
+
cumulative_gains = [
|
| 1327 |
+
_tie_averaged_dcg(y_t, y_s, discount_cumsum)
|
| 1328 |
+
for y_t, y_s in zip(y_true, y_score)
|
| 1329 |
+
]
|
| 1330 |
+
cumulative_gains = np.asarray(cumulative_gains)
|
| 1331 |
+
return cumulative_gains
|
| 1332 |
+
|
| 1333 |
+
|
| 1334 |
+
def _tie_averaged_dcg(y_true, y_score, discount_cumsum):
|
| 1335 |
+
"""
|
| 1336 |
+
Compute DCG by averaging over possible permutations of ties.
|
| 1337 |
+
|
| 1338 |
+
The gain (`y_true`) of an index falling inside a tied group (in the order
|
| 1339 |
+
induced by `y_score`) is replaced by the average gain within this group.
|
| 1340 |
+
The discounted gain for a tied group is then the average `y_true` within
|
| 1341 |
+
this group times the sum of discounts of the corresponding ranks.
|
| 1342 |
+
|
| 1343 |
+
This amounts to averaging scores for all possible orderings of the tied
|
| 1344 |
+
groups.
|
| 1345 |
+
|
| 1346 |
+
(note in the case of dcg@k the discount is 0 after index k)
|
| 1347 |
+
|
| 1348 |
+
Parameters
|
| 1349 |
+
----------
|
| 1350 |
+
y_true : ndarray
|
| 1351 |
+
The true relevance scores.
|
| 1352 |
+
|
| 1353 |
+
y_score : ndarray
|
| 1354 |
+
Predicted scores.
|
| 1355 |
+
|
| 1356 |
+
discount_cumsum : ndarray
|
| 1357 |
+
Precomputed cumulative sum of the discounts.
|
| 1358 |
+
|
| 1359 |
+
Returns
|
| 1360 |
+
-------
|
| 1361 |
+
discounted_cumulative_gain : float
|
| 1362 |
+
The discounted cumulative gain.
|
| 1363 |
+
|
| 1364 |
+
References
|
| 1365 |
+
----------
|
| 1366 |
+
McSherry, F., & Najork, M. (2008, March). Computing information retrieval
|
| 1367 |
+
performance measures efficiently in the presence of tied scores. In
|
| 1368 |
+
European conference on information retrieval (pp. 414-421). Springer,
|
| 1369 |
+
Berlin, Heidelberg.
|
| 1370 |
+
"""
|
| 1371 |
+
_, inv, counts = np.unique(-y_score, return_inverse=True, return_counts=True)
|
| 1372 |
+
ranked = np.zeros(len(counts))
|
| 1373 |
+
np.add.at(ranked, inv, y_true)
|
| 1374 |
+
ranked /= counts
|
| 1375 |
+
groups = np.cumsum(counts) - 1
|
| 1376 |
+
discount_sums = np.empty(len(counts))
|
| 1377 |
+
discount_sums[0] = discount_cumsum[groups[0]]
|
| 1378 |
+
discount_sums[1:] = np.diff(discount_cumsum[groups])
|
| 1379 |
+
return (ranked * discount_sums).sum()
|
| 1380 |
+
|
| 1381 |
+
|
| 1382 |
+
def _check_dcg_target_type(y_true):
|
| 1383 |
+
y_type = type_of_target(y_true, input_name="y_true")
|
| 1384 |
+
supported_fmt = (
|
| 1385 |
+
"multilabel-indicator",
|
| 1386 |
+
"continuous-multioutput",
|
| 1387 |
+
"multiclass-multioutput",
|
| 1388 |
+
)
|
| 1389 |
+
if y_type not in supported_fmt:
|
| 1390 |
+
raise ValueError(
|
| 1391 |
+
"Only {} formats are supported. Got {} instead".format(
|
| 1392 |
+
supported_fmt, y_type
|
| 1393 |
+
)
|
| 1394 |
+
)
|
| 1395 |
+
|
| 1396 |
+
|
| 1397 |
+
def dcg_score(
|
| 1398 |
+
y_true, y_score, *, k=None, log_base=2, sample_weight=None, ignore_ties=False
|
| 1399 |
+
):
|
| 1400 |
+
"""Compute Discounted Cumulative Gain.
|
| 1401 |
+
|
| 1402 |
+
Sum the true scores ranked in the order induced by the predicted scores,
|
| 1403 |
+
after applying a logarithmic discount.
|
| 1404 |
+
|
| 1405 |
+
This ranking metric yields a high value if true labels are ranked high by
|
| 1406 |
+
``y_score``.
|
| 1407 |
+
|
| 1408 |
+
Usually the Normalized Discounted Cumulative Gain (NDCG, computed by
|
| 1409 |
+
ndcg_score) is preferred.
|
| 1410 |
+
|
| 1411 |
+
Parameters
|
| 1412 |
+
----------
|
| 1413 |
+
y_true : ndarray of shape (n_samples, n_labels)
|
| 1414 |
+
True targets of multilabel classification, or true scores of entities
|
| 1415 |
+
to be ranked.
|
| 1416 |
+
|
| 1417 |
+
y_score : ndarray of shape (n_samples, n_labels)
|
| 1418 |
+
Target scores, can either be probability estimates, confidence values,
|
| 1419 |
+
or non-thresholded measure of decisions (as returned by
|
| 1420 |
+
"decision_function" on some classifiers).
|
| 1421 |
+
|
| 1422 |
+
k : int, default=None
|
| 1423 |
+
Only consider the highest k scores in the ranking. If None, use all
|
| 1424 |
+
outputs.
|
| 1425 |
+
|
| 1426 |
+
log_base : float, default=2
|
| 1427 |
+
Base of the logarithm used for the discount. A low value means a
|
| 1428 |
+
sharper discount (top results are more important).
|
| 1429 |
+
|
| 1430 |
+
sample_weight : ndarray of shape (n_samples,), default=None
|
| 1431 |
+
Sample weights. If `None`, all samples are given the same weight.
|
| 1432 |
+
|
| 1433 |
+
ignore_ties : bool, default=False
|
| 1434 |
+
Assume that there are no ties in y_score (which is likely to be the
|
| 1435 |
+
case if y_score is continuous) for efficiency gains.
|
| 1436 |
+
|
| 1437 |
+
Returns
|
| 1438 |
+
-------
|
| 1439 |
+
discounted_cumulative_gain : float
|
| 1440 |
+
The averaged sample DCG scores.
|
| 1441 |
+
|
| 1442 |
+
See Also
|
| 1443 |
+
--------
|
| 1444 |
+
ndcg_score : The Discounted Cumulative Gain divided by the Ideal Discounted
|
| 1445 |
+
Cumulative Gain (the DCG obtained for a perfect ranking), in order to
|
| 1446 |
+
have a score between 0 and 1.
|
| 1447 |
+
|
| 1448 |
+
References
|
| 1449 |
+
----------
|
| 1450 |
+
`Wikipedia entry for Discounted Cumulative Gain
|
| 1451 |
+
<https://en.wikipedia.org/wiki/Discounted_cumulative_gain>`_.
|
| 1452 |
+
|
| 1453 |
+
Jarvelin, K., & Kekalainen, J. (2002).
|
| 1454 |
+
Cumulated gain-based evaluation of IR techniques. ACM Transactions on
|
| 1455 |
+
Information Systems (TOIS), 20(4), 422-446.
|
| 1456 |
+
|
| 1457 |
+
Wang, Y., Wang, L., Li, Y., He, D., Chen, W., & Liu, T. Y. (2013, May).
|
| 1458 |
+
A theoretical analysis of NDCG ranking measures. In Proceedings of the 26th
|
| 1459 |
+
Annual Conference on Learning Theory (COLT 2013).
|
| 1460 |
+
|
| 1461 |
+
McSherry, F., & Najork, M. (2008, March). Computing information retrieval
|
| 1462 |
+
performance measures efficiently in the presence of tied scores. In
|
| 1463 |
+
European conference on information retrieval (pp. 414-421). Springer,
|
| 1464 |
+
Berlin, Heidelberg.
|
| 1465 |
+
|
| 1466 |
+
Examples
|
| 1467 |
+
--------
|
| 1468 |
+
>>> import numpy as np
|
| 1469 |
+
>>> from sklearn.metrics import dcg_score
|
| 1470 |
+
>>> # we have groud-truth relevance of some answers to a query:
|
| 1471 |
+
>>> true_relevance = np.asarray([[10, 0, 0, 1, 5]])
|
| 1472 |
+
>>> # we predict scores for the answers
|
| 1473 |
+
>>> scores = np.asarray([[.1, .2, .3, 4, 70]])
|
| 1474 |
+
>>> dcg_score(true_relevance, scores)
|
| 1475 |
+
9.49...
|
| 1476 |
+
>>> # we can set k to truncate the sum; only top k answers contribute
|
| 1477 |
+
>>> dcg_score(true_relevance, scores, k=2)
|
| 1478 |
+
5.63...
|
| 1479 |
+
>>> # now we have some ties in our prediction
|
| 1480 |
+
>>> scores = np.asarray([[1, 0, 0, 0, 1]])
|
| 1481 |
+
>>> # by default ties are averaged, so here we get the average true
|
| 1482 |
+
>>> # relevance of our top predictions: (10 + 5) / 2 = 7.5
|
| 1483 |
+
>>> dcg_score(true_relevance, scores, k=1)
|
| 1484 |
+
7.5
|
| 1485 |
+
>>> # we can choose to ignore ties for faster results, but only
|
| 1486 |
+
>>> # if we know there aren't ties in our scores, otherwise we get
|
| 1487 |
+
>>> # wrong results:
|
| 1488 |
+
>>> dcg_score(true_relevance,
|
| 1489 |
+
... scores, k=1, ignore_ties=True)
|
| 1490 |
+
5.0
|
| 1491 |
+
"""
|
| 1492 |
+
y_true = check_array(y_true, ensure_2d=False)
|
| 1493 |
+
y_score = check_array(y_score, ensure_2d=False)
|
| 1494 |
+
check_consistent_length(y_true, y_score, sample_weight)
|
| 1495 |
+
_check_dcg_target_type(y_true)
|
| 1496 |
+
return np.average(
|
| 1497 |
+
_dcg_sample_scores(
|
| 1498 |
+
y_true, y_score, k=k, log_base=log_base, ignore_ties=ignore_ties
|
| 1499 |
+
),
|
| 1500 |
+
weights=sample_weight,
|
| 1501 |
+
)
|
| 1502 |
+
|
| 1503 |
+
|
| 1504 |
+
def _ndcg_sample_scores(y_true, y_score, k=None, ignore_ties=False):
|
| 1505 |
+
"""Compute Normalized Discounted Cumulative Gain.
|
| 1506 |
+
|
| 1507 |
+
Sum the true scores ranked in the order induced by the predicted scores,
|
| 1508 |
+
after applying a logarithmic discount. Then divide by the best possible
|
| 1509 |
+
score (Ideal DCG, obtained for a perfect ranking) to obtain a score between
|
| 1510 |
+
0 and 1.
|
| 1511 |
+
|
| 1512 |
+
This ranking metric yields a high value if true labels are ranked high by
|
| 1513 |
+
``y_score``.
|
| 1514 |
+
|
| 1515 |
+
Parameters
|
| 1516 |
+
----------
|
| 1517 |
+
y_true : ndarray of shape (n_samples, n_labels)
|
| 1518 |
+
True targets of multilabel classification, or true scores of entities
|
| 1519 |
+
to be ranked.
|
| 1520 |
+
|
| 1521 |
+
y_score : ndarray of shape (n_samples, n_labels)
|
| 1522 |
+
Target scores, can either be probability estimates, confidence values,
|
| 1523 |
+
or non-thresholded measure of decisions (as returned by
|
| 1524 |
+
"decision_function" on some classifiers).
|
| 1525 |
+
|
| 1526 |
+
k : int, default=None
|
| 1527 |
+
Only consider the highest k scores in the ranking. If None, use all
|
| 1528 |
+
outputs.
|
| 1529 |
+
|
| 1530 |
+
ignore_ties : bool, default=False
|
| 1531 |
+
Assume that there are no ties in y_score (which is likely to be the
|
| 1532 |
+
case if y_score is continuous) for efficiency gains.
|
| 1533 |
+
|
| 1534 |
+
Returns
|
| 1535 |
+
-------
|
| 1536 |
+
normalized_discounted_cumulative_gain : ndarray of shape (n_samples,)
|
| 1537 |
+
The NDCG score for each sample (float in [0., 1.]).
|
| 1538 |
+
|
| 1539 |
+
See Also
|
| 1540 |
+
--------
|
| 1541 |
+
dcg_score : Discounted Cumulative Gain (not normalized).
|
| 1542 |
+
|
| 1543 |
+
"""
|
| 1544 |
+
gain = _dcg_sample_scores(y_true, y_score, k, ignore_ties=ignore_ties)
|
| 1545 |
+
# Here we use the order induced by y_true so we can ignore ties since
|
| 1546 |
+
# the gain associated to tied indices is the same (permuting ties doesn't
|
| 1547 |
+
# change the value of the re-ordered y_true)
|
| 1548 |
+
normalizing_gain = _dcg_sample_scores(y_true, y_true, k, ignore_ties=True)
|
| 1549 |
+
all_irrelevant = normalizing_gain == 0
|
| 1550 |
+
gain[all_irrelevant] = 0
|
| 1551 |
+
gain[~all_irrelevant] /= normalizing_gain[~all_irrelevant]
|
| 1552 |
+
return gain
|
| 1553 |
+
|
| 1554 |
+
|
| 1555 |
+
def ndcg_score(y_true, y_score, *, k=None, sample_weight=None, ignore_ties=False):
|
| 1556 |
+
"""Compute Normalized Discounted Cumulative Gain.
|
| 1557 |
+
|
| 1558 |
+
Sum the true scores ranked in the order induced by the predicted scores,
|
| 1559 |
+
after applying a logarithmic discount. Then divide by the best possible
|
| 1560 |
+
score (Ideal DCG, obtained for a perfect ranking) to obtain a score between
|
| 1561 |
+
0 and 1.
|
| 1562 |
+
|
| 1563 |
+
This ranking metric returns a high value if true labels are ranked high by
|
| 1564 |
+
``y_score``.
|
| 1565 |
+
|
| 1566 |
+
Parameters
|
| 1567 |
+
----------
|
| 1568 |
+
y_true : ndarray of shape (n_samples, n_labels)
|
| 1569 |
+
True targets of multilabel classification, or true scores of entities
|
| 1570 |
+
to be ranked. Negative values in `y_true` may result in an output
|
| 1571 |
+
that is not between 0 and 1.
|
| 1572 |
+
|
| 1573 |
+
.. versionchanged:: 1.2
|
| 1574 |
+
These negative values are deprecated, and will raise an error in v1.4.
|
| 1575 |
+
|
| 1576 |
+
y_score : ndarray of shape (n_samples, n_labels)
|
| 1577 |
+
Target scores, can either be probability estimates, confidence values,
|
| 1578 |
+
or non-thresholded measure of decisions (as returned by
|
| 1579 |
+
"decision_function" on some classifiers).
|
| 1580 |
+
|
| 1581 |
+
k : int, default=None
|
| 1582 |
+
Only consider the highest k scores in the ranking. If `None`, use all
|
| 1583 |
+
outputs.
|
| 1584 |
+
|
| 1585 |
+
sample_weight : ndarray of shape (n_samples,), default=None
|
| 1586 |
+
Sample weights. If `None`, all samples are given the same weight.
|
| 1587 |
+
|
| 1588 |
+
ignore_ties : bool, default=False
|
| 1589 |
+
Assume that there are no ties in y_score (which is likely to be the
|
| 1590 |
+
case if y_score is continuous) for efficiency gains.
|
| 1591 |
+
|
| 1592 |
+
Returns
|
| 1593 |
+
-------
|
| 1594 |
+
normalized_discounted_cumulative_gain : float in [0., 1.]
|
| 1595 |
+
The averaged NDCG scores for all samples.
|
| 1596 |
+
|
| 1597 |
+
See Also
|
| 1598 |
+
--------
|
| 1599 |
+
dcg_score : Discounted Cumulative Gain (not normalized).
|
| 1600 |
+
|
| 1601 |
+
References
|
| 1602 |
+
----------
|
| 1603 |
+
`Wikipedia entry for Discounted Cumulative Gain
|
| 1604 |
+
<https://en.wikipedia.org/wiki/Discounted_cumulative_gain>`_
|
| 1605 |
+
|
| 1606 |
+
Jarvelin, K., & Kekalainen, J. (2002).
|
| 1607 |
+
Cumulated gain-based evaluation of IR techniques. ACM Transactions on
|
| 1608 |
+
Information Systems (TOIS), 20(4), 422-446.
|
| 1609 |
+
|
| 1610 |
+
Wang, Y., Wang, L., Li, Y., He, D., Chen, W., & Liu, T. Y. (2013, May).
|
| 1611 |
+
A theoretical analysis of NDCG ranking measures. In Proceedings of the 26th
|
| 1612 |
+
Annual Conference on Learning Theory (COLT 2013)
|
| 1613 |
+
|
| 1614 |
+
McSherry, F., & Najork, M. (2008, March). Computing information retrieval
|
| 1615 |
+
performance measures efficiently in the presence of tied scores. In
|
| 1616 |
+
European conference on information retrieval (pp. 414-421). Springer,
|
| 1617 |
+
Berlin, Heidelberg.
|
| 1618 |
+
|
| 1619 |
+
Examples
|
| 1620 |
+
--------
|
| 1621 |
+
>>> import numpy as np
|
| 1622 |
+
>>> from sklearn.metrics import ndcg_score
|
| 1623 |
+
>>> # we have groud-truth relevance of some answers to a query:
|
| 1624 |
+
>>> true_relevance = np.asarray([[10, 0, 0, 1, 5]])
|
| 1625 |
+
>>> # we predict some scores (relevance) for the answers
|
| 1626 |
+
>>> scores = np.asarray([[.1, .2, .3, 4, 70]])
|
| 1627 |
+
>>> ndcg_score(true_relevance, scores)
|
| 1628 |
+
0.69...
|
| 1629 |
+
>>> scores = np.asarray([[.05, 1.1, 1., .5, .0]])
|
| 1630 |
+
>>> ndcg_score(true_relevance, scores)
|
| 1631 |
+
0.49...
|
| 1632 |
+
>>> # we can set k to truncate the sum; only top k answers contribute.
|
| 1633 |
+
>>> ndcg_score(true_relevance, scores, k=4)
|
| 1634 |
+
0.35...
|
| 1635 |
+
>>> # the normalization takes k into account so a perfect answer
|
| 1636 |
+
>>> # would still get 1.0
|
| 1637 |
+
>>> ndcg_score(true_relevance, true_relevance, k=4)
|
| 1638 |
+
1.0...
|
| 1639 |
+
>>> # now we have some ties in our prediction
|
| 1640 |
+
>>> scores = np.asarray([[1, 0, 0, 0, 1]])
|
| 1641 |
+
>>> # by default ties are averaged, so here we get the average (normalized)
|
| 1642 |
+
>>> # true relevance of our top predictions: (10 / 10 + 5 / 10) / 2 = .75
|
| 1643 |
+
>>> ndcg_score(true_relevance, scores, k=1)
|
| 1644 |
+
0.75...
|
| 1645 |
+
>>> # we can choose to ignore ties for faster results, but only
|
| 1646 |
+
>>> # if we know there aren't ties in our scores, otherwise we get
|
| 1647 |
+
>>> # wrong results:
|
| 1648 |
+
>>> ndcg_score(true_relevance,
|
| 1649 |
+
... scores, k=1, ignore_ties=True)
|
| 1650 |
+
0.5...
|
| 1651 |
+
"""
|
| 1652 |
+
y_true = check_array(y_true, ensure_2d=False)
|
| 1653 |
+
y_score = check_array(y_score, ensure_2d=False)
|
| 1654 |
+
check_consistent_length(y_true, y_score, sample_weight)
|
| 1655 |
+
|
| 1656 |
+
if y_true.min() < 0:
|
| 1657 |
+
# TODO(1.4): Replace warning w/ ValueError
|
| 1658 |
+
warnings.warn(
|
| 1659 |
+
"ndcg_score should not be used on negative y_true values. ndcg_score will"
|
| 1660 |
+
" raise a ValueError on negative y_true values starting from version 1.4.",
|
| 1661 |
+
FutureWarning,
|
| 1662 |
+
)
|
| 1663 |
+
_check_dcg_target_type(y_true)
|
| 1664 |
+
gain = _ndcg_sample_scores(y_true, y_score, k=k, ignore_ties=ignore_ties)
|
| 1665 |
+
return np.average(gain, weights=sample_weight)
|
| 1666 |
+
|
| 1667 |
+
|
| 1668 |
+
def top_k_accuracy_score(
|
| 1669 |
+
y_true, y_score, *, k=2, normalize=True, sample_weight=None, labels=None
|
| 1670 |
+
):
|
| 1671 |
+
"""Top-k Accuracy classification score.
|
| 1672 |
+
|
| 1673 |
+
This metric computes the number of times where the correct label is among
|
| 1674 |
+
the top `k` labels predicted (ranked by predicted scores). Note that the
|
| 1675 |
+
multilabel case isn't covered here.
|
| 1676 |
+
|
| 1677 |
+
Read more in the :ref:`User Guide <top_k_accuracy_score>`
|
| 1678 |
+
|
| 1679 |
+
Parameters
|
| 1680 |
+
----------
|
| 1681 |
+
y_true : array-like of shape (n_samples,)
|
| 1682 |
+
True labels.
|
| 1683 |
+
|
| 1684 |
+
y_score : array-like of shape (n_samples,) or (n_samples, n_classes)
|
| 1685 |
+
Target scores. These can be either probability estimates or
|
| 1686 |
+
non-thresholded decision values (as returned by
|
| 1687 |
+
:term:`decision_function` on some classifiers).
|
| 1688 |
+
The binary case expects scores with shape (n_samples,) while the
|
| 1689 |
+
multiclass case expects scores with shape (n_samples, n_classes).
|
| 1690 |
+
In the multiclass case, the order of the class scores must
|
| 1691 |
+
correspond to the order of ``labels``, if provided, or else to
|
| 1692 |
+
the numerical or lexicographical order of the labels in ``y_true``.
|
| 1693 |
+
If ``y_true`` does not contain all the labels, ``labels`` must be
|
| 1694 |
+
provided.
|
| 1695 |
+
|
| 1696 |
+
k : int, default=2
|
| 1697 |
+
Number of most likely outcomes considered to find the correct label.
|
| 1698 |
+
|
| 1699 |
+
normalize : bool, default=True
|
| 1700 |
+
If `True`, return the fraction of correctly classified samples.
|
| 1701 |
+
Otherwise, return the number of correctly classified samples.
|
| 1702 |
+
|
| 1703 |
+
sample_weight : array-like of shape (n_samples,), default=None
|
| 1704 |
+
Sample weights. If `None`, all samples are given the same weight.
|
| 1705 |
+
|
| 1706 |
+
labels : array-like of shape (n_classes,), default=None
|
| 1707 |
+
Multiclass only. List of labels that index the classes in ``y_score``.
|
| 1708 |
+
If ``None``, the numerical or lexicographical order of the labels in
|
| 1709 |
+
``y_true`` is used. If ``y_true`` does not contain all the labels,
|
| 1710 |
+
``labels`` must be provided.
|
| 1711 |
+
|
| 1712 |
+
Returns
|
| 1713 |
+
-------
|
| 1714 |
+
score : float
|
| 1715 |
+
The top-k accuracy score. The best performance is 1 with
|
| 1716 |
+
`normalize == True` and the number of samples with
|
| 1717 |
+
`normalize == False`.
|
| 1718 |
+
|
| 1719 |
+
See Also
|
| 1720 |
+
--------
|
| 1721 |
+
accuracy_score : Compute the accuracy score. By default, the function will
|
| 1722 |
+
return the fraction of correct predictions divided by the total number
|
| 1723 |
+
of predictions.
|
| 1724 |
+
|
| 1725 |
+
Notes
|
| 1726 |
+
-----
|
| 1727 |
+
In cases where two or more labels are assigned equal predicted scores,
|
| 1728 |
+
the labels with the highest indices will be chosen first. This might
|
| 1729 |
+
impact the result if the correct label falls after the threshold because
|
| 1730 |
+
of that.
|
| 1731 |
+
|
| 1732 |
+
Examples
|
| 1733 |
+
--------
|
| 1734 |
+
>>> import numpy as np
|
| 1735 |
+
>>> from sklearn.metrics import top_k_accuracy_score
|
| 1736 |
+
>>> y_true = np.array([0, 1, 2, 2])
|
| 1737 |
+
>>> y_score = np.array([[0.5, 0.2, 0.2], # 0 is in top 2
|
| 1738 |
+
... [0.3, 0.4, 0.2], # 1 is in top 2
|
| 1739 |
+
... [0.2, 0.4, 0.3], # 2 is in top 2
|
| 1740 |
+
... [0.7, 0.2, 0.1]]) # 2 isn't in top 2
|
| 1741 |
+
>>> top_k_accuracy_score(y_true, y_score, k=2)
|
| 1742 |
+
0.75
|
| 1743 |
+
>>> # Not normalizing gives the number of "correctly" classified samples
|
| 1744 |
+
>>> top_k_accuracy_score(y_true, y_score, k=2, normalize=False)
|
| 1745 |
+
3
|
| 1746 |
+
"""
|
| 1747 |
+
y_true = check_array(y_true, ensure_2d=False, dtype=None)
|
| 1748 |
+
y_true = column_or_1d(y_true)
|
| 1749 |
+
y_type = type_of_target(y_true, input_name="y_true")
|
| 1750 |
+
if y_type == "binary" and labels is not None and len(labels) > 2:
|
| 1751 |
+
y_type = "multiclass"
|
| 1752 |
+
if y_type not in {"binary", "multiclass"}:
|
| 1753 |
+
raise ValueError(
|
| 1754 |
+
f"y type must be 'binary' or 'multiclass', got '{y_type}' instead."
|
| 1755 |
+
)
|
| 1756 |
+
y_score = check_array(y_score, ensure_2d=False)
|
| 1757 |
+
if y_type == "binary":
|
| 1758 |
+
if y_score.ndim == 2 and y_score.shape[1] != 1:
|
| 1759 |
+
raise ValueError(
|
| 1760 |
+
"`y_true` is binary while y_score is 2d with"
|
| 1761 |
+
f" {y_score.shape[1]} classes. If `y_true` does not contain all the"
|
| 1762 |
+
" labels, `labels` must be provided."
|
| 1763 |
+
)
|
| 1764 |
+
y_score = column_or_1d(y_score)
|
| 1765 |
+
|
| 1766 |
+
check_consistent_length(y_true, y_score, sample_weight)
|
| 1767 |
+
y_score_n_classes = y_score.shape[1] if y_score.ndim == 2 else 2
|
| 1768 |
+
|
| 1769 |
+
if labels is None:
|
| 1770 |
+
classes = _unique(y_true)
|
| 1771 |
+
n_classes = len(classes)
|
| 1772 |
+
|
| 1773 |
+
if n_classes != y_score_n_classes:
|
| 1774 |
+
raise ValueError(
|
| 1775 |
+
f"Number of classes in 'y_true' ({n_classes}) not equal "
|
| 1776 |
+
f"to the number of classes in 'y_score' ({y_score_n_classes})."
|
| 1777 |
+
"You can provide a list of all known classes by assigning it "
|
| 1778 |
+
"to the `labels` parameter."
|
| 1779 |
+
)
|
| 1780 |
+
else:
|
| 1781 |
+
labels = column_or_1d(labels)
|
| 1782 |
+
classes = _unique(labels)
|
| 1783 |
+
n_labels = len(labels)
|
| 1784 |
+
n_classes = len(classes)
|
| 1785 |
+
|
| 1786 |
+
if n_classes != n_labels:
|
| 1787 |
+
raise ValueError("Parameter 'labels' must be unique.")
|
| 1788 |
+
|
| 1789 |
+
if not np.array_equal(classes, labels):
|
| 1790 |
+
raise ValueError("Parameter 'labels' must be ordered.")
|
| 1791 |
+
|
| 1792 |
+
if n_classes != y_score_n_classes:
|
| 1793 |
+
raise ValueError(
|
| 1794 |
+
f"Number of given labels ({n_classes}) not equal to the "
|
| 1795 |
+
f"number of classes in 'y_score' ({y_score_n_classes})."
|
| 1796 |
+
)
|
| 1797 |
+
|
| 1798 |
+
if len(np.setdiff1d(y_true, classes)):
|
| 1799 |
+
raise ValueError("'y_true' contains labels not in parameter 'labels'.")
|
| 1800 |
+
|
| 1801 |
+
if k >= n_classes:
|
| 1802 |
+
warnings.warn(
|
| 1803 |
+
f"'k' ({k}) greater than or equal to 'n_classes' ({n_classes}) "
|
| 1804 |
+
"will result in a perfect score and is therefore meaningless.",
|
| 1805 |
+
UndefinedMetricWarning,
|
| 1806 |
+
)
|
| 1807 |
+
|
| 1808 |
+
y_true_encoded = _encode(y_true, uniques=classes)
|
| 1809 |
+
|
| 1810 |
+
if y_type == "binary":
|
| 1811 |
+
if k == 1:
|
| 1812 |
+
threshold = 0.5 if y_score.min() >= 0 and y_score.max() <= 1 else 0
|
| 1813 |
+
y_pred = (y_score > threshold).astype(np.int64)
|
| 1814 |
+
hits = y_pred == y_true_encoded
|
| 1815 |
+
else:
|
| 1816 |
+
hits = np.ones_like(y_score, dtype=np.bool_)
|
| 1817 |
+
elif y_type == "multiclass":
|
| 1818 |
+
sorted_pred = np.argsort(y_score, axis=1, kind="mergesort")[:, ::-1]
|
| 1819 |
+
hits = (y_true_encoded == sorted_pred[:, :k].T).any(axis=0)
|
| 1820 |
+
|
| 1821 |
+
if normalize:
|
| 1822 |
+
return np.average(hits, weights=sample_weight)
|
| 1823 |
+
elif sample_weight is None:
|
| 1824 |
+
return np.sum(hits)
|
| 1825 |
+
else:
|
| 1826 |
+
return np.dot(hits, sample_weight)
|
videollama2/lib/python3.10/site-packages/sklearn/tree/_utils.cpython-310-x86_64-linux-gnu.so
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:f61b3136021cc79f20389c80b3c58dc74558c0604ae25ee04ebbe6bdd3db52b0
|
| 3 |
+
size 241201
|
videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/_available_if.cpython-310.pyc
ADDED
|
Binary file (2.92 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/_encode.cpython-310.pyc
ADDED
|
Binary file (10.2 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/_mask.cpython-310.pyc
ADDED
|
Binary file (1.65 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/_mocking.cpython-310.pyc
ADDED
|
Binary file (11.3 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/_set_output.cpython-310.pyc
ADDED
|
Binary file (7.71 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/_show_versions.cpython-310.pyc
ADDED
|
Binary file (2.27 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/_tags.cpython-310.pyc
ADDED
|
Binary file (1.98 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/_testing.cpython-310.pyc
ADDED
|
Binary file (30.7 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/deprecation.cpython-310.pyc
ADDED
|
Binary file (3.3 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/estimator_checks.cpython-310.pyc
ADDED
|
Binary file (94.6 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/extmath.cpython-310.pyc
ADDED
|
Binary file (33.8 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/fixes.cpython-310.pyc
ADDED
|
Binary file (5.53 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/graph.cpython-310.pyc
ADDED
|
Binary file (4.64 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/multiclass.cpython-310.pyc
ADDED
|
Binary file (14.1 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/optimize.cpython-310.pyc
ADDED
|
Binary file (5.96 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/stats.cpython-310.pyc
ADDED
|
Binary file (2.13 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/validation.cpython-310.pyc
ADDED
|
Binary file (57.9 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/_estimator_html_repr.py
ADDED
|
@@ -0,0 +1,428 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from contextlib import closing
|
| 2 |
+
from io import StringIO
|
| 3 |
+
from inspect import isclass
|
| 4 |
+
from string import Template
|
| 5 |
+
import html
|
| 6 |
+
|
| 7 |
+
from .. import config_context
|
| 8 |
+
|
| 9 |
+
|
| 10 |
+
class _IDCounter:
|
| 11 |
+
"""Generate sequential ids with a prefix."""
|
| 12 |
+
|
| 13 |
+
def __init__(self, prefix):
|
| 14 |
+
self.prefix = prefix
|
| 15 |
+
self.count = 0
|
| 16 |
+
|
| 17 |
+
def get_id(self):
|
| 18 |
+
self.count += 1
|
| 19 |
+
return f"{self.prefix}-{self.count}"
|
| 20 |
+
|
| 21 |
+
|
| 22 |
+
_CONTAINER_ID_COUNTER = _IDCounter("sk-container-id")
|
| 23 |
+
_ESTIMATOR_ID_COUNTER = _IDCounter("sk-estimator-id")
|
| 24 |
+
|
| 25 |
+
|
| 26 |
+
class _VisualBlock:
|
| 27 |
+
"""HTML Representation of Estimator
|
| 28 |
+
|
| 29 |
+
Parameters
|
| 30 |
+
----------
|
| 31 |
+
kind : {'serial', 'parallel', 'single'}
|
| 32 |
+
kind of HTML block
|
| 33 |
+
|
| 34 |
+
estimators : list of estimators or `_VisualBlock`s or a single estimator
|
| 35 |
+
If kind != 'single', then `estimators` is a list of
|
| 36 |
+
estimators.
|
| 37 |
+
If kind == 'single', then `estimators` is a single estimator.
|
| 38 |
+
|
| 39 |
+
names : list of str, default=None
|
| 40 |
+
If kind != 'single', then `names` corresponds to estimators.
|
| 41 |
+
If kind == 'single', then `names` is a single string corresponding to
|
| 42 |
+
the single estimator.
|
| 43 |
+
|
| 44 |
+
name_details : list of str, str, or None, default=None
|
| 45 |
+
If kind != 'single', then `name_details` corresponds to `names`.
|
| 46 |
+
If kind == 'single', then `name_details` is a single string
|
| 47 |
+
corresponding to the single estimator.
|
| 48 |
+
|
| 49 |
+
dash_wrapped : bool, default=True
|
| 50 |
+
If true, wrapped HTML element will be wrapped with a dashed border.
|
| 51 |
+
Only active when kind != 'single'.
|
| 52 |
+
"""
|
| 53 |
+
|
| 54 |
+
def __init__(
|
| 55 |
+
self, kind, estimators, *, names=None, name_details=None, dash_wrapped=True
|
| 56 |
+
):
|
| 57 |
+
self.kind = kind
|
| 58 |
+
self.estimators = estimators
|
| 59 |
+
self.dash_wrapped = dash_wrapped
|
| 60 |
+
|
| 61 |
+
if self.kind in ("parallel", "serial"):
|
| 62 |
+
if names is None:
|
| 63 |
+
names = (None,) * len(estimators)
|
| 64 |
+
if name_details is None:
|
| 65 |
+
name_details = (None,) * len(estimators)
|
| 66 |
+
|
| 67 |
+
self.names = names
|
| 68 |
+
self.name_details = name_details
|
| 69 |
+
|
| 70 |
+
def _sk_visual_block_(self):
|
| 71 |
+
return self
|
| 72 |
+
|
| 73 |
+
|
| 74 |
+
def _write_label_html(
|
| 75 |
+
out,
|
| 76 |
+
name,
|
| 77 |
+
name_details,
|
| 78 |
+
outer_class="sk-label-container",
|
| 79 |
+
inner_class="sk-label",
|
| 80 |
+
checked=False,
|
| 81 |
+
):
|
| 82 |
+
"""Write labeled html with or without a dropdown with named details"""
|
| 83 |
+
out.write(f'<div class="{outer_class}"><div class="{inner_class} sk-toggleable">')
|
| 84 |
+
name = html.escape(name)
|
| 85 |
+
|
| 86 |
+
if name_details is not None:
|
| 87 |
+
name_details = html.escape(str(name_details))
|
| 88 |
+
label_class = "sk-toggleable__label sk-toggleable__label-arrow"
|
| 89 |
+
|
| 90 |
+
checked_str = "checked" if checked else ""
|
| 91 |
+
est_id = _ESTIMATOR_ID_COUNTER.get_id()
|
| 92 |
+
out.write(
|
| 93 |
+
'<input class="sk-toggleable__control sk-hidden--visually" '
|
| 94 |
+
f'id="{est_id}" type="checkbox" {checked_str}>'
|
| 95 |
+
f'<label for="{est_id}" class="{label_class}">{name}</label>'
|
| 96 |
+
f'<div class="sk-toggleable__content"><pre>{name_details}'
|
| 97 |
+
"</pre></div>"
|
| 98 |
+
)
|
| 99 |
+
else:
|
| 100 |
+
out.write(f"<label>{name}</label>")
|
| 101 |
+
out.write("</div></div>") # outer_class inner_class
|
| 102 |
+
|
| 103 |
+
|
| 104 |
+
def _get_visual_block(estimator):
|
| 105 |
+
"""Generate information about how to display an estimator."""
|
| 106 |
+
if hasattr(estimator, "_sk_visual_block_"):
|
| 107 |
+
try:
|
| 108 |
+
return estimator._sk_visual_block_()
|
| 109 |
+
except Exception:
|
| 110 |
+
return _VisualBlock(
|
| 111 |
+
"single",
|
| 112 |
+
estimator,
|
| 113 |
+
names=estimator.__class__.__name__,
|
| 114 |
+
name_details=str(estimator),
|
| 115 |
+
)
|
| 116 |
+
|
| 117 |
+
if isinstance(estimator, str):
|
| 118 |
+
return _VisualBlock(
|
| 119 |
+
"single", estimator, names=estimator, name_details=estimator
|
| 120 |
+
)
|
| 121 |
+
elif estimator is None:
|
| 122 |
+
return _VisualBlock("single", estimator, names="None", name_details="None")
|
| 123 |
+
|
| 124 |
+
# check if estimator looks like a meta estimator wraps estimators
|
| 125 |
+
if hasattr(estimator, "get_params") and not isclass(estimator):
|
| 126 |
+
estimators = [
|
| 127 |
+
(key, est)
|
| 128 |
+
for key, est in estimator.get_params(deep=False).items()
|
| 129 |
+
if hasattr(est, "get_params") and hasattr(est, "fit") and not isclass(est)
|
| 130 |
+
]
|
| 131 |
+
if estimators:
|
| 132 |
+
return _VisualBlock(
|
| 133 |
+
"parallel",
|
| 134 |
+
[est for _, est in estimators],
|
| 135 |
+
names=[f"{key}: {est.__class__.__name__}" for key, est in estimators],
|
| 136 |
+
name_details=[str(est) for _, est in estimators],
|
| 137 |
+
)
|
| 138 |
+
|
| 139 |
+
return _VisualBlock(
|
| 140 |
+
"single",
|
| 141 |
+
estimator,
|
| 142 |
+
names=estimator.__class__.__name__,
|
| 143 |
+
name_details=str(estimator),
|
| 144 |
+
)
|
| 145 |
+
|
| 146 |
+
|
| 147 |
+
def _write_estimator_html(
|
| 148 |
+
out, estimator, estimator_label, estimator_label_details, first_call=False
|
| 149 |
+
):
|
| 150 |
+
"""Write estimator to html in serial, parallel, or by itself (single)."""
|
| 151 |
+
if first_call:
|
| 152 |
+
est_block = _get_visual_block(estimator)
|
| 153 |
+
else:
|
| 154 |
+
with config_context(print_changed_only=True):
|
| 155 |
+
est_block = _get_visual_block(estimator)
|
| 156 |
+
|
| 157 |
+
if est_block.kind in ("serial", "parallel"):
|
| 158 |
+
dashed_wrapped = first_call or est_block.dash_wrapped
|
| 159 |
+
dash_cls = " sk-dashed-wrapped" if dashed_wrapped else ""
|
| 160 |
+
out.write(f'<div class="sk-item{dash_cls}">')
|
| 161 |
+
|
| 162 |
+
if estimator_label:
|
| 163 |
+
_write_label_html(out, estimator_label, estimator_label_details)
|
| 164 |
+
|
| 165 |
+
kind = est_block.kind
|
| 166 |
+
out.write(f'<div class="sk-{kind}">')
|
| 167 |
+
est_infos = zip(est_block.estimators, est_block.names, est_block.name_details)
|
| 168 |
+
|
| 169 |
+
for est, name, name_details in est_infos:
|
| 170 |
+
if kind == "serial":
|
| 171 |
+
_write_estimator_html(out, est, name, name_details)
|
| 172 |
+
else: # parallel
|
| 173 |
+
out.write('<div class="sk-parallel-item">')
|
| 174 |
+
# wrap element in a serial visualblock
|
| 175 |
+
serial_block = _VisualBlock("serial", [est], dash_wrapped=False)
|
| 176 |
+
_write_estimator_html(out, serial_block, name, name_details)
|
| 177 |
+
out.write("</div>") # sk-parallel-item
|
| 178 |
+
|
| 179 |
+
out.write("</div></div>")
|
| 180 |
+
elif est_block.kind == "single":
|
| 181 |
+
_write_label_html(
|
| 182 |
+
out,
|
| 183 |
+
est_block.names,
|
| 184 |
+
est_block.name_details,
|
| 185 |
+
outer_class="sk-item",
|
| 186 |
+
inner_class="sk-estimator",
|
| 187 |
+
checked=first_call,
|
| 188 |
+
)
|
| 189 |
+
|
| 190 |
+
|
| 191 |
+
_STYLE = """
|
| 192 |
+
#$id {
|
| 193 |
+
color: black;
|
| 194 |
+
background-color: white;
|
| 195 |
+
}
|
| 196 |
+
#$id pre{
|
| 197 |
+
padding: 0;
|
| 198 |
+
}
|
| 199 |
+
#$id div.sk-toggleable {
|
| 200 |
+
background-color: white;
|
| 201 |
+
}
|
| 202 |
+
#$id label.sk-toggleable__label {
|
| 203 |
+
cursor: pointer;
|
| 204 |
+
display: block;
|
| 205 |
+
width: 100%;
|
| 206 |
+
margin-bottom: 0;
|
| 207 |
+
padding: 0.3em;
|
| 208 |
+
box-sizing: border-box;
|
| 209 |
+
text-align: center;
|
| 210 |
+
}
|
| 211 |
+
#$id label.sk-toggleable__label-arrow:before {
|
| 212 |
+
content: "▸";
|
| 213 |
+
float: left;
|
| 214 |
+
margin-right: 0.25em;
|
| 215 |
+
color: #696969;
|
| 216 |
+
}
|
| 217 |
+
#$id label.sk-toggleable__label-arrow:hover:before {
|
| 218 |
+
color: black;
|
| 219 |
+
}
|
| 220 |
+
#$id div.sk-estimator:hover label.sk-toggleable__label-arrow:before {
|
| 221 |
+
color: black;
|
| 222 |
+
}
|
| 223 |
+
#$id div.sk-toggleable__content {
|
| 224 |
+
max-height: 0;
|
| 225 |
+
max-width: 0;
|
| 226 |
+
overflow: hidden;
|
| 227 |
+
text-align: left;
|
| 228 |
+
background-color: #f0f8ff;
|
| 229 |
+
}
|
| 230 |
+
#$id div.sk-toggleable__content pre {
|
| 231 |
+
margin: 0.2em;
|
| 232 |
+
color: black;
|
| 233 |
+
border-radius: 0.25em;
|
| 234 |
+
background-color: #f0f8ff;
|
| 235 |
+
}
|
| 236 |
+
#$id input.sk-toggleable__control:checked~div.sk-toggleable__content {
|
| 237 |
+
max-height: 200px;
|
| 238 |
+
max-width: 100%;
|
| 239 |
+
overflow: auto;
|
| 240 |
+
}
|
| 241 |
+
#$id input.sk-toggleable__control:checked~label.sk-toggleable__label-arrow:before {
|
| 242 |
+
content: "▾";
|
| 243 |
+
}
|
| 244 |
+
#$id div.sk-estimator input.sk-toggleable__control:checked~label.sk-toggleable__label {
|
| 245 |
+
background-color: #d4ebff;
|
| 246 |
+
}
|
| 247 |
+
#$id div.sk-label input.sk-toggleable__control:checked~label.sk-toggleable__label {
|
| 248 |
+
background-color: #d4ebff;
|
| 249 |
+
}
|
| 250 |
+
#$id input.sk-hidden--visually {
|
| 251 |
+
border: 0;
|
| 252 |
+
clip: rect(1px 1px 1px 1px);
|
| 253 |
+
clip: rect(1px, 1px, 1px, 1px);
|
| 254 |
+
height: 1px;
|
| 255 |
+
margin: -1px;
|
| 256 |
+
overflow: hidden;
|
| 257 |
+
padding: 0;
|
| 258 |
+
position: absolute;
|
| 259 |
+
width: 1px;
|
| 260 |
+
}
|
| 261 |
+
#$id div.sk-estimator {
|
| 262 |
+
font-family: monospace;
|
| 263 |
+
background-color: #f0f8ff;
|
| 264 |
+
border: 1px dotted black;
|
| 265 |
+
border-radius: 0.25em;
|
| 266 |
+
box-sizing: border-box;
|
| 267 |
+
margin-bottom: 0.5em;
|
| 268 |
+
}
|
| 269 |
+
#$id div.sk-estimator:hover {
|
| 270 |
+
background-color: #d4ebff;
|
| 271 |
+
}
|
| 272 |
+
#$id div.sk-parallel-item::after {
|
| 273 |
+
content: "";
|
| 274 |
+
width: 100%;
|
| 275 |
+
border-bottom: 1px solid gray;
|
| 276 |
+
flex-grow: 1;
|
| 277 |
+
}
|
| 278 |
+
#$id div.sk-label:hover label.sk-toggleable__label {
|
| 279 |
+
background-color: #d4ebff;
|
| 280 |
+
}
|
| 281 |
+
#$id div.sk-serial::before {
|
| 282 |
+
content: "";
|
| 283 |
+
position: absolute;
|
| 284 |
+
border-left: 1px solid gray;
|
| 285 |
+
box-sizing: border-box;
|
| 286 |
+
top: 0;
|
| 287 |
+
bottom: 0;
|
| 288 |
+
left: 50%;
|
| 289 |
+
z-index: 0;
|
| 290 |
+
}
|
| 291 |
+
#$id div.sk-serial {
|
| 292 |
+
display: flex;
|
| 293 |
+
flex-direction: column;
|
| 294 |
+
align-items: center;
|
| 295 |
+
background-color: white;
|
| 296 |
+
padding-right: 0.2em;
|
| 297 |
+
padding-left: 0.2em;
|
| 298 |
+
position: relative;
|
| 299 |
+
}
|
| 300 |
+
#$id div.sk-item {
|
| 301 |
+
position: relative;
|
| 302 |
+
z-index: 1;
|
| 303 |
+
}
|
| 304 |
+
#$id div.sk-parallel {
|
| 305 |
+
display: flex;
|
| 306 |
+
align-items: stretch;
|
| 307 |
+
justify-content: center;
|
| 308 |
+
background-color: white;
|
| 309 |
+
position: relative;
|
| 310 |
+
}
|
| 311 |
+
#$id div.sk-item::before, #$id div.sk-parallel-item::before {
|
| 312 |
+
content: "";
|
| 313 |
+
position: absolute;
|
| 314 |
+
border-left: 1px solid gray;
|
| 315 |
+
box-sizing: border-box;
|
| 316 |
+
top: 0;
|
| 317 |
+
bottom: 0;
|
| 318 |
+
left: 50%;
|
| 319 |
+
z-index: -1;
|
| 320 |
+
}
|
| 321 |
+
#$id div.sk-parallel-item {
|
| 322 |
+
display: flex;
|
| 323 |
+
flex-direction: column;
|
| 324 |
+
z-index: 1;
|
| 325 |
+
position: relative;
|
| 326 |
+
background-color: white;
|
| 327 |
+
}
|
| 328 |
+
#$id div.sk-parallel-item:first-child::after {
|
| 329 |
+
align-self: flex-end;
|
| 330 |
+
width: 50%;
|
| 331 |
+
}
|
| 332 |
+
#$id div.sk-parallel-item:last-child::after {
|
| 333 |
+
align-self: flex-start;
|
| 334 |
+
width: 50%;
|
| 335 |
+
}
|
| 336 |
+
#$id div.sk-parallel-item:only-child::after {
|
| 337 |
+
width: 0;
|
| 338 |
+
}
|
| 339 |
+
#$id div.sk-dashed-wrapped {
|
| 340 |
+
border: 1px dashed gray;
|
| 341 |
+
margin: 0 0.4em 0.5em 0.4em;
|
| 342 |
+
box-sizing: border-box;
|
| 343 |
+
padding-bottom: 0.4em;
|
| 344 |
+
background-color: white;
|
| 345 |
+
}
|
| 346 |
+
#$id div.sk-label label {
|
| 347 |
+
font-family: monospace;
|
| 348 |
+
font-weight: bold;
|
| 349 |
+
display: inline-block;
|
| 350 |
+
line-height: 1.2em;
|
| 351 |
+
}
|
| 352 |
+
#$id div.sk-label-container {
|
| 353 |
+
text-align: center;
|
| 354 |
+
}
|
| 355 |
+
#$id div.sk-container {
|
| 356 |
+
/* jupyter's `normalize.less` sets `[hidden] { display: none; }`
|
| 357 |
+
but bootstrap.min.css set `[hidden] { display: none !important; }`
|
| 358 |
+
so we also need the `!important` here to be able to override the
|
| 359 |
+
default hidden behavior on the sphinx rendered scikit-learn.org.
|
| 360 |
+
See: https://github.com/scikit-learn/scikit-learn/issues/21755 */
|
| 361 |
+
display: inline-block !important;
|
| 362 |
+
position: relative;
|
| 363 |
+
}
|
| 364 |
+
#$id div.sk-text-repr-fallback {
|
| 365 |
+
display: none;
|
| 366 |
+
}
|
| 367 |
+
""".replace(
|
| 368 |
+
" ", ""
|
| 369 |
+
).replace(
|
| 370 |
+
"\n", ""
|
| 371 |
+
) # noqa
|
| 372 |
+
|
| 373 |
+
|
| 374 |
+
def estimator_html_repr(estimator):
|
| 375 |
+
"""Build a HTML representation of an estimator.
|
| 376 |
+
|
| 377 |
+
Read more in the :ref:`User Guide <visualizing_composite_estimators>`.
|
| 378 |
+
|
| 379 |
+
Parameters
|
| 380 |
+
----------
|
| 381 |
+
estimator : estimator object
|
| 382 |
+
The estimator to visualize.
|
| 383 |
+
|
| 384 |
+
Returns
|
| 385 |
+
-------
|
| 386 |
+
html: str
|
| 387 |
+
HTML representation of estimator.
|
| 388 |
+
"""
|
| 389 |
+
with closing(StringIO()) as out:
|
| 390 |
+
container_id = _CONTAINER_ID_COUNTER.get_id()
|
| 391 |
+
style_template = Template(_STYLE)
|
| 392 |
+
style_with_id = style_template.substitute(id=container_id)
|
| 393 |
+
estimator_str = str(estimator)
|
| 394 |
+
|
| 395 |
+
# The fallback message is shown by default and loading the CSS sets
|
| 396 |
+
# div.sk-text-repr-fallback to display: none to hide the fallback message.
|
| 397 |
+
#
|
| 398 |
+
# If the notebook is trusted, the CSS is loaded which hides the fallback
|
| 399 |
+
# message. If the notebook is not trusted, then the CSS is not loaded and the
|
| 400 |
+
# fallback message is shown by default.
|
| 401 |
+
#
|
| 402 |
+
# The reverse logic applies to HTML repr div.sk-container.
|
| 403 |
+
# div.sk-container is hidden by default and the loading the CSS displays it.
|
| 404 |
+
fallback_msg = (
|
| 405 |
+
"In a Jupyter environment, please rerun this cell to show the HTML"
|
| 406 |
+
" representation or trust the notebook. <br />On GitHub, the"
|
| 407 |
+
" HTML representation is unable to render, please try loading this page"
|
| 408 |
+
" with nbviewer.org."
|
| 409 |
+
)
|
| 410 |
+
out.write(
|
| 411 |
+
f"<style>{style_with_id}</style>"
|
| 412 |
+
f'<div id="{container_id}" class="sk-top-container">'
|
| 413 |
+
'<div class="sk-text-repr-fallback">'
|
| 414 |
+
f"<pre>{html.escape(estimator_str)}</pre><b>{fallback_msg}</b>"
|
| 415 |
+
"</div>"
|
| 416 |
+
'<div class="sk-container" hidden>'
|
| 417 |
+
)
|
| 418 |
+
_write_estimator_html(
|
| 419 |
+
out,
|
| 420 |
+
estimator,
|
| 421 |
+
estimator.__class__.__name__,
|
| 422 |
+
estimator_str,
|
| 423 |
+
first_call=True,
|
| 424 |
+
)
|
| 425 |
+
out.write("</div></div>")
|
| 426 |
+
|
| 427 |
+
html_output = out.getvalue()
|
| 428 |
+
return html_output
|
videollama2/lib/python3.10/site-packages/sklearn/utils/_fast_dict.pxd
ADDED
|
@@ -0,0 +1,22 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Author: Gael Varoquaux
|
| 2 |
+
# License: BSD
|
| 3 |
+
"""
|
| 4 |
+
Uses C++ map containers for fast dict-like behavior with keys being
|
| 5 |
+
integers, and values float.
|
| 6 |
+
"""
|
| 7 |
+
|
| 8 |
+
from libcpp.map cimport map as cpp_map
|
| 9 |
+
|
| 10 |
+
# Import the C-level symbols of numpy
|
| 11 |
+
cimport numpy as cnp
|
| 12 |
+
|
| 13 |
+
ctypedef cnp.float64_t DTYPE_t
|
| 14 |
+
|
| 15 |
+
ctypedef cnp.intp_t ITYPE_t
|
| 16 |
+
|
| 17 |
+
###############################################################################
|
| 18 |
+
# An object to be used in Python
|
| 19 |
+
|
| 20 |
+
cdef class IntFloatDict:
|
| 21 |
+
cdef cpp_map[ITYPE_t, DTYPE_t] my_map
|
| 22 |
+
cdef _to_arrays(self, ITYPE_t [:] keys, DTYPE_t [:] values)
|
videollama2/lib/python3.10/site-packages/sklearn/utils/_isfinite.cpython-310-x86_64-linux-gnu.so
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:6297f649f6414e2131de7a03bcbf7cea238832bb3bd1a13bfce9469a7aa2346d
|
| 3 |
+
size 253625
|
videollama2/lib/python3.10/site-packages/sklearn/utils/_random.pxd
ADDED
|
@@ -0,0 +1,39 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Authors: Arnaud Joly
|
| 2 |
+
#
|
| 3 |
+
# License: BSD 3 clause
|
| 4 |
+
|
| 5 |
+
|
| 6 |
+
cimport numpy as cnp
|
| 7 |
+
ctypedef cnp.npy_uint32 UINT32_t
|
| 8 |
+
|
| 9 |
+
cdef inline UINT32_t DEFAULT_SEED = 1
|
| 10 |
+
|
| 11 |
+
cdef enum:
|
| 12 |
+
# Max value for our rand_r replacement (near the bottom).
|
| 13 |
+
# We don't use RAND_MAX because it's different across platforms and
|
| 14 |
+
# particularly tiny on Windows/MSVC.
|
| 15 |
+
# It corresponds to the maximum representable value for
|
| 16 |
+
# 32-bit signed integers (i.e. 2^31 - 1).
|
| 17 |
+
RAND_R_MAX = 0x7FFFFFFF
|
| 18 |
+
|
| 19 |
+
cpdef sample_without_replacement(cnp.int_t n_population,
|
| 20 |
+
cnp.int_t n_samples,
|
| 21 |
+
method=*,
|
| 22 |
+
random_state=*)
|
| 23 |
+
|
| 24 |
+
# rand_r replacement using a 32bit XorShift generator
|
| 25 |
+
# See http://www.jstatsoft.org/v08/i14/paper for details
|
| 26 |
+
cdef inline UINT32_t our_rand_r(UINT32_t* seed) nogil:
|
| 27 |
+
"""Generate a pseudo-random np.uint32 from a np.uint32 seed"""
|
| 28 |
+
# seed shouldn't ever be 0.
|
| 29 |
+
if (seed[0] == 0): seed[0] = DEFAULT_SEED
|
| 30 |
+
|
| 31 |
+
seed[0] ^= <UINT32_t>(seed[0] << 13)
|
| 32 |
+
seed[0] ^= <UINT32_t>(seed[0] >> 17)
|
| 33 |
+
seed[0] ^= <UINT32_t>(seed[0] << 5)
|
| 34 |
+
|
| 35 |
+
# Use the modulo to make sure that we don't return a values greater than the
|
| 36 |
+
# maximum representable value for signed 32bit integers (i.e. 2^31 - 1).
|
| 37 |
+
# Note that the parenthesis are needed to avoid overflow: here
|
| 38 |
+
# RAND_R_MAX is cast to UINT32_t before 1 is added.
|
| 39 |
+
return seed[0] % ((<UINT32_t>RAND_R_MAX) + 1)
|
videollama2/lib/python3.10/site-packages/sklearn/utils/_seq_dataset.pxd
ADDED
|
@@ -0,0 +1,116 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
|
| 2 |
+
#------------------------------------------------------------------------------
|
| 3 |
+
|
| 4 |
+
"""
|
| 5 |
+
Dataset abstractions for sequential data access.
|
| 6 |
+
WARNING: Do not edit .pxd file directly, it is generated from .pxd.tp
|
| 7 |
+
"""
|
| 8 |
+
|
| 9 |
+
cimport numpy as cnp
|
| 10 |
+
|
| 11 |
+
# SequentialDataset and its two concrete subclasses are (optionally randomized)
|
| 12 |
+
# iterators over the rows of a matrix X and corresponding target values y.
|
| 13 |
+
|
| 14 |
+
|
| 15 |
+
cdef class SequentialDataset64:
|
| 16 |
+
cdef int current_index
|
| 17 |
+
cdef cnp.ndarray index
|
| 18 |
+
cdef int *index_data_ptr
|
| 19 |
+
cdef Py_ssize_t n_samples
|
| 20 |
+
cdef cnp.uint32_t seed
|
| 21 |
+
|
| 22 |
+
cdef void shuffle(self, cnp.uint32_t seed) nogil
|
| 23 |
+
cdef int _get_next_index(self) nogil
|
| 24 |
+
cdef int _get_random_index(self) nogil
|
| 25 |
+
|
| 26 |
+
cdef void _sample(self, double **x_data_ptr, int **x_ind_ptr,
|
| 27 |
+
int *nnz, double *y, double *sample_weight,
|
| 28 |
+
int current_index) nogil
|
| 29 |
+
cdef void next(self, double **x_data_ptr, int **x_ind_ptr,
|
| 30 |
+
int *nnz, double *y, double *sample_weight) nogil
|
| 31 |
+
cdef int random(self, double **x_data_ptr, int **x_ind_ptr,
|
| 32 |
+
int *nnz, double *y, double *sample_weight) nogil
|
| 33 |
+
|
| 34 |
+
|
| 35 |
+
cdef class ArrayDataset64(SequentialDataset64):
|
| 36 |
+
cdef cnp.ndarray X
|
| 37 |
+
cdef cnp.ndarray Y
|
| 38 |
+
cdef cnp.ndarray sample_weights
|
| 39 |
+
cdef Py_ssize_t n_features
|
| 40 |
+
cdef cnp.npy_intp X_stride
|
| 41 |
+
cdef double *X_data_ptr
|
| 42 |
+
cdef double *Y_data_ptr
|
| 43 |
+
cdef cnp.ndarray feature_indices
|
| 44 |
+
cdef int *feature_indices_ptr
|
| 45 |
+
cdef double *sample_weight_data
|
| 46 |
+
|
| 47 |
+
|
| 48 |
+
cdef class CSRDataset64(SequentialDataset64):
|
| 49 |
+
cdef cnp.ndarray X_data
|
| 50 |
+
cdef cnp.ndarray X_indptr
|
| 51 |
+
cdef cnp.ndarray X_indices
|
| 52 |
+
cdef cnp.ndarray Y
|
| 53 |
+
cdef cnp.ndarray sample_weights
|
| 54 |
+
cdef double *X_data_ptr
|
| 55 |
+
cdef int *X_indptr_ptr
|
| 56 |
+
cdef int *X_indices_ptr
|
| 57 |
+
cdef double *Y_data_ptr
|
| 58 |
+
cdef double *sample_weight_data
|
| 59 |
+
|
| 60 |
+
#------------------------------------------------------------------------------
|
| 61 |
+
|
| 62 |
+
"""
|
| 63 |
+
Dataset abstractions for sequential data access.
|
| 64 |
+
WARNING: Do not edit .pxd file directly, it is generated from .pxd.tp
|
| 65 |
+
"""
|
| 66 |
+
|
| 67 |
+
cimport numpy as cnp
|
| 68 |
+
|
| 69 |
+
# SequentialDataset and its two concrete subclasses are (optionally randomized)
|
| 70 |
+
# iterators over the rows of a matrix X and corresponding target values y.
|
| 71 |
+
|
| 72 |
+
|
| 73 |
+
cdef class SequentialDataset32:
|
| 74 |
+
cdef int current_index
|
| 75 |
+
cdef cnp.ndarray index
|
| 76 |
+
cdef int *index_data_ptr
|
| 77 |
+
cdef Py_ssize_t n_samples
|
| 78 |
+
cdef cnp.uint32_t seed
|
| 79 |
+
|
| 80 |
+
cdef void shuffle(self, cnp.uint32_t seed) nogil
|
| 81 |
+
cdef int _get_next_index(self) nogil
|
| 82 |
+
cdef int _get_random_index(self) nogil
|
| 83 |
+
|
| 84 |
+
cdef void _sample(self, float **x_data_ptr, int **x_ind_ptr,
|
| 85 |
+
int *nnz, float *y, float *sample_weight,
|
| 86 |
+
int current_index) nogil
|
| 87 |
+
cdef void next(self, float **x_data_ptr, int **x_ind_ptr,
|
| 88 |
+
int *nnz, float *y, float *sample_weight) nogil
|
| 89 |
+
cdef int random(self, float **x_data_ptr, int **x_ind_ptr,
|
| 90 |
+
int *nnz, float *y, float *sample_weight) nogil
|
| 91 |
+
|
| 92 |
+
|
| 93 |
+
cdef class ArrayDataset32(SequentialDataset32):
|
| 94 |
+
cdef cnp.ndarray X
|
| 95 |
+
cdef cnp.ndarray Y
|
| 96 |
+
cdef cnp.ndarray sample_weights
|
| 97 |
+
cdef Py_ssize_t n_features
|
| 98 |
+
cdef cnp.npy_intp X_stride
|
| 99 |
+
cdef float *X_data_ptr
|
| 100 |
+
cdef float *Y_data_ptr
|
| 101 |
+
cdef cnp.ndarray feature_indices
|
| 102 |
+
cdef int *feature_indices_ptr
|
| 103 |
+
cdef float *sample_weight_data
|
| 104 |
+
|
| 105 |
+
|
| 106 |
+
cdef class CSRDataset32(SequentialDataset32):
|
| 107 |
+
cdef cnp.ndarray X_data
|
| 108 |
+
cdef cnp.ndarray X_indptr
|
| 109 |
+
cdef cnp.ndarray X_indices
|
| 110 |
+
cdef cnp.ndarray Y
|
| 111 |
+
cdef cnp.ndarray sample_weights
|
| 112 |
+
cdef float *X_data_ptr
|
| 113 |
+
cdef int *X_indptr_ptr
|
| 114 |
+
cdef int *X_indices_ptr
|
| 115 |
+
cdef float *Y_data_ptr
|
| 116 |
+
cdef float *sample_weight_data
|
videollama2/lib/python3.10/site-packages/sklearn/utils/_sorting.pxd
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from ._typedefs cimport DTYPE_t, ITYPE_t
|
| 2 |
+
|
| 3 |
+
from cython cimport floating
|
| 4 |
+
|
| 5 |
+
cdef int simultaneous_sort(
|
| 6 |
+
floating *dist,
|
| 7 |
+
ITYPE_t *idx,
|
| 8 |
+
ITYPE_t size,
|
| 9 |
+
) nogil
|
videollama2/lib/python3.10/site-packages/sklearn/utils/_typedefs.cpython-310-x86_64-linux-gnu.so
ADDED
|
Binary file (38.7 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/_vector_sentinel.cpython-310-x86_64-linux-gnu.so
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:3ad4d62de9cdbea5d9f4005dd26d8b9c59939b60a2f0ee27bd663a92ae288774
|
| 3 |
+
size 145705
|
videollama2/lib/python3.10/site-packages/sklearn/utils/deprecation.py
ADDED
|
@@ -0,0 +1,115 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import warnings
|
| 2 |
+
import functools
|
| 3 |
+
|
| 4 |
+
|
| 5 |
+
__all__ = ["deprecated"]
|
| 6 |
+
|
| 7 |
+
|
| 8 |
+
class deprecated:
|
| 9 |
+
"""Decorator to mark a function or class as deprecated.
|
| 10 |
+
|
| 11 |
+
Issue a warning when the function is called/the class is instantiated and
|
| 12 |
+
adds a warning to the docstring.
|
| 13 |
+
|
| 14 |
+
The optional extra argument will be appended to the deprecation message
|
| 15 |
+
and the docstring. Note: to use this with the default value for extra, put
|
| 16 |
+
in an empty of parentheses:
|
| 17 |
+
|
| 18 |
+
>>> from sklearn.utils import deprecated
|
| 19 |
+
>>> deprecated()
|
| 20 |
+
<sklearn.utils.deprecation.deprecated object at ...>
|
| 21 |
+
|
| 22 |
+
>>> @deprecated()
|
| 23 |
+
... def some_function(): pass
|
| 24 |
+
|
| 25 |
+
Parameters
|
| 26 |
+
----------
|
| 27 |
+
extra : str, default=''
|
| 28 |
+
To be added to the deprecation messages.
|
| 29 |
+
"""
|
| 30 |
+
|
| 31 |
+
# Adapted from https://wiki.python.org/moin/PythonDecoratorLibrary,
|
| 32 |
+
# but with many changes.
|
| 33 |
+
|
| 34 |
+
def __init__(self, extra=""):
|
| 35 |
+
self.extra = extra
|
| 36 |
+
|
| 37 |
+
def __call__(self, obj):
|
| 38 |
+
"""Call method
|
| 39 |
+
|
| 40 |
+
Parameters
|
| 41 |
+
----------
|
| 42 |
+
obj : object
|
| 43 |
+
"""
|
| 44 |
+
if isinstance(obj, type):
|
| 45 |
+
return self._decorate_class(obj)
|
| 46 |
+
elif isinstance(obj, property):
|
| 47 |
+
# Note that this is only triggered properly if the `property`
|
| 48 |
+
# decorator comes before the `deprecated` decorator, like so:
|
| 49 |
+
#
|
| 50 |
+
# @deprecated(msg)
|
| 51 |
+
# @property
|
| 52 |
+
# def deprecated_attribute_(self):
|
| 53 |
+
# ...
|
| 54 |
+
return self._decorate_property(obj)
|
| 55 |
+
else:
|
| 56 |
+
return self._decorate_fun(obj)
|
| 57 |
+
|
| 58 |
+
def _decorate_class(self, cls):
|
| 59 |
+
msg = "Class %s is deprecated" % cls.__name__
|
| 60 |
+
if self.extra:
|
| 61 |
+
msg += "; %s" % self.extra
|
| 62 |
+
|
| 63 |
+
# FIXME: we should probably reset __new__ for full generality
|
| 64 |
+
init = cls.__init__
|
| 65 |
+
|
| 66 |
+
def wrapped(*args, **kwargs):
|
| 67 |
+
warnings.warn(msg, category=FutureWarning)
|
| 68 |
+
return init(*args, **kwargs)
|
| 69 |
+
|
| 70 |
+
cls.__init__ = wrapped
|
| 71 |
+
|
| 72 |
+
wrapped.__name__ = "__init__"
|
| 73 |
+
wrapped.deprecated_original = init
|
| 74 |
+
|
| 75 |
+
return cls
|
| 76 |
+
|
| 77 |
+
def _decorate_fun(self, fun):
|
| 78 |
+
"""Decorate function fun"""
|
| 79 |
+
|
| 80 |
+
msg = "Function %s is deprecated" % fun.__name__
|
| 81 |
+
if self.extra:
|
| 82 |
+
msg += "; %s" % self.extra
|
| 83 |
+
|
| 84 |
+
@functools.wraps(fun)
|
| 85 |
+
def wrapped(*args, **kwargs):
|
| 86 |
+
warnings.warn(msg, category=FutureWarning)
|
| 87 |
+
return fun(*args, **kwargs)
|
| 88 |
+
|
| 89 |
+
# Add a reference to the wrapped function so that we can introspect
|
| 90 |
+
# on function arguments in Python 2 (already works in Python 3)
|
| 91 |
+
wrapped.__wrapped__ = fun
|
| 92 |
+
|
| 93 |
+
return wrapped
|
| 94 |
+
|
| 95 |
+
def _decorate_property(self, prop):
|
| 96 |
+
msg = self.extra
|
| 97 |
+
|
| 98 |
+
@property
|
| 99 |
+
@functools.wraps(prop)
|
| 100 |
+
def wrapped(*args, **kwargs):
|
| 101 |
+
warnings.warn(msg, category=FutureWarning)
|
| 102 |
+
return prop.fget(*args, **kwargs)
|
| 103 |
+
|
| 104 |
+
return wrapped
|
| 105 |
+
|
| 106 |
+
|
| 107 |
+
def _is_deprecated(func):
|
| 108 |
+
"""Helper to check if func is wrapped by our deprecated decorator"""
|
| 109 |
+
closures = getattr(func, "__closure__", [])
|
| 110 |
+
if closures is None:
|
| 111 |
+
closures = []
|
| 112 |
+
is_deprecated = "deprecated" in "".join(
|
| 113 |
+
[c.cell_contents for c in closures if isinstance(c.cell_contents, str)]
|
| 114 |
+
)
|
| 115 |
+
return is_deprecated
|
videollama2/lib/python3.10/site-packages/sklearn/utils/fixes.py
ADDED
|
@@ -0,0 +1,216 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""Compatibility fixes for older version of python, numpy and scipy
|
| 2 |
+
|
| 3 |
+
If you add content to this file, please give the version of the package
|
| 4 |
+
at which the fix is no longer needed.
|
| 5 |
+
"""
|
| 6 |
+
# Authors: Emmanuelle Gouillart <emmanuelle.gouillart@normalesup.org>
|
| 7 |
+
# Gael Varoquaux <gael.varoquaux@normalesup.org>
|
| 8 |
+
# Fabian Pedregosa <fpedregosa@acm.org>
|
| 9 |
+
# Lars Buitinck
|
| 10 |
+
#
|
| 11 |
+
# License: BSD 3 clause
|
| 12 |
+
|
| 13 |
+
from importlib import resources
|
| 14 |
+
import sys
|
| 15 |
+
|
| 16 |
+
import sklearn
|
| 17 |
+
import numpy as np
|
| 18 |
+
import scipy
|
| 19 |
+
import scipy.stats
|
| 20 |
+
import threadpoolctl
|
| 21 |
+
|
| 22 |
+
from .deprecation import deprecated
|
| 23 |
+
from ..externals._packaging.version import parse as parse_version
|
| 24 |
+
|
| 25 |
+
|
| 26 |
+
np_version = parse_version(np.__version__)
|
| 27 |
+
sp_version = parse_version(scipy.__version__)
|
| 28 |
+
|
| 29 |
+
|
| 30 |
+
if sp_version >= parse_version("1.4"):
|
| 31 |
+
from scipy.sparse.linalg import lobpcg
|
| 32 |
+
else:
|
| 33 |
+
# Backport of lobpcg functionality from scipy 1.4.0, can be removed
|
| 34 |
+
# once support for sp_version < parse_version('1.4') is dropped
|
| 35 |
+
# mypy error: Name 'lobpcg' already defined (possibly by an import)
|
| 36 |
+
from ..externals._lobpcg import lobpcg # type: ignore # noqa
|
| 37 |
+
|
| 38 |
+
try:
|
| 39 |
+
from scipy.optimize._linesearch import line_search_wolfe2, line_search_wolfe1
|
| 40 |
+
except ImportError: # SciPy < 1.8
|
| 41 |
+
from scipy.optimize.linesearch import line_search_wolfe2, line_search_wolfe1 # type: ignore # noqa
|
| 42 |
+
|
| 43 |
+
|
| 44 |
+
def _object_dtype_isnan(X):
|
| 45 |
+
return X != X
|
| 46 |
+
|
| 47 |
+
|
| 48 |
+
class loguniform(scipy.stats.reciprocal):
|
| 49 |
+
"""A class supporting log-uniform random variables.
|
| 50 |
+
|
| 51 |
+
Parameters
|
| 52 |
+
----------
|
| 53 |
+
low : float
|
| 54 |
+
The minimum value
|
| 55 |
+
high : float
|
| 56 |
+
The maximum value
|
| 57 |
+
|
| 58 |
+
Methods
|
| 59 |
+
-------
|
| 60 |
+
rvs(self, size=None, random_state=None)
|
| 61 |
+
Generate log-uniform random variables
|
| 62 |
+
|
| 63 |
+
The most useful method for Scikit-learn usage is highlighted here.
|
| 64 |
+
For a full list, see
|
| 65 |
+
`scipy.stats.reciprocal
|
| 66 |
+
<https://docs.scipy.org/doc/scipy/reference/generated/scipy.stats.reciprocal.html>`_.
|
| 67 |
+
This list includes all functions of ``scipy.stats`` continuous
|
| 68 |
+
distributions such as ``pdf``.
|
| 69 |
+
|
| 70 |
+
Notes
|
| 71 |
+
-----
|
| 72 |
+
This class generates values between ``low`` and ``high`` or
|
| 73 |
+
|
| 74 |
+
low <= loguniform(low, high).rvs() <= high
|
| 75 |
+
|
| 76 |
+
The logarithmic probability density function (PDF) is uniform. When
|
| 77 |
+
``x`` is a uniformly distributed random variable between 0 and 1, ``10**x``
|
| 78 |
+
are random variables that are equally likely to be returned.
|
| 79 |
+
|
| 80 |
+
This class is an alias to ``scipy.stats.reciprocal``, which uses the
|
| 81 |
+
reciprocal distribution:
|
| 82 |
+
https://en.wikipedia.org/wiki/Reciprocal_distribution
|
| 83 |
+
|
| 84 |
+
Examples
|
| 85 |
+
--------
|
| 86 |
+
|
| 87 |
+
>>> from sklearn.utils.fixes import loguniform
|
| 88 |
+
>>> rv = loguniform(1e-3, 1e1)
|
| 89 |
+
>>> rvs = rv.rvs(random_state=42, size=1000)
|
| 90 |
+
>>> rvs.min() # doctest: +SKIP
|
| 91 |
+
0.0010435856341129003
|
| 92 |
+
>>> rvs.max() # doctest: +SKIP
|
| 93 |
+
9.97403052786026
|
| 94 |
+
"""
|
| 95 |
+
|
| 96 |
+
|
| 97 |
+
# TODO: remove when the minimum scipy version is >= 1.5
|
| 98 |
+
if sp_version >= parse_version("1.5"):
|
| 99 |
+
from scipy.linalg import eigh as _eigh # noqa
|
| 100 |
+
else:
|
| 101 |
+
|
| 102 |
+
def _eigh(*args, **kwargs):
|
| 103 |
+
"""Wrapper for `scipy.linalg.eigh` that handles the deprecation of `eigvals`."""
|
| 104 |
+
eigvals = kwargs.pop("subset_by_index", None)
|
| 105 |
+
return scipy.linalg.eigh(*args, eigvals=eigvals, **kwargs)
|
| 106 |
+
|
| 107 |
+
|
| 108 |
+
# Rename the `method` kwarg to `interpolation` for NumPy < 1.22, because
|
| 109 |
+
# `interpolation` kwarg was deprecated in favor of `method` in NumPy >= 1.22.
|
| 110 |
+
def _percentile(a, q, *, method="linear", **kwargs):
|
| 111 |
+
return np.percentile(a, q, interpolation=method, **kwargs)
|
| 112 |
+
|
| 113 |
+
|
| 114 |
+
if np_version < parse_version("1.22"):
|
| 115 |
+
percentile = _percentile
|
| 116 |
+
else: # >= 1.22
|
| 117 |
+
from numpy import percentile # type: ignore # noqa
|
| 118 |
+
|
| 119 |
+
|
| 120 |
+
# compatibility fix for threadpoolctl >= 3.0.0
|
| 121 |
+
# since version 3 it's possible to setup a global threadpool controller to avoid
|
| 122 |
+
# looping through all loaded shared libraries each time.
|
| 123 |
+
# the global controller is created during the first call to threadpoolctl.
|
| 124 |
+
def _get_threadpool_controller():
|
| 125 |
+
if not hasattr(threadpoolctl, "ThreadpoolController"):
|
| 126 |
+
return None
|
| 127 |
+
|
| 128 |
+
if not hasattr(sklearn, "_sklearn_threadpool_controller"):
|
| 129 |
+
sklearn._sklearn_threadpool_controller = threadpoolctl.ThreadpoolController()
|
| 130 |
+
|
| 131 |
+
return sklearn._sklearn_threadpool_controller
|
| 132 |
+
|
| 133 |
+
|
| 134 |
+
def threadpool_limits(limits=None, user_api=None):
|
| 135 |
+
controller = _get_threadpool_controller()
|
| 136 |
+
if controller is not None:
|
| 137 |
+
return controller.limit(limits=limits, user_api=user_api)
|
| 138 |
+
else:
|
| 139 |
+
return threadpoolctl.threadpool_limits(limits=limits, user_api=user_api)
|
| 140 |
+
|
| 141 |
+
|
| 142 |
+
threadpool_limits.__doc__ = threadpoolctl.threadpool_limits.__doc__
|
| 143 |
+
|
| 144 |
+
|
| 145 |
+
def threadpool_info():
|
| 146 |
+
controller = _get_threadpool_controller()
|
| 147 |
+
if controller is not None:
|
| 148 |
+
return controller.info()
|
| 149 |
+
else:
|
| 150 |
+
return threadpoolctl.threadpool_info()
|
| 151 |
+
|
| 152 |
+
|
| 153 |
+
threadpool_info.__doc__ = threadpoolctl.threadpool_info.__doc__
|
| 154 |
+
|
| 155 |
+
|
| 156 |
+
@deprecated(
|
| 157 |
+
"The function `delayed` has been moved from `sklearn.utils.fixes` to "
|
| 158 |
+
"`sklearn.utils.parallel`. This import path will be removed in 1.5."
|
| 159 |
+
)
|
| 160 |
+
def delayed(function):
|
| 161 |
+
from sklearn.utils.parallel import delayed
|
| 162 |
+
|
| 163 |
+
return delayed(function)
|
| 164 |
+
|
| 165 |
+
|
| 166 |
+
# TODO: Remove when SciPy 1.11 is the minimum supported version
|
| 167 |
+
def _mode(a, axis=0):
|
| 168 |
+
if sp_version >= parse_version("1.9.0"):
|
| 169 |
+
mode = scipy.stats.mode(a, axis=axis, keepdims=True)
|
| 170 |
+
if sp_version >= parse_version("1.10.999"):
|
| 171 |
+
# scipy.stats.mode has changed returned array shape with axis=None
|
| 172 |
+
# and keepdims=True, see https://github.com/scipy/scipy/pull/17561
|
| 173 |
+
if axis is None:
|
| 174 |
+
mode = np.ravel(mode)
|
| 175 |
+
return mode
|
| 176 |
+
return scipy.stats.mode(a, axis=axis)
|
| 177 |
+
|
| 178 |
+
|
| 179 |
+
###############################################################################
|
| 180 |
+
# Backport of Python 3.9's importlib.resources
|
| 181 |
+
# TODO: Remove when Python 3.9 is the minimum supported version
|
| 182 |
+
|
| 183 |
+
|
| 184 |
+
def _open_text(data_module, data_file_name):
|
| 185 |
+
if sys.version_info >= (3, 9):
|
| 186 |
+
return resources.files(data_module).joinpath(data_file_name).open("r")
|
| 187 |
+
else:
|
| 188 |
+
return resources.open_text(data_module, data_file_name)
|
| 189 |
+
|
| 190 |
+
|
| 191 |
+
def _open_binary(data_module, data_file_name):
|
| 192 |
+
if sys.version_info >= (3, 9):
|
| 193 |
+
return resources.files(data_module).joinpath(data_file_name).open("rb")
|
| 194 |
+
else:
|
| 195 |
+
return resources.open_binary(data_module, data_file_name)
|
| 196 |
+
|
| 197 |
+
|
| 198 |
+
def _read_text(descr_module, descr_file_name):
|
| 199 |
+
if sys.version_info >= (3, 9):
|
| 200 |
+
return resources.files(descr_module).joinpath(descr_file_name).read_text()
|
| 201 |
+
else:
|
| 202 |
+
return resources.read_text(descr_module, descr_file_name)
|
| 203 |
+
|
| 204 |
+
|
| 205 |
+
def _path(data_module, data_file_name):
|
| 206 |
+
if sys.version_info >= (3, 9):
|
| 207 |
+
return resources.as_file(resources.files(data_module).joinpath(data_file_name))
|
| 208 |
+
else:
|
| 209 |
+
return resources.path(data_module, data_file_name)
|
| 210 |
+
|
| 211 |
+
|
| 212 |
+
def _is_resource(data_module, data_file_name):
|
| 213 |
+
if sys.version_info >= (3, 9):
|
| 214 |
+
return resources.files(data_module).joinpath(data_file_name).is_file()
|
| 215 |
+
else:
|
| 216 |
+
return resources.is_resource(data_module, data_file_name)
|
videollama2/lib/python3.10/site-packages/sklearn/utils/tests/__pycache__/__init__.cpython-310.pyc
ADDED
|
Binary file (176 Bytes). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/tests/__pycache__/conftest.cpython-310.pyc
ADDED
|
Binary file (425 Bytes). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/tests/__pycache__/test_array_api.cpython-310.pyc
ADDED
|
Binary file (6.77 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/tests/__pycache__/test_arrayfuncs.cpython-310.pyc
ADDED
|
Binary file (900 Bytes). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/tests/__pycache__/test_class_weight.cpython-310.pyc
ADDED
|
Binary file (7.93 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/tests/__pycache__/test_cython_blas.cpython-310.pyc
ADDED
|
Binary file (6.12 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/tests/__pycache__/test_cython_templating.cpython-310.pyc
ADDED
|
Binary file (1.04 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/tests/__pycache__/test_deprecation.cpython-310.pyc
ADDED
|
Binary file (2.3 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/tests/__pycache__/test_encode.cpython-310.pyc
ADDED
|
Binary file (6.42 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/tests/__pycache__/test_estimator_checks.cpython-310.pyc
ADDED
|
Binary file (41.6 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/tests/__pycache__/test_estimator_html_repr.cpython-310.pyc
ADDED
|
Binary file (10.4 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/tests/__pycache__/test_fast_dict.cpython-310.pyc
ADDED
|
Binary file (1.12 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/tests/__pycache__/test_mocking.cpython-310.pyc
ADDED
|
Binary file (4.94 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/tests/__pycache__/test_murmurhash.cpython-310.pyc
ADDED
|
Binary file (2.55 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/tests/__pycache__/test_optimize.cpython-310.pyc
ADDED
|
Binary file (1.48 kB). View file
|
|
|