Add files using upload-large-folder tool
Browse filesThis view is limited to 50 files because it contains too many changes. See raw diff
- videollama2/lib/python3.10/site-packages/sklearn/metrics/__init__.py +181 -0
- videollama2/lib/python3.10/site-packages/sklearn/metrics/_classification.py +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/metrics/_pairwise_distances_reduction/__pycache__/__init__.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/metrics/_pairwise_distances_reduction/__pycache__/_dispatcher.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/metrics/_pairwise_distances_reduction/_argkmin.pxd +56 -0
- videollama2/lib/python3.10/site-packages/sklearn/metrics/_plot/__init__.py +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/metrics/_plot/__pycache__/roc_curve.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/metrics/_plot/base.py +116 -0
- videollama2/lib/python3.10/site-packages/sklearn/metrics/_plot/confusion_matrix.py +482 -0
- videollama2/lib/python3.10/site-packages/sklearn/metrics/_plot/det_curve.py +346 -0
- videollama2/lib/python3.10/site-packages/sklearn/metrics/_plot/precision_recall_curve.py +393 -0
- videollama2/lib/python3.10/site-packages/sklearn/metrics/_plot/regression.py +406 -0
- videollama2/lib/python3.10/site-packages/sklearn/metrics/_plot/roc_curve.py +352 -0
- videollama2/lib/python3.10/site-packages/sklearn/metrics/_plot/tests/__init__.py +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/metrics/_plot/tests/test_common_curve_display.py +152 -0
- videollama2/lib/python3.10/site-packages/sklearn/metrics/_plot/tests/test_confusion_matrix_display.py +407 -0
- videollama2/lib/python3.10/site-packages/sklearn/metrics/_plot/tests/test_precision_recall_display.py +292 -0
- videollama2/lib/python3.10/site-packages/sklearn/metrics/_plot/tests/test_predict_error_display.py +163 -0
- videollama2/lib/python3.10/site-packages/sklearn/metrics/_regression.py +1497 -0
- videollama2/lib/python3.10/site-packages/sklearn/metrics/_scorer.py +867 -0
- videollama2/lib/python3.10/site-packages/sklearn/metrics/pairwise.py +2205 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/__init__.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/_arpack.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/_array_api.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/_bunch.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/_estimator_html_repr.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/_joblib.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/_param_validation.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/_pprint.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/class_weight.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/discovery.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/metaestimators.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/parallel.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/random.cpython-310.pyc +0 -0
- videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/sparsefuncs.cpython-310.pyc +0 -0
- vllm/lib/python3.10/site-packages/torch/include/ATen/ops/_assert_async_cpu_dispatch.h +24 -0
- vllm/lib/python3.10/site-packages/torch/include/ATen/ops/_cast_Char_compositeimplicitautograd_dispatch.h +23 -0
- vllm/lib/python3.10/site-packages/torch/include/ATen/ops/_copy_from_native.h +21 -0
- vllm/lib/python3.10/site-packages/torch/include/ATen/ops/_embedding_bag_ops.h +39 -0
- vllm/lib/python3.10/site-packages/torch/include/ATen/ops/_foreach_tanh_ops.h +50 -0
- vllm/lib/python3.10/site-packages/torch/include/ATen/ops/_sobol_engine_draw_ops.h +28 -0
- vllm/lib/python3.10/site-packages/torch/include/ATen/ops/_sparse_semi_structured_tile_ops.h +28 -0
- vllm/lib/python3.10/site-packages/torch/include/ATen/ops/_thnn_fused_lstm_cell_cuda_dispatch.h +23 -0
- vllm/lib/python3.10/site-packages/torch/include/ATen/ops/arccosh_native.h +23 -0
- vllm/lib/python3.10/site-packages/torch/include/ATen/ops/batch_norm_update_stats.h +39 -0
- vllm/lib/python3.10/site-packages/torch/include/ATen/ops/binary_cross_entropy_with_logits_compositeexplicitautograd_dispatch.h +25 -0
- vllm/lib/python3.10/site-packages/torch/include/ATen/ops/conv_transpose3d_compositeimplicitautograd_dispatch.h +24 -0
- vllm/lib/python3.10/site-packages/torch/include/ATen/ops/corrcoef_native.h +21 -0
- vllm/lib/python3.10/site-packages/torch/include/ATen/ops/cudnn_convolution_add_relu.h +91 -0
- vllm/lib/python3.10/site-packages/torch/include/ATen/ops/deg2rad.h +44 -0
videollama2/lib/python3.10/site-packages/sklearn/metrics/__init__.py
ADDED
|
@@ -0,0 +1,181 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""
|
| 2 |
+
The :mod:`sklearn.metrics` module includes score functions, performance metrics
|
| 3 |
+
and pairwise metrics and distance computations.
|
| 4 |
+
"""
|
| 5 |
+
|
| 6 |
+
|
| 7 |
+
from ._ranking import auc
|
| 8 |
+
from ._ranking import average_precision_score
|
| 9 |
+
from ._ranking import coverage_error
|
| 10 |
+
from ._ranking import det_curve
|
| 11 |
+
from ._ranking import dcg_score
|
| 12 |
+
from ._ranking import label_ranking_average_precision_score
|
| 13 |
+
from ._ranking import label_ranking_loss
|
| 14 |
+
from ._ranking import ndcg_score
|
| 15 |
+
from ._ranking import precision_recall_curve
|
| 16 |
+
from ._ranking import roc_auc_score
|
| 17 |
+
from ._ranking import roc_curve
|
| 18 |
+
from ._ranking import top_k_accuracy_score
|
| 19 |
+
|
| 20 |
+
from ._classification import accuracy_score
|
| 21 |
+
from ._classification import balanced_accuracy_score
|
| 22 |
+
from ._classification import class_likelihood_ratios
|
| 23 |
+
from ._classification import classification_report
|
| 24 |
+
from ._classification import cohen_kappa_score
|
| 25 |
+
from ._classification import confusion_matrix
|
| 26 |
+
from ._classification import f1_score
|
| 27 |
+
from ._classification import fbeta_score
|
| 28 |
+
from ._classification import hamming_loss
|
| 29 |
+
from ._classification import hinge_loss
|
| 30 |
+
from ._classification import jaccard_score
|
| 31 |
+
from ._classification import log_loss
|
| 32 |
+
from ._classification import matthews_corrcoef
|
| 33 |
+
from ._classification import precision_recall_fscore_support
|
| 34 |
+
from ._classification import precision_score
|
| 35 |
+
from ._classification import recall_score
|
| 36 |
+
from ._classification import zero_one_loss
|
| 37 |
+
from ._classification import brier_score_loss
|
| 38 |
+
from ._classification import multilabel_confusion_matrix
|
| 39 |
+
|
| 40 |
+
from ._dist_metrics import DistanceMetric
|
| 41 |
+
|
| 42 |
+
from . import cluster
|
| 43 |
+
from .cluster import adjusted_mutual_info_score
|
| 44 |
+
from .cluster import adjusted_rand_score
|
| 45 |
+
from .cluster import rand_score
|
| 46 |
+
from .cluster import pair_confusion_matrix
|
| 47 |
+
from .cluster import completeness_score
|
| 48 |
+
from .cluster import consensus_score
|
| 49 |
+
from .cluster import homogeneity_completeness_v_measure
|
| 50 |
+
from .cluster import homogeneity_score
|
| 51 |
+
from .cluster import mutual_info_score
|
| 52 |
+
from .cluster import normalized_mutual_info_score
|
| 53 |
+
from .cluster import fowlkes_mallows_score
|
| 54 |
+
from .cluster import silhouette_samples
|
| 55 |
+
from .cluster import silhouette_score
|
| 56 |
+
from .cluster import calinski_harabasz_score
|
| 57 |
+
from .cluster import v_measure_score
|
| 58 |
+
from .cluster import davies_bouldin_score
|
| 59 |
+
|
| 60 |
+
from .pairwise import euclidean_distances
|
| 61 |
+
from .pairwise import nan_euclidean_distances
|
| 62 |
+
from .pairwise import pairwise_distances
|
| 63 |
+
from .pairwise import pairwise_distances_argmin
|
| 64 |
+
from .pairwise import pairwise_distances_argmin_min
|
| 65 |
+
from .pairwise import pairwise_kernels
|
| 66 |
+
from .pairwise import pairwise_distances_chunked
|
| 67 |
+
|
| 68 |
+
from ._regression import explained_variance_score
|
| 69 |
+
from ._regression import max_error
|
| 70 |
+
from ._regression import mean_absolute_error
|
| 71 |
+
from ._regression import mean_squared_error
|
| 72 |
+
from ._regression import mean_squared_log_error
|
| 73 |
+
from ._regression import median_absolute_error
|
| 74 |
+
from ._regression import mean_absolute_percentage_error
|
| 75 |
+
from ._regression import mean_pinball_loss
|
| 76 |
+
from ._regression import r2_score
|
| 77 |
+
from ._regression import mean_tweedie_deviance
|
| 78 |
+
from ._regression import mean_poisson_deviance
|
| 79 |
+
from ._regression import mean_gamma_deviance
|
| 80 |
+
from ._regression import d2_tweedie_score
|
| 81 |
+
from ._regression import d2_pinball_score
|
| 82 |
+
from ._regression import d2_absolute_error_score
|
| 83 |
+
|
| 84 |
+
|
| 85 |
+
from ._scorer import check_scoring
|
| 86 |
+
from ._scorer import make_scorer
|
| 87 |
+
from ._scorer import SCORERS
|
| 88 |
+
from ._scorer import get_scorer
|
| 89 |
+
from ._scorer import get_scorer_names
|
| 90 |
+
|
| 91 |
+
|
| 92 |
+
from ._plot.det_curve import DetCurveDisplay
|
| 93 |
+
from ._plot.roc_curve import RocCurveDisplay
|
| 94 |
+
from ._plot.precision_recall_curve import PrecisionRecallDisplay
|
| 95 |
+
from ._plot.confusion_matrix import ConfusionMatrixDisplay
|
| 96 |
+
from ._plot.regression import PredictionErrorDisplay
|
| 97 |
+
|
| 98 |
+
|
| 99 |
+
__all__ = [
|
| 100 |
+
"accuracy_score",
|
| 101 |
+
"adjusted_mutual_info_score",
|
| 102 |
+
"adjusted_rand_score",
|
| 103 |
+
"auc",
|
| 104 |
+
"average_precision_score",
|
| 105 |
+
"balanced_accuracy_score",
|
| 106 |
+
"calinski_harabasz_score",
|
| 107 |
+
"check_scoring",
|
| 108 |
+
"class_likelihood_ratios",
|
| 109 |
+
"classification_report",
|
| 110 |
+
"cluster",
|
| 111 |
+
"cohen_kappa_score",
|
| 112 |
+
"completeness_score",
|
| 113 |
+
"ConfusionMatrixDisplay",
|
| 114 |
+
"confusion_matrix",
|
| 115 |
+
"consensus_score",
|
| 116 |
+
"coverage_error",
|
| 117 |
+
"d2_tweedie_score",
|
| 118 |
+
"d2_absolute_error_score",
|
| 119 |
+
"d2_pinball_score",
|
| 120 |
+
"dcg_score",
|
| 121 |
+
"davies_bouldin_score",
|
| 122 |
+
"DetCurveDisplay",
|
| 123 |
+
"det_curve",
|
| 124 |
+
"DistanceMetric",
|
| 125 |
+
"euclidean_distances",
|
| 126 |
+
"explained_variance_score",
|
| 127 |
+
"f1_score",
|
| 128 |
+
"fbeta_score",
|
| 129 |
+
"fowlkes_mallows_score",
|
| 130 |
+
"get_scorer",
|
| 131 |
+
"hamming_loss",
|
| 132 |
+
"hinge_loss",
|
| 133 |
+
"homogeneity_completeness_v_measure",
|
| 134 |
+
"homogeneity_score",
|
| 135 |
+
"jaccard_score",
|
| 136 |
+
"label_ranking_average_precision_score",
|
| 137 |
+
"label_ranking_loss",
|
| 138 |
+
"log_loss",
|
| 139 |
+
"make_scorer",
|
| 140 |
+
"nan_euclidean_distances",
|
| 141 |
+
"matthews_corrcoef",
|
| 142 |
+
"max_error",
|
| 143 |
+
"mean_absolute_error",
|
| 144 |
+
"mean_squared_error",
|
| 145 |
+
"mean_squared_log_error",
|
| 146 |
+
"mean_pinball_loss",
|
| 147 |
+
"mean_poisson_deviance",
|
| 148 |
+
"mean_gamma_deviance",
|
| 149 |
+
"mean_tweedie_deviance",
|
| 150 |
+
"median_absolute_error",
|
| 151 |
+
"mean_absolute_percentage_error",
|
| 152 |
+
"multilabel_confusion_matrix",
|
| 153 |
+
"mutual_info_score",
|
| 154 |
+
"ndcg_score",
|
| 155 |
+
"normalized_mutual_info_score",
|
| 156 |
+
"pair_confusion_matrix",
|
| 157 |
+
"pairwise_distances",
|
| 158 |
+
"pairwise_distances_argmin",
|
| 159 |
+
"pairwise_distances_argmin_min",
|
| 160 |
+
"pairwise_distances_chunked",
|
| 161 |
+
"pairwise_kernels",
|
| 162 |
+
"PrecisionRecallDisplay",
|
| 163 |
+
"precision_recall_curve",
|
| 164 |
+
"precision_recall_fscore_support",
|
| 165 |
+
"precision_score",
|
| 166 |
+
"PredictionErrorDisplay",
|
| 167 |
+
"r2_score",
|
| 168 |
+
"rand_score",
|
| 169 |
+
"recall_score",
|
| 170 |
+
"RocCurveDisplay",
|
| 171 |
+
"roc_auc_score",
|
| 172 |
+
"roc_curve",
|
| 173 |
+
"SCORERS",
|
| 174 |
+
"get_scorer_names",
|
| 175 |
+
"silhouette_samples",
|
| 176 |
+
"silhouette_score",
|
| 177 |
+
"top_k_accuracy_score",
|
| 178 |
+
"v_measure_score",
|
| 179 |
+
"zero_one_loss",
|
| 180 |
+
"brier_score_loss",
|
| 181 |
+
]
|
videollama2/lib/python3.10/site-packages/sklearn/metrics/_classification.py
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/metrics/_pairwise_distances_reduction/__pycache__/__init__.cpython-310.pyc
ADDED
|
Binary file (365 Bytes). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/metrics/_pairwise_distances_reduction/__pycache__/_dispatcher.cpython-310.pyc
ADDED
|
Binary file (14.4 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/metrics/_pairwise_distances_reduction/_argkmin.pxd
ADDED
|
@@ -0,0 +1,56 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
cimport numpy as cnp
|
| 2 |
+
from ...utils._typedefs cimport ITYPE_t, DTYPE_t
|
| 3 |
+
|
| 4 |
+
cnp.import_array()
|
| 5 |
+
|
| 6 |
+
from ._base cimport BaseDistancesReduction64
|
| 7 |
+
from ._middle_term_computer cimport MiddleTermComputer64
|
| 8 |
+
|
| 9 |
+
cdef class ArgKmin64(BaseDistancesReduction64):
|
| 10 |
+
"""float64 implementation of the ArgKmin."""
|
| 11 |
+
|
| 12 |
+
cdef:
|
| 13 |
+
ITYPE_t k
|
| 14 |
+
|
| 15 |
+
ITYPE_t[:, ::1] argkmin_indices
|
| 16 |
+
DTYPE_t[:, ::1] argkmin_distances
|
| 17 |
+
|
| 18 |
+
# Used as array of pointers to private datastructures used in threads.
|
| 19 |
+
DTYPE_t ** heaps_r_distances_chunks
|
| 20 |
+
ITYPE_t ** heaps_indices_chunks
|
| 21 |
+
|
| 22 |
+
|
| 23 |
+
cdef class EuclideanArgKmin64(ArgKmin64):
|
| 24 |
+
"""EuclideanDistance-specialisation of ArgKmin64."""
|
| 25 |
+
cdef:
|
| 26 |
+
MiddleTermComputer64 middle_term_computer
|
| 27 |
+
const DTYPE_t[::1] X_norm_squared
|
| 28 |
+
const DTYPE_t[::1] Y_norm_squared
|
| 29 |
+
|
| 30 |
+
bint use_squared_distances
|
| 31 |
+
|
| 32 |
+
from ._base cimport BaseDistancesReduction32
|
| 33 |
+
from ._middle_term_computer cimport MiddleTermComputer32
|
| 34 |
+
|
| 35 |
+
cdef class ArgKmin32(BaseDistancesReduction32):
|
| 36 |
+
"""float32 implementation of the ArgKmin."""
|
| 37 |
+
|
| 38 |
+
cdef:
|
| 39 |
+
ITYPE_t k
|
| 40 |
+
|
| 41 |
+
ITYPE_t[:, ::1] argkmin_indices
|
| 42 |
+
DTYPE_t[:, ::1] argkmin_distances
|
| 43 |
+
|
| 44 |
+
# Used as array of pointers to private datastructures used in threads.
|
| 45 |
+
DTYPE_t ** heaps_r_distances_chunks
|
| 46 |
+
ITYPE_t ** heaps_indices_chunks
|
| 47 |
+
|
| 48 |
+
|
| 49 |
+
cdef class EuclideanArgKmin32(ArgKmin32):
|
| 50 |
+
"""EuclideanDistance-specialisation of ArgKmin32."""
|
| 51 |
+
cdef:
|
| 52 |
+
MiddleTermComputer32 middle_term_computer
|
| 53 |
+
const DTYPE_t[::1] X_norm_squared
|
| 54 |
+
const DTYPE_t[::1] Y_norm_squared
|
| 55 |
+
|
| 56 |
+
bint use_squared_distances
|
videollama2/lib/python3.10/site-packages/sklearn/metrics/_plot/__init__.py
ADDED
|
File without changes
|
videollama2/lib/python3.10/site-packages/sklearn/metrics/_plot/__pycache__/roc_curve.cpython-310.pyc
ADDED
|
Binary file (10.9 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/metrics/_plot/base.py
ADDED
|
@@ -0,0 +1,116 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from ...base import is_classifier
|
| 2 |
+
|
| 3 |
+
|
| 4 |
+
def _check_classifier_response_method(estimator, response_method):
|
| 5 |
+
"""Return prediction method from the response_method
|
| 6 |
+
|
| 7 |
+
Parameters
|
| 8 |
+
----------
|
| 9 |
+
estimator: object
|
| 10 |
+
Classifier to check
|
| 11 |
+
|
| 12 |
+
response_method: {'auto', 'predict_proba', 'decision_function'}
|
| 13 |
+
Specifies whether to use :term:`predict_proba` or
|
| 14 |
+
:term:`decision_function` as the target response. If set to 'auto',
|
| 15 |
+
:term:`predict_proba` is tried first and if it does not exist
|
| 16 |
+
:term:`decision_function` is tried next.
|
| 17 |
+
|
| 18 |
+
Returns
|
| 19 |
+
-------
|
| 20 |
+
prediction_method: callable
|
| 21 |
+
prediction method of estimator
|
| 22 |
+
"""
|
| 23 |
+
|
| 24 |
+
if response_method not in ("predict_proba", "decision_function", "auto"):
|
| 25 |
+
raise ValueError(
|
| 26 |
+
"response_method must be 'predict_proba', 'decision_function' or 'auto'"
|
| 27 |
+
)
|
| 28 |
+
|
| 29 |
+
error_msg = "response method {} is not defined in {}"
|
| 30 |
+
if response_method != "auto":
|
| 31 |
+
prediction_method = getattr(estimator, response_method, None)
|
| 32 |
+
if prediction_method is None:
|
| 33 |
+
raise ValueError(
|
| 34 |
+
error_msg.format(response_method, estimator.__class__.__name__)
|
| 35 |
+
)
|
| 36 |
+
else:
|
| 37 |
+
predict_proba = getattr(estimator, "predict_proba", None)
|
| 38 |
+
decision_function = getattr(estimator, "decision_function", None)
|
| 39 |
+
prediction_method = predict_proba or decision_function
|
| 40 |
+
if prediction_method is None:
|
| 41 |
+
raise ValueError(
|
| 42 |
+
error_msg.format(
|
| 43 |
+
"decision_function or predict_proba", estimator.__class__.__name__
|
| 44 |
+
)
|
| 45 |
+
)
|
| 46 |
+
|
| 47 |
+
return prediction_method
|
| 48 |
+
|
| 49 |
+
|
| 50 |
+
def _get_response(X, estimator, response_method, pos_label=None):
|
| 51 |
+
"""Return response and positive label.
|
| 52 |
+
|
| 53 |
+
Parameters
|
| 54 |
+
----------
|
| 55 |
+
X : {array-like, sparse matrix} of shape (n_samples, n_features)
|
| 56 |
+
Input values.
|
| 57 |
+
|
| 58 |
+
estimator : estimator instance
|
| 59 |
+
Fitted classifier or a fitted :class:`~sklearn.pipeline.Pipeline`
|
| 60 |
+
in which the last estimator is a classifier.
|
| 61 |
+
|
| 62 |
+
response_method: {'auto', 'predict_proba', 'decision_function'}
|
| 63 |
+
Specifies whether to use :term:`predict_proba` or
|
| 64 |
+
:term:`decision_function` as the target response. If set to 'auto',
|
| 65 |
+
:term:`predict_proba` is tried first and if it does not exist
|
| 66 |
+
:term:`decision_function` is tried next.
|
| 67 |
+
|
| 68 |
+
pos_label : str or int, default=None
|
| 69 |
+
The class considered as the positive class when computing
|
| 70 |
+
the metrics. By default, `estimators.classes_[1]` is
|
| 71 |
+
considered as the positive class.
|
| 72 |
+
|
| 73 |
+
Returns
|
| 74 |
+
-------
|
| 75 |
+
y_pred: ndarray of shape (n_samples,)
|
| 76 |
+
Target scores calculated from the provided response_method
|
| 77 |
+
and pos_label.
|
| 78 |
+
|
| 79 |
+
pos_label: str or int
|
| 80 |
+
The class considered as the positive class when computing
|
| 81 |
+
the metrics.
|
| 82 |
+
"""
|
| 83 |
+
classification_error = (
|
| 84 |
+
"Expected 'estimator' to be a binary classifier, but got"
|
| 85 |
+
f" {estimator.__class__.__name__}"
|
| 86 |
+
)
|
| 87 |
+
|
| 88 |
+
if not is_classifier(estimator):
|
| 89 |
+
raise ValueError(classification_error)
|
| 90 |
+
|
| 91 |
+
prediction_method = _check_classifier_response_method(estimator, response_method)
|
| 92 |
+
y_pred = prediction_method(X)
|
| 93 |
+
if pos_label is not None:
|
| 94 |
+
try:
|
| 95 |
+
class_idx = estimator.classes_.tolist().index(pos_label)
|
| 96 |
+
except ValueError as e:
|
| 97 |
+
raise ValueError(
|
| 98 |
+
"The class provided by 'pos_label' is unknown. Got "
|
| 99 |
+
f"{pos_label} instead of one of {set(estimator.classes_)}"
|
| 100 |
+
) from e
|
| 101 |
+
else:
|
| 102 |
+
class_idx = 1
|
| 103 |
+
pos_label = estimator.classes_[class_idx]
|
| 104 |
+
|
| 105 |
+
if y_pred.ndim != 1: # `predict_proba`
|
| 106 |
+
y_pred_shape = y_pred.shape[1]
|
| 107 |
+
if y_pred_shape != 2:
|
| 108 |
+
raise ValueError(
|
| 109 |
+
f"{classification_error} fit on multiclass ({y_pred_shape} classes)"
|
| 110 |
+
" data"
|
| 111 |
+
)
|
| 112 |
+
y_pred = y_pred[:, class_idx]
|
| 113 |
+
elif pos_label == estimator.classes_[0]: # `decision_function`
|
| 114 |
+
y_pred *= -1
|
| 115 |
+
|
| 116 |
+
return y_pred, pos_label
|
videollama2/lib/python3.10/site-packages/sklearn/metrics/_plot/confusion_matrix.py
ADDED
|
@@ -0,0 +1,482 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from itertools import product
|
| 2 |
+
|
| 3 |
+
import numpy as np
|
| 4 |
+
|
| 5 |
+
from .. import confusion_matrix
|
| 6 |
+
from ...utils import check_matplotlib_support
|
| 7 |
+
from ...utils.multiclass import unique_labels
|
| 8 |
+
from ...base import is_classifier
|
| 9 |
+
|
| 10 |
+
|
| 11 |
+
class ConfusionMatrixDisplay:
|
| 12 |
+
"""Confusion Matrix visualization.
|
| 13 |
+
|
| 14 |
+
It is recommend to use
|
| 15 |
+
:func:`~sklearn.metrics.ConfusionMatrixDisplay.from_estimator` or
|
| 16 |
+
:func:`~sklearn.metrics.ConfusionMatrixDisplay.from_predictions` to
|
| 17 |
+
create a :class:`ConfusionMatrixDisplay`. All parameters are stored as
|
| 18 |
+
attributes.
|
| 19 |
+
|
| 20 |
+
Read more in the :ref:`User Guide <visualizations>`.
|
| 21 |
+
|
| 22 |
+
Parameters
|
| 23 |
+
----------
|
| 24 |
+
confusion_matrix : ndarray of shape (n_classes, n_classes)
|
| 25 |
+
Confusion matrix.
|
| 26 |
+
|
| 27 |
+
display_labels : ndarray of shape (n_classes,), default=None
|
| 28 |
+
Display labels for plot. If None, display labels are set from 0 to
|
| 29 |
+
`n_classes - 1`.
|
| 30 |
+
|
| 31 |
+
Attributes
|
| 32 |
+
----------
|
| 33 |
+
im_ : matplotlib AxesImage
|
| 34 |
+
Image representing the confusion matrix.
|
| 35 |
+
|
| 36 |
+
text_ : ndarray of shape (n_classes, n_classes), dtype=matplotlib Text, \
|
| 37 |
+
or None
|
| 38 |
+
Array of matplotlib axes. `None` if `include_values` is false.
|
| 39 |
+
|
| 40 |
+
ax_ : matplotlib Axes
|
| 41 |
+
Axes with confusion matrix.
|
| 42 |
+
|
| 43 |
+
figure_ : matplotlib Figure
|
| 44 |
+
Figure containing the confusion matrix.
|
| 45 |
+
|
| 46 |
+
See Also
|
| 47 |
+
--------
|
| 48 |
+
confusion_matrix : Compute Confusion Matrix to evaluate the accuracy of a
|
| 49 |
+
classification.
|
| 50 |
+
ConfusionMatrixDisplay.from_estimator : Plot the confusion matrix
|
| 51 |
+
given an estimator, the data, and the label.
|
| 52 |
+
ConfusionMatrixDisplay.from_predictions : Plot the confusion matrix
|
| 53 |
+
given the true and predicted labels.
|
| 54 |
+
|
| 55 |
+
Examples
|
| 56 |
+
--------
|
| 57 |
+
>>> import matplotlib.pyplot as plt
|
| 58 |
+
>>> from sklearn.datasets import make_classification
|
| 59 |
+
>>> from sklearn.metrics import confusion_matrix, ConfusionMatrixDisplay
|
| 60 |
+
>>> from sklearn.model_selection import train_test_split
|
| 61 |
+
>>> from sklearn.svm import SVC
|
| 62 |
+
>>> X, y = make_classification(random_state=0)
|
| 63 |
+
>>> X_train, X_test, y_train, y_test = train_test_split(X, y,
|
| 64 |
+
... random_state=0)
|
| 65 |
+
>>> clf = SVC(random_state=0)
|
| 66 |
+
>>> clf.fit(X_train, y_train)
|
| 67 |
+
SVC(random_state=0)
|
| 68 |
+
>>> predictions = clf.predict(X_test)
|
| 69 |
+
>>> cm = confusion_matrix(y_test, predictions, labels=clf.classes_)
|
| 70 |
+
>>> disp = ConfusionMatrixDisplay(confusion_matrix=cm,
|
| 71 |
+
... display_labels=clf.classes_)
|
| 72 |
+
>>> disp.plot()
|
| 73 |
+
<...>
|
| 74 |
+
>>> plt.show()
|
| 75 |
+
"""
|
| 76 |
+
|
| 77 |
+
def __init__(self, confusion_matrix, *, display_labels=None):
|
| 78 |
+
self.confusion_matrix = confusion_matrix
|
| 79 |
+
self.display_labels = display_labels
|
| 80 |
+
|
| 81 |
+
def plot(
|
| 82 |
+
self,
|
| 83 |
+
*,
|
| 84 |
+
include_values=True,
|
| 85 |
+
cmap="viridis",
|
| 86 |
+
xticks_rotation="horizontal",
|
| 87 |
+
values_format=None,
|
| 88 |
+
ax=None,
|
| 89 |
+
colorbar=True,
|
| 90 |
+
im_kw=None,
|
| 91 |
+
text_kw=None,
|
| 92 |
+
):
|
| 93 |
+
"""Plot visualization.
|
| 94 |
+
|
| 95 |
+
Parameters
|
| 96 |
+
----------
|
| 97 |
+
include_values : bool, default=True
|
| 98 |
+
Includes values in confusion matrix.
|
| 99 |
+
|
| 100 |
+
cmap : str or matplotlib Colormap, default='viridis'
|
| 101 |
+
Colormap recognized by matplotlib.
|
| 102 |
+
|
| 103 |
+
xticks_rotation : {'vertical', 'horizontal'} or float, \
|
| 104 |
+
default='horizontal'
|
| 105 |
+
Rotation of xtick labels.
|
| 106 |
+
|
| 107 |
+
values_format : str, default=None
|
| 108 |
+
Format specification for values in confusion matrix. If `None`,
|
| 109 |
+
the format specification is 'd' or '.2g' whichever is shorter.
|
| 110 |
+
|
| 111 |
+
ax : matplotlib axes, default=None
|
| 112 |
+
Axes object to plot on. If `None`, a new figure and axes is
|
| 113 |
+
created.
|
| 114 |
+
|
| 115 |
+
colorbar : bool, default=True
|
| 116 |
+
Whether or not to add a colorbar to the plot.
|
| 117 |
+
|
| 118 |
+
im_kw : dict, default=None
|
| 119 |
+
Dict with keywords passed to `matplotlib.pyplot.imshow` call.
|
| 120 |
+
|
| 121 |
+
text_kw : dict, default=None
|
| 122 |
+
Dict with keywords passed to `matplotlib.pyplot.text` call.
|
| 123 |
+
|
| 124 |
+
.. versionadded:: 1.2
|
| 125 |
+
|
| 126 |
+
Returns
|
| 127 |
+
-------
|
| 128 |
+
display : :class:`~sklearn.metrics.ConfusionMatrixDisplay`
|
| 129 |
+
Returns a :class:`~sklearn.metrics.ConfusionMatrixDisplay` instance
|
| 130 |
+
that contains all the information to plot the confusion matrix.
|
| 131 |
+
"""
|
| 132 |
+
check_matplotlib_support("ConfusionMatrixDisplay.plot")
|
| 133 |
+
import matplotlib.pyplot as plt
|
| 134 |
+
|
| 135 |
+
if ax is None:
|
| 136 |
+
fig, ax = plt.subplots()
|
| 137 |
+
else:
|
| 138 |
+
fig = ax.figure
|
| 139 |
+
|
| 140 |
+
cm = self.confusion_matrix
|
| 141 |
+
n_classes = cm.shape[0]
|
| 142 |
+
|
| 143 |
+
default_im_kw = dict(interpolation="nearest", cmap=cmap)
|
| 144 |
+
im_kw = im_kw or {}
|
| 145 |
+
im_kw = {**default_im_kw, **im_kw}
|
| 146 |
+
text_kw = text_kw or {}
|
| 147 |
+
|
| 148 |
+
self.im_ = ax.imshow(cm, **im_kw)
|
| 149 |
+
self.text_ = None
|
| 150 |
+
cmap_min, cmap_max = self.im_.cmap(0), self.im_.cmap(1.0)
|
| 151 |
+
|
| 152 |
+
if include_values:
|
| 153 |
+
self.text_ = np.empty_like(cm, dtype=object)
|
| 154 |
+
|
| 155 |
+
# print text with appropriate color depending on background
|
| 156 |
+
thresh = (cm.max() + cm.min()) / 2.0
|
| 157 |
+
|
| 158 |
+
for i, j in product(range(n_classes), range(n_classes)):
|
| 159 |
+
color = cmap_max if cm[i, j] < thresh else cmap_min
|
| 160 |
+
|
| 161 |
+
if values_format is None:
|
| 162 |
+
text_cm = format(cm[i, j], ".2g")
|
| 163 |
+
if cm.dtype.kind != "f":
|
| 164 |
+
text_d = format(cm[i, j], "d")
|
| 165 |
+
if len(text_d) < len(text_cm):
|
| 166 |
+
text_cm = text_d
|
| 167 |
+
else:
|
| 168 |
+
text_cm = format(cm[i, j], values_format)
|
| 169 |
+
|
| 170 |
+
default_text_kwargs = dict(ha="center", va="center", color=color)
|
| 171 |
+
text_kwargs = {**default_text_kwargs, **text_kw}
|
| 172 |
+
|
| 173 |
+
self.text_[i, j] = ax.text(j, i, text_cm, **text_kwargs)
|
| 174 |
+
|
| 175 |
+
if self.display_labels is None:
|
| 176 |
+
display_labels = np.arange(n_classes)
|
| 177 |
+
else:
|
| 178 |
+
display_labels = self.display_labels
|
| 179 |
+
if colorbar:
|
| 180 |
+
fig.colorbar(self.im_, ax=ax)
|
| 181 |
+
ax.set(
|
| 182 |
+
xticks=np.arange(n_classes),
|
| 183 |
+
yticks=np.arange(n_classes),
|
| 184 |
+
xticklabels=display_labels,
|
| 185 |
+
yticklabels=display_labels,
|
| 186 |
+
ylabel="True label",
|
| 187 |
+
xlabel="Predicted label",
|
| 188 |
+
)
|
| 189 |
+
|
| 190 |
+
ax.set_ylim((n_classes - 0.5, -0.5))
|
| 191 |
+
plt.setp(ax.get_xticklabels(), rotation=xticks_rotation)
|
| 192 |
+
|
| 193 |
+
self.figure_ = fig
|
| 194 |
+
self.ax_ = ax
|
| 195 |
+
return self
|
| 196 |
+
|
| 197 |
+
@classmethod
|
| 198 |
+
def from_estimator(
|
| 199 |
+
cls,
|
| 200 |
+
estimator,
|
| 201 |
+
X,
|
| 202 |
+
y,
|
| 203 |
+
*,
|
| 204 |
+
labels=None,
|
| 205 |
+
sample_weight=None,
|
| 206 |
+
normalize=None,
|
| 207 |
+
display_labels=None,
|
| 208 |
+
include_values=True,
|
| 209 |
+
xticks_rotation="horizontal",
|
| 210 |
+
values_format=None,
|
| 211 |
+
cmap="viridis",
|
| 212 |
+
ax=None,
|
| 213 |
+
colorbar=True,
|
| 214 |
+
im_kw=None,
|
| 215 |
+
text_kw=None,
|
| 216 |
+
):
|
| 217 |
+
"""Plot Confusion Matrix given an estimator and some data.
|
| 218 |
+
|
| 219 |
+
Read more in the :ref:`User Guide <confusion_matrix>`.
|
| 220 |
+
|
| 221 |
+
.. versionadded:: 1.0
|
| 222 |
+
|
| 223 |
+
Parameters
|
| 224 |
+
----------
|
| 225 |
+
estimator : estimator instance
|
| 226 |
+
Fitted classifier or a fitted :class:`~sklearn.pipeline.Pipeline`
|
| 227 |
+
in which the last estimator is a classifier.
|
| 228 |
+
|
| 229 |
+
X : {array-like, sparse matrix} of shape (n_samples, n_features)
|
| 230 |
+
Input values.
|
| 231 |
+
|
| 232 |
+
y : array-like of shape (n_samples,)
|
| 233 |
+
Target values.
|
| 234 |
+
|
| 235 |
+
labels : array-like of shape (n_classes,), default=None
|
| 236 |
+
List of labels to index the confusion matrix. This may be used to
|
| 237 |
+
reorder or select a subset of labels. If `None` is given, those
|
| 238 |
+
that appear at least once in `y_true` or `y_pred` are used in
|
| 239 |
+
sorted order.
|
| 240 |
+
|
| 241 |
+
sample_weight : array-like of shape (n_samples,), default=None
|
| 242 |
+
Sample weights.
|
| 243 |
+
|
| 244 |
+
normalize : {'true', 'pred', 'all'}, default=None
|
| 245 |
+
Either to normalize the counts display in the matrix:
|
| 246 |
+
|
| 247 |
+
- if `'true'`, the confusion matrix is normalized over the true
|
| 248 |
+
conditions (e.g. rows);
|
| 249 |
+
- if `'pred'`, the confusion matrix is normalized over the
|
| 250 |
+
predicted conditions (e.g. columns);
|
| 251 |
+
- if `'all'`, the confusion matrix is normalized by the total
|
| 252 |
+
number of samples;
|
| 253 |
+
- if `None` (default), the confusion matrix will not be normalized.
|
| 254 |
+
|
| 255 |
+
display_labels : array-like of shape (n_classes,), default=None
|
| 256 |
+
Target names used for plotting. By default, `labels` will be used
|
| 257 |
+
if it is defined, otherwise the unique labels of `y_true` and
|
| 258 |
+
`y_pred` will be used.
|
| 259 |
+
|
| 260 |
+
include_values : bool, default=True
|
| 261 |
+
Includes values in confusion matrix.
|
| 262 |
+
|
| 263 |
+
xticks_rotation : {'vertical', 'horizontal'} or float, \
|
| 264 |
+
default='horizontal'
|
| 265 |
+
Rotation of xtick labels.
|
| 266 |
+
|
| 267 |
+
values_format : str, default=None
|
| 268 |
+
Format specification for values in confusion matrix. If `None`, the
|
| 269 |
+
format specification is 'd' or '.2g' whichever is shorter.
|
| 270 |
+
|
| 271 |
+
cmap : str or matplotlib Colormap, default='viridis'
|
| 272 |
+
Colormap recognized by matplotlib.
|
| 273 |
+
|
| 274 |
+
ax : matplotlib Axes, default=None
|
| 275 |
+
Axes object to plot on. If `None`, a new figure and axes is
|
| 276 |
+
created.
|
| 277 |
+
|
| 278 |
+
colorbar : bool, default=True
|
| 279 |
+
Whether or not to add a colorbar to the plot.
|
| 280 |
+
|
| 281 |
+
im_kw : dict, default=None
|
| 282 |
+
Dict with keywords passed to `matplotlib.pyplot.imshow` call.
|
| 283 |
+
|
| 284 |
+
text_kw : dict, default=None
|
| 285 |
+
Dict with keywords passed to `matplotlib.pyplot.text` call.
|
| 286 |
+
|
| 287 |
+
.. versionadded:: 1.2
|
| 288 |
+
|
| 289 |
+
Returns
|
| 290 |
+
-------
|
| 291 |
+
display : :class:`~sklearn.metrics.ConfusionMatrixDisplay`
|
| 292 |
+
|
| 293 |
+
See Also
|
| 294 |
+
--------
|
| 295 |
+
ConfusionMatrixDisplay.from_predictions : Plot the confusion matrix
|
| 296 |
+
given the true and predicted labels.
|
| 297 |
+
|
| 298 |
+
Examples
|
| 299 |
+
--------
|
| 300 |
+
>>> import matplotlib.pyplot as plt
|
| 301 |
+
>>> from sklearn.datasets import make_classification
|
| 302 |
+
>>> from sklearn.metrics import ConfusionMatrixDisplay
|
| 303 |
+
>>> from sklearn.model_selection import train_test_split
|
| 304 |
+
>>> from sklearn.svm import SVC
|
| 305 |
+
>>> X, y = make_classification(random_state=0)
|
| 306 |
+
>>> X_train, X_test, y_train, y_test = train_test_split(
|
| 307 |
+
... X, y, random_state=0)
|
| 308 |
+
>>> clf = SVC(random_state=0)
|
| 309 |
+
>>> clf.fit(X_train, y_train)
|
| 310 |
+
SVC(random_state=0)
|
| 311 |
+
>>> ConfusionMatrixDisplay.from_estimator(
|
| 312 |
+
... clf, X_test, y_test)
|
| 313 |
+
<...>
|
| 314 |
+
>>> plt.show()
|
| 315 |
+
"""
|
| 316 |
+
method_name = f"{cls.__name__}.from_estimator"
|
| 317 |
+
check_matplotlib_support(method_name)
|
| 318 |
+
if not is_classifier(estimator):
|
| 319 |
+
raise ValueError(f"{method_name} only supports classifiers")
|
| 320 |
+
y_pred = estimator.predict(X)
|
| 321 |
+
|
| 322 |
+
return cls.from_predictions(
|
| 323 |
+
y,
|
| 324 |
+
y_pred,
|
| 325 |
+
sample_weight=sample_weight,
|
| 326 |
+
labels=labels,
|
| 327 |
+
normalize=normalize,
|
| 328 |
+
display_labels=display_labels,
|
| 329 |
+
include_values=include_values,
|
| 330 |
+
cmap=cmap,
|
| 331 |
+
ax=ax,
|
| 332 |
+
xticks_rotation=xticks_rotation,
|
| 333 |
+
values_format=values_format,
|
| 334 |
+
colorbar=colorbar,
|
| 335 |
+
im_kw=im_kw,
|
| 336 |
+
text_kw=text_kw,
|
| 337 |
+
)
|
| 338 |
+
|
| 339 |
+
@classmethod
|
| 340 |
+
def from_predictions(
|
| 341 |
+
cls,
|
| 342 |
+
y_true,
|
| 343 |
+
y_pred,
|
| 344 |
+
*,
|
| 345 |
+
labels=None,
|
| 346 |
+
sample_weight=None,
|
| 347 |
+
normalize=None,
|
| 348 |
+
display_labels=None,
|
| 349 |
+
include_values=True,
|
| 350 |
+
xticks_rotation="horizontal",
|
| 351 |
+
values_format=None,
|
| 352 |
+
cmap="viridis",
|
| 353 |
+
ax=None,
|
| 354 |
+
colorbar=True,
|
| 355 |
+
im_kw=None,
|
| 356 |
+
text_kw=None,
|
| 357 |
+
):
|
| 358 |
+
"""Plot Confusion Matrix given true and predicted labels.
|
| 359 |
+
|
| 360 |
+
Read more in the :ref:`User Guide <confusion_matrix>`.
|
| 361 |
+
|
| 362 |
+
.. versionadded:: 1.0
|
| 363 |
+
|
| 364 |
+
Parameters
|
| 365 |
+
----------
|
| 366 |
+
y_true : array-like of shape (n_samples,)
|
| 367 |
+
True labels.
|
| 368 |
+
|
| 369 |
+
y_pred : array-like of shape (n_samples,)
|
| 370 |
+
The predicted labels given by the method `predict` of an
|
| 371 |
+
classifier.
|
| 372 |
+
|
| 373 |
+
labels : array-like of shape (n_classes,), default=None
|
| 374 |
+
List of labels to index the confusion matrix. This may be used to
|
| 375 |
+
reorder or select a subset of labels. If `None` is given, those
|
| 376 |
+
that appear at least once in `y_true` or `y_pred` are used in
|
| 377 |
+
sorted order.
|
| 378 |
+
|
| 379 |
+
sample_weight : array-like of shape (n_samples,), default=None
|
| 380 |
+
Sample weights.
|
| 381 |
+
|
| 382 |
+
normalize : {'true', 'pred', 'all'}, default=None
|
| 383 |
+
Either to normalize the counts display in the matrix:
|
| 384 |
+
|
| 385 |
+
- if `'true'`, the confusion matrix is normalized over the true
|
| 386 |
+
conditions (e.g. rows);
|
| 387 |
+
- if `'pred'`, the confusion matrix is normalized over the
|
| 388 |
+
predicted conditions (e.g. columns);
|
| 389 |
+
- if `'all'`, the confusion matrix is normalized by the total
|
| 390 |
+
number of samples;
|
| 391 |
+
- if `None` (default), the confusion matrix will not be normalized.
|
| 392 |
+
|
| 393 |
+
display_labels : array-like of shape (n_classes,), default=None
|
| 394 |
+
Target names used for plotting. By default, `labels` will be used
|
| 395 |
+
if it is defined, otherwise the unique labels of `y_true` and
|
| 396 |
+
`y_pred` will be used.
|
| 397 |
+
|
| 398 |
+
include_values : bool, default=True
|
| 399 |
+
Includes values in confusion matrix.
|
| 400 |
+
|
| 401 |
+
xticks_rotation : {'vertical', 'horizontal'} or float, \
|
| 402 |
+
default='horizontal'
|
| 403 |
+
Rotation of xtick labels.
|
| 404 |
+
|
| 405 |
+
values_format : str, default=None
|
| 406 |
+
Format specification for values in confusion matrix. If `None`, the
|
| 407 |
+
format specification is 'd' or '.2g' whichever is shorter.
|
| 408 |
+
|
| 409 |
+
cmap : str or matplotlib Colormap, default='viridis'
|
| 410 |
+
Colormap recognized by matplotlib.
|
| 411 |
+
|
| 412 |
+
ax : matplotlib Axes, default=None
|
| 413 |
+
Axes object to plot on. If `None`, a new figure and axes is
|
| 414 |
+
created.
|
| 415 |
+
|
| 416 |
+
colorbar : bool, default=True
|
| 417 |
+
Whether or not to add a colorbar to the plot.
|
| 418 |
+
|
| 419 |
+
im_kw : dict, default=None
|
| 420 |
+
Dict with keywords passed to `matplotlib.pyplot.imshow` call.
|
| 421 |
+
|
| 422 |
+
text_kw : dict, default=None
|
| 423 |
+
Dict with keywords passed to `matplotlib.pyplot.text` call.
|
| 424 |
+
|
| 425 |
+
.. versionadded:: 1.2
|
| 426 |
+
|
| 427 |
+
Returns
|
| 428 |
+
-------
|
| 429 |
+
display : :class:`~sklearn.metrics.ConfusionMatrixDisplay`
|
| 430 |
+
|
| 431 |
+
See Also
|
| 432 |
+
--------
|
| 433 |
+
ConfusionMatrixDisplay.from_estimator : Plot the confusion matrix
|
| 434 |
+
given an estimator, the data, and the label.
|
| 435 |
+
|
| 436 |
+
Examples
|
| 437 |
+
--------
|
| 438 |
+
>>> import matplotlib.pyplot as plt
|
| 439 |
+
>>> from sklearn.datasets import make_classification
|
| 440 |
+
>>> from sklearn.metrics import ConfusionMatrixDisplay
|
| 441 |
+
>>> from sklearn.model_selection import train_test_split
|
| 442 |
+
>>> from sklearn.svm import SVC
|
| 443 |
+
>>> X, y = make_classification(random_state=0)
|
| 444 |
+
>>> X_train, X_test, y_train, y_test = train_test_split(
|
| 445 |
+
... X, y, random_state=0)
|
| 446 |
+
>>> clf = SVC(random_state=0)
|
| 447 |
+
>>> clf.fit(X_train, y_train)
|
| 448 |
+
SVC(random_state=0)
|
| 449 |
+
>>> y_pred = clf.predict(X_test)
|
| 450 |
+
>>> ConfusionMatrixDisplay.from_predictions(
|
| 451 |
+
... y_test, y_pred)
|
| 452 |
+
<...>
|
| 453 |
+
>>> plt.show()
|
| 454 |
+
"""
|
| 455 |
+
check_matplotlib_support(f"{cls.__name__}.from_predictions")
|
| 456 |
+
|
| 457 |
+
if display_labels is None:
|
| 458 |
+
if labels is None:
|
| 459 |
+
display_labels = unique_labels(y_true, y_pred)
|
| 460 |
+
else:
|
| 461 |
+
display_labels = labels
|
| 462 |
+
|
| 463 |
+
cm = confusion_matrix(
|
| 464 |
+
y_true,
|
| 465 |
+
y_pred,
|
| 466 |
+
sample_weight=sample_weight,
|
| 467 |
+
labels=labels,
|
| 468 |
+
normalize=normalize,
|
| 469 |
+
)
|
| 470 |
+
|
| 471 |
+
disp = cls(confusion_matrix=cm, display_labels=display_labels)
|
| 472 |
+
|
| 473 |
+
return disp.plot(
|
| 474 |
+
include_values=include_values,
|
| 475 |
+
cmap=cmap,
|
| 476 |
+
ax=ax,
|
| 477 |
+
xticks_rotation=xticks_rotation,
|
| 478 |
+
values_format=values_format,
|
| 479 |
+
colorbar=colorbar,
|
| 480 |
+
im_kw=im_kw,
|
| 481 |
+
text_kw=text_kw,
|
| 482 |
+
)
|
videollama2/lib/python3.10/site-packages/sklearn/metrics/_plot/det_curve.py
ADDED
|
@@ -0,0 +1,346 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import scipy as sp
|
| 2 |
+
|
| 3 |
+
from .base import _get_response
|
| 4 |
+
|
| 5 |
+
from .. import det_curve
|
| 6 |
+
from .._base import _check_pos_label_consistency
|
| 7 |
+
|
| 8 |
+
from ...utils import check_matplotlib_support
|
| 9 |
+
|
| 10 |
+
|
| 11 |
+
class DetCurveDisplay:
|
| 12 |
+
"""DET curve visualization.
|
| 13 |
+
|
| 14 |
+
It is recommend to use :func:`~sklearn.metrics.DetCurveDisplay.from_estimator`
|
| 15 |
+
or :func:`~sklearn.metrics.DetCurveDisplay.from_predictions` to create a
|
| 16 |
+
visualizer. All parameters are stored as attributes.
|
| 17 |
+
|
| 18 |
+
Read more in the :ref:`User Guide <visualizations>`.
|
| 19 |
+
|
| 20 |
+
.. versionadded:: 0.24
|
| 21 |
+
|
| 22 |
+
Parameters
|
| 23 |
+
----------
|
| 24 |
+
fpr : ndarray
|
| 25 |
+
False positive rate.
|
| 26 |
+
|
| 27 |
+
fnr : ndarray
|
| 28 |
+
False negative rate.
|
| 29 |
+
|
| 30 |
+
estimator_name : str, default=None
|
| 31 |
+
Name of estimator. If None, the estimator name is not shown.
|
| 32 |
+
|
| 33 |
+
pos_label : str or int, default=None
|
| 34 |
+
The label of the positive class.
|
| 35 |
+
|
| 36 |
+
Attributes
|
| 37 |
+
----------
|
| 38 |
+
line_ : matplotlib Artist
|
| 39 |
+
DET Curve.
|
| 40 |
+
|
| 41 |
+
ax_ : matplotlib Axes
|
| 42 |
+
Axes with DET Curve.
|
| 43 |
+
|
| 44 |
+
figure_ : matplotlib Figure
|
| 45 |
+
Figure containing the curve.
|
| 46 |
+
|
| 47 |
+
See Also
|
| 48 |
+
--------
|
| 49 |
+
det_curve : Compute error rates for different probability thresholds.
|
| 50 |
+
DetCurveDisplay.from_estimator : Plot DET curve given an estimator and
|
| 51 |
+
some data.
|
| 52 |
+
DetCurveDisplay.from_predictions : Plot DET curve given the true and
|
| 53 |
+
predicted labels.
|
| 54 |
+
|
| 55 |
+
Examples
|
| 56 |
+
--------
|
| 57 |
+
>>> import matplotlib.pyplot as plt
|
| 58 |
+
>>> from sklearn.datasets import make_classification
|
| 59 |
+
>>> from sklearn.metrics import det_curve, DetCurveDisplay
|
| 60 |
+
>>> from sklearn.model_selection import train_test_split
|
| 61 |
+
>>> from sklearn.svm import SVC
|
| 62 |
+
>>> X, y = make_classification(n_samples=1000, random_state=0)
|
| 63 |
+
>>> X_train, X_test, y_train, y_test = train_test_split(
|
| 64 |
+
... X, y, test_size=0.4, random_state=0)
|
| 65 |
+
>>> clf = SVC(random_state=0).fit(X_train, y_train)
|
| 66 |
+
>>> y_pred = clf.decision_function(X_test)
|
| 67 |
+
>>> fpr, fnr, _ = det_curve(y_test, y_pred)
|
| 68 |
+
>>> display = DetCurveDisplay(
|
| 69 |
+
... fpr=fpr, fnr=fnr, estimator_name="SVC"
|
| 70 |
+
... )
|
| 71 |
+
>>> display.plot()
|
| 72 |
+
<...>
|
| 73 |
+
>>> plt.show()
|
| 74 |
+
"""
|
| 75 |
+
|
| 76 |
+
def __init__(self, *, fpr, fnr, estimator_name=None, pos_label=None):
|
| 77 |
+
self.fpr = fpr
|
| 78 |
+
self.fnr = fnr
|
| 79 |
+
self.estimator_name = estimator_name
|
| 80 |
+
self.pos_label = pos_label
|
| 81 |
+
|
| 82 |
+
@classmethod
|
| 83 |
+
def from_estimator(
|
| 84 |
+
cls,
|
| 85 |
+
estimator,
|
| 86 |
+
X,
|
| 87 |
+
y,
|
| 88 |
+
*,
|
| 89 |
+
sample_weight=None,
|
| 90 |
+
response_method="auto",
|
| 91 |
+
pos_label=None,
|
| 92 |
+
name=None,
|
| 93 |
+
ax=None,
|
| 94 |
+
**kwargs,
|
| 95 |
+
):
|
| 96 |
+
"""Plot DET curve given an estimator and data.
|
| 97 |
+
|
| 98 |
+
Read more in the :ref:`User Guide <visualizations>`.
|
| 99 |
+
|
| 100 |
+
.. versionadded:: 1.0
|
| 101 |
+
|
| 102 |
+
Parameters
|
| 103 |
+
----------
|
| 104 |
+
estimator : estimator instance
|
| 105 |
+
Fitted classifier or a fitted :class:`~sklearn.pipeline.Pipeline`
|
| 106 |
+
in which the last estimator is a classifier.
|
| 107 |
+
|
| 108 |
+
X : {array-like, sparse matrix} of shape (n_samples, n_features)
|
| 109 |
+
Input values.
|
| 110 |
+
|
| 111 |
+
y : array-like of shape (n_samples,)
|
| 112 |
+
Target values.
|
| 113 |
+
|
| 114 |
+
sample_weight : array-like of shape (n_samples,), default=None
|
| 115 |
+
Sample weights.
|
| 116 |
+
|
| 117 |
+
response_method : {'predict_proba', 'decision_function', 'auto'} \
|
| 118 |
+
default='auto'
|
| 119 |
+
Specifies whether to use :term:`predict_proba` or
|
| 120 |
+
:term:`decision_function` as the predicted target response. If set
|
| 121 |
+
to 'auto', :term:`predict_proba` is tried first and if it does not
|
| 122 |
+
exist :term:`decision_function` is tried next.
|
| 123 |
+
|
| 124 |
+
pos_label : str or int, default=None
|
| 125 |
+
The label of the positive class. When `pos_label=None`, if `y_true`
|
| 126 |
+
is in {-1, 1} or {0, 1}, `pos_label` is set to 1, otherwise an
|
| 127 |
+
error will be raised.
|
| 128 |
+
|
| 129 |
+
name : str, default=None
|
| 130 |
+
Name of DET curve for labeling. If `None`, use the name of the
|
| 131 |
+
estimator.
|
| 132 |
+
|
| 133 |
+
ax : matplotlib axes, default=None
|
| 134 |
+
Axes object to plot on. If `None`, a new figure and axes is
|
| 135 |
+
created.
|
| 136 |
+
|
| 137 |
+
**kwargs : dict
|
| 138 |
+
Additional keywords arguments passed to matplotlib `plot` function.
|
| 139 |
+
|
| 140 |
+
Returns
|
| 141 |
+
-------
|
| 142 |
+
display : :class:`~sklearn.metrics.DetCurveDisplay`
|
| 143 |
+
Object that stores computed values.
|
| 144 |
+
|
| 145 |
+
See Also
|
| 146 |
+
--------
|
| 147 |
+
det_curve : Compute error rates for different probability thresholds.
|
| 148 |
+
DetCurveDisplay.from_predictions : Plot DET curve given the true and
|
| 149 |
+
predicted labels.
|
| 150 |
+
|
| 151 |
+
Examples
|
| 152 |
+
--------
|
| 153 |
+
>>> import matplotlib.pyplot as plt
|
| 154 |
+
>>> from sklearn.datasets import make_classification
|
| 155 |
+
>>> from sklearn.metrics import DetCurveDisplay
|
| 156 |
+
>>> from sklearn.model_selection import train_test_split
|
| 157 |
+
>>> from sklearn.svm import SVC
|
| 158 |
+
>>> X, y = make_classification(n_samples=1000, random_state=0)
|
| 159 |
+
>>> X_train, X_test, y_train, y_test = train_test_split(
|
| 160 |
+
... X, y, test_size=0.4, random_state=0)
|
| 161 |
+
>>> clf = SVC(random_state=0).fit(X_train, y_train)
|
| 162 |
+
>>> DetCurveDisplay.from_estimator(
|
| 163 |
+
... clf, X_test, y_test)
|
| 164 |
+
<...>
|
| 165 |
+
>>> plt.show()
|
| 166 |
+
"""
|
| 167 |
+
check_matplotlib_support(f"{cls.__name__}.from_estimator")
|
| 168 |
+
|
| 169 |
+
name = estimator.__class__.__name__ if name is None else name
|
| 170 |
+
|
| 171 |
+
y_pred, pos_label = _get_response(
|
| 172 |
+
X,
|
| 173 |
+
estimator,
|
| 174 |
+
response_method,
|
| 175 |
+
pos_label=pos_label,
|
| 176 |
+
)
|
| 177 |
+
|
| 178 |
+
return cls.from_predictions(
|
| 179 |
+
y_true=y,
|
| 180 |
+
y_pred=y_pred,
|
| 181 |
+
sample_weight=sample_weight,
|
| 182 |
+
name=name,
|
| 183 |
+
ax=ax,
|
| 184 |
+
pos_label=pos_label,
|
| 185 |
+
**kwargs,
|
| 186 |
+
)
|
| 187 |
+
|
| 188 |
+
@classmethod
|
| 189 |
+
def from_predictions(
|
| 190 |
+
cls,
|
| 191 |
+
y_true,
|
| 192 |
+
y_pred,
|
| 193 |
+
*,
|
| 194 |
+
sample_weight=None,
|
| 195 |
+
pos_label=None,
|
| 196 |
+
name=None,
|
| 197 |
+
ax=None,
|
| 198 |
+
**kwargs,
|
| 199 |
+
):
|
| 200 |
+
"""Plot the DET curve given the true and predicted labels.
|
| 201 |
+
|
| 202 |
+
Read more in the :ref:`User Guide <visualizations>`.
|
| 203 |
+
|
| 204 |
+
.. versionadded:: 1.0
|
| 205 |
+
|
| 206 |
+
Parameters
|
| 207 |
+
----------
|
| 208 |
+
y_true : array-like of shape (n_samples,)
|
| 209 |
+
True labels.
|
| 210 |
+
|
| 211 |
+
y_pred : array-like of shape (n_samples,)
|
| 212 |
+
Target scores, can either be probability estimates of the positive
|
| 213 |
+
class, confidence values, or non-thresholded measure of decisions
|
| 214 |
+
(as returned by `decision_function` on some classifiers).
|
| 215 |
+
|
| 216 |
+
sample_weight : array-like of shape (n_samples,), default=None
|
| 217 |
+
Sample weights.
|
| 218 |
+
|
| 219 |
+
pos_label : str or int, default=None
|
| 220 |
+
The label of the positive class. When `pos_label=None`, if `y_true`
|
| 221 |
+
is in {-1, 1} or {0, 1}, `pos_label` is set to 1, otherwise an
|
| 222 |
+
error will be raised.
|
| 223 |
+
|
| 224 |
+
name : str, default=None
|
| 225 |
+
Name of DET curve for labeling. If `None`, name will be set to
|
| 226 |
+
`"Classifier"`.
|
| 227 |
+
|
| 228 |
+
ax : matplotlib axes, default=None
|
| 229 |
+
Axes object to plot on. If `None`, a new figure and axes is
|
| 230 |
+
created.
|
| 231 |
+
|
| 232 |
+
**kwargs : dict
|
| 233 |
+
Additional keywords arguments passed to matplotlib `plot` function.
|
| 234 |
+
|
| 235 |
+
Returns
|
| 236 |
+
-------
|
| 237 |
+
display : :class:`~sklearn.metrics.DetCurveDisplay`
|
| 238 |
+
Object that stores computed values.
|
| 239 |
+
|
| 240 |
+
See Also
|
| 241 |
+
--------
|
| 242 |
+
det_curve : Compute error rates for different probability thresholds.
|
| 243 |
+
DetCurveDisplay.from_estimator : Plot DET curve given an estimator and
|
| 244 |
+
some data.
|
| 245 |
+
|
| 246 |
+
Examples
|
| 247 |
+
--------
|
| 248 |
+
>>> import matplotlib.pyplot as plt
|
| 249 |
+
>>> from sklearn.datasets import make_classification
|
| 250 |
+
>>> from sklearn.metrics import DetCurveDisplay
|
| 251 |
+
>>> from sklearn.model_selection import train_test_split
|
| 252 |
+
>>> from sklearn.svm import SVC
|
| 253 |
+
>>> X, y = make_classification(n_samples=1000, random_state=0)
|
| 254 |
+
>>> X_train, X_test, y_train, y_test = train_test_split(
|
| 255 |
+
... X, y, test_size=0.4, random_state=0)
|
| 256 |
+
>>> clf = SVC(random_state=0).fit(X_train, y_train)
|
| 257 |
+
>>> y_pred = clf.decision_function(X_test)
|
| 258 |
+
>>> DetCurveDisplay.from_predictions(
|
| 259 |
+
... y_test, y_pred)
|
| 260 |
+
<...>
|
| 261 |
+
>>> plt.show()
|
| 262 |
+
"""
|
| 263 |
+
check_matplotlib_support(f"{cls.__name__}.from_predictions")
|
| 264 |
+
fpr, fnr, _ = det_curve(
|
| 265 |
+
y_true,
|
| 266 |
+
y_pred,
|
| 267 |
+
pos_label=pos_label,
|
| 268 |
+
sample_weight=sample_weight,
|
| 269 |
+
)
|
| 270 |
+
|
| 271 |
+
pos_label = _check_pos_label_consistency(pos_label, y_true)
|
| 272 |
+
name = "Classifier" if name is None else name
|
| 273 |
+
|
| 274 |
+
viz = DetCurveDisplay(
|
| 275 |
+
fpr=fpr,
|
| 276 |
+
fnr=fnr,
|
| 277 |
+
estimator_name=name,
|
| 278 |
+
pos_label=pos_label,
|
| 279 |
+
)
|
| 280 |
+
|
| 281 |
+
return viz.plot(ax=ax, name=name, **kwargs)
|
| 282 |
+
|
| 283 |
+
def plot(self, ax=None, *, name=None, **kwargs):
|
| 284 |
+
"""Plot visualization.
|
| 285 |
+
|
| 286 |
+
Parameters
|
| 287 |
+
----------
|
| 288 |
+
ax : matplotlib axes, default=None
|
| 289 |
+
Axes object to plot on. If `None`, a new figure and axes is
|
| 290 |
+
created.
|
| 291 |
+
|
| 292 |
+
name : str, default=None
|
| 293 |
+
Name of DET curve for labeling. If `None`, use `estimator_name` if
|
| 294 |
+
it is not `None`, otherwise no labeling is shown.
|
| 295 |
+
|
| 296 |
+
**kwargs : dict
|
| 297 |
+
Additional keywords arguments passed to matplotlib `plot` function.
|
| 298 |
+
|
| 299 |
+
Returns
|
| 300 |
+
-------
|
| 301 |
+
display : :class:`~sklearn.metrics.plot.DetCurveDisplay`
|
| 302 |
+
Object that stores computed values.
|
| 303 |
+
"""
|
| 304 |
+
check_matplotlib_support("DetCurveDisplay.plot")
|
| 305 |
+
|
| 306 |
+
name = self.estimator_name if name is None else name
|
| 307 |
+
line_kwargs = {} if name is None else {"label": name}
|
| 308 |
+
line_kwargs.update(**kwargs)
|
| 309 |
+
|
| 310 |
+
import matplotlib.pyplot as plt
|
| 311 |
+
|
| 312 |
+
if ax is None:
|
| 313 |
+
_, ax = plt.subplots()
|
| 314 |
+
|
| 315 |
+
(self.line_,) = ax.plot(
|
| 316 |
+
sp.stats.norm.ppf(self.fpr),
|
| 317 |
+
sp.stats.norm.ppf(self.fnr),
|
| 318 |
+
**line_kwargs,
|
| 319 |
+
)
|
| 320 |
+
info_pos_label = (
|
| 321 |
+
f" (Positive label: {self.pos_label})" if self.pos_label is not None else ""
|
| 322 |
+
)
|
| 323 |
+
|
| 324 |
+
xlabel = "False Positive Rate" + info_pos_label
|
| 325 |
+
ylabel = "False Negative Rate" + info_pos_label
|
| 326 |
+
ax.set(xlabel=xlabel, ylabel=ylabel)
|
| 327 |
+
|
| 328 |
+
if "label" in line_kwargs:
|
| 329 |
+
ax.legend(loc="lower right")
|
| 330 |
+
|
| 331 |
+
ticks = [0.001, 0.01, 0.05, 0.20, 0.5, 0.80, 0.95, 0.99, 0.999]
|
| 332 |
+
tick_locations = sp.stats.norm.ppf(ticks)
|
| 333 |
+
tick_labels = [
|
| 334 |
+
"{:.0%}".format(s) if (100 * s).is_integer() else "{:.1%}".format(s)
|
| 335 |
+
for s in ticks
|
| 336 |
+
]
|
| 337 |
+
ax.set_xticks(tick_locations)
|
| 338 |
+
ax.set_xticklabels(tick_labels)
|
| 339 |
+
ax.set_xlim(-3, 3)
|
| 340 |
+
ax.set_yticks(tick_locations)
|
| 341 |
+
ax.set_yticklabels(tick_labels)
|
| 342 |
+
ax.set_ylim(-3, 3)
|
| 343 |
+
|
| 344 |
+
self.ax_ = ax
|
| 345 |
+
self.figure_ = ax.figure
|
| 346 |
+
return self
|
videollama2/lib/python3.10/site-packages/sklearn/metrics/_plot/precision_recall_curve.py
ADDED
|
@@ -0,0 +1,393 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from sklearn.base import is_classifier
|
| 2 |
+
from .base import _get_response
|
| 3 |
+
|
| 4 |
+
from .. import average_precision_score
|
| 5 |
+
from .. import precision_recall_curve
|
| 6 |
+
from .._base import _check_pos_label_consistency
|
| 7 |
+
from .._classification import check_consistent_length
|
| 8 |
+
|
| 9 |
+
from ...utils import check_matplotlib_support
|
| 10 |
+
|
| 11 |
+
|
| 12 |
+
class PrecisionRecallDisplay:
|
| 13 |
+
"""Precision Recall visualization.
|
| 14 |
+
|
| 15 |
+
It is recommend to use
|
| 16 |
+
:func:`~sklearn.metrics.PrecisionRecallDisplay.from_estimator` or
|
| 17 |
+
:func:`~sklearn.metrics.PrecisionRecallDisplay.from_predictions` to create
|
| 18 |
+
a :class:`~sklearn.metrics.PredictionRecallDisplay`. All parameters are
|
| 19 |
+
stored as attributes.
|
| 20 |
+
|
| 21 |
+
Read more in the :ref:`User Guide <visualizations>`.
|
| 22 |
+
|
| 23 |
+
Parameters
|
| 24 |
+
----------
|
| 25 |
+
precision : ndarray
|
| 26 |
+
Precision values.
|
| 27 |
+
|
| 28 |
+
recall : ndarray
|
| 29 |
+
Recall values.
|
| 30 |
+
|
| 31 |
+
average_precision : float, default=None
|
| 32 |
+
Average precision. If None, the average precision is not shown.
|
| 33 |
+
|
| 34 |
+
estimator_name : str, default=None
|
| 35 |
+
Name of estimator. If None, then the estimator name is not shown.
|
| 36 |
+
|
| 37 |
+
pos_label : str or int, default=None
|
| 38 |
+
The class considered as the positive class. If None, the class will not
|
| 39 |
+
be shown in the legend.
|
| 40 |
+
|
| 41 |
+
.. versionadded:: 0.24
|
| 42 |
+
|
| 43 |
+
Attributes
|
| 44 |
+
----------
|
| 45 |
+
line_ : matplotlib Artist
|
| 46 |
+
Precision recall curve.
|
| 47 |
+
|
| 48 |
+
ax_ : matplotlib Axes
|
| 49 |
+
Axes with precision recall curve.
|
| 50 |
+
|
| 51 |
+
figure_ : matplotlib Figure
|
| 52 |
+
Figure containing the curve.
|
| 53 |
+
|
| 54 |
+
See Also
|
| 55 |
+
--------
|
| 56 |
+
precision_recall_curve : Compute precision-recall pairs for different
|
| 57 |
+
probability thresholds.
|
| 58 |
+
PrecisionRecallDisplay.from_estimator : Plot Precision Recall Curve given
|
| 59 |
+
a binary classifier.
|
| 60 |
+
PrecisionRecallDisplay.from_predictions : Plot Precision Recall Curve
|
| 61 |
+
using predictions from a binary classifier.
|
| 62 |
+
|
| 63 |
+
Notes
|
| 64 |
+
-----
|
| 65 |
+
The average precision (cf. :func:`~sklearn.metrics.average_precision`) in
|
| 66 |
+
scikit-learn is computed without any interpolation. To be consistent with
|
| 67 |
+
this metric, the precision-recall curve is plotted without any
|
| 68 |
+
interpolation as well (step-wise style).
|
| 69 |
+
|
| 70 |
+
You can change this style by passing the keyword argument
|
| 71 |
+
`drawstyle="default"` in :meth:`plot`, :meth:`from_estimator`, or
|
| 72 |
+
:meth:`from_predictions`. However, the curve will not be strictly
|
| 73 |
+
consistent with the reported average precision.
|
| 74 |
+
|
| 75 |
+
Examples
|
| 76 |
+
--------
|
| 77 |
+
>>> import matplotlib.pyplot as plt
|
| 78 |
+
>>> from sklearn.datasets import make_classification
|
| 79 |
+
>>> from sklearn.metrics import (precision_recall_curve,
|
| 80 |
+
... PrecisionRecallDisplay)
|
| 81 |
+
>>> from sklearn.model_selection import train_test_split
|
| 82 |
+
>>> from sklearn.svm import SVC
|
| 83 |
+
>>> X, y = make_classification(random_state=0)
|
| 84 |
+
>>> X_train, X_test, y_train, y_test = train_test_split(X, y,
|
| 85 |
+
... random_state=0)
|
| 86 |
+
>>> clf = SVC(random_state=0)
|
| 87 |
+
>>> clf.fit(X_train, y_train)
|
| 88 |
+
SVC(random_state=0)
|
| 89 |
+
>>> predictions = clf.predict(X_test)
|
| 90 |
+
>>> precision, recall, _ = precision_recall_curve(y_test, predictions)
|
| 91 |
+
>>> disp = PrecisionRecallDisplay(precision=precision, recall=recall)
|
| 92 |
+
>>> disp.plot()
|
| 93 |
+
<...>
|
| 94 |
+
>>> plt.show()
|
| 95 |
+
"""
|
| 96 |
+
|
| 97 |
+
def __init__(
|
| 98 |
+
self,
|
| 99 |
+
precision,
|
| 100 |
+
recall,
|
| 101 |
+
*,
|
| 102 |
+
average_precision=None,
|
| 103 |
+
estimator_name=None,
|
| 104 |
+
pos_label=None,
|
| 105 |
+
):
|
| 106 |
+
self.estimator_name = estimator_name
|
| 107 |
+
self.precision = precision
|
| 108 |
+
self.recall = recall
|
| 109 |
+
self.average_precision = average_precision
|
| 110 |
+
self.pos_label = pos_label
|
| 111 |
+
|
| 112 |
+
def plot(self, ax=None, *, name=None, **kwargs):
|
| 113 |
+
"""Plot visualization.
|
| 114 |
+
|
| 115 |
+
Extra keyword arguments will be passed to matplotlib's `plot`.
|
| 116 |
+
|
| 117 |
+
Parameters
|
| 118 |
+
----------
|
| 119 |
+
ax : Matplotlib Axes, default=None
|
| 120 |
+
Axes object to plot on. If `None`, a new figure and axes is
|
| 121 |
+
created.
|
| 122 |
+
|
| 123 |
+
name : str, default=None
|
| 124 |
+
Name of precision recall curve for labeling. If `None`, use
|
| 125 |
+
`estimator_name` if not `None`, otherwise no labeling is shown.
|
| 126 |
+
|
| 127 |
+
**kwargs : dict
|
| 128 |
+
Keyword arguments to be passed to matplotlib's `plot`.
|
| 129 |
+
|
| 130 |
+
Returns
|
| 131 |
+
-------
|
| 132 |
+
display : :class:`~sklearn.metrics.PrecisionRecallDisplay`
|
| 133 |
+
Object that stores computed values.
|
| 134 |
+
|
| 135 |
+
Notes
|
| 136 |
+
-----
|
| 137 |
+
The average precision (cf. :func:`~sklearn.metrics.average_precision`)
|
| 138 |
+
in scikit-learn is computed without any interpolation. To be consistent
|
| 139 |
+
with this metric, the precision-recall curve is plotted without any
|
| 140 |
+
interpolation as well (step-wise style).
|
| 141 |
+
|
| 142 |
+
You can change this style by passing the keyword argument
|
| 143 |
+
`drawstyle="default"`. However, the curve will not be strictly
|
| 144 |
+
consistent with the reported average precision.
|
| 145 |
+
"""
|
| 146 |
+
check_matplotlib_support("PrecisionRecallDisplay.plot")
|
| 147 |
+
|
| 148 |
+
name = self.estimator_name if name is None else name
|
| 149 |
+
|
| 150 |
+
line_kwargs = {"drawstyle": "steps-post"}
|
| 151 |
+
if self.average_precision is not None and name is not None:
|
| 152 |
+
line_kwargs["label"] = f"{name} (AP = {self.average_precision:0.2f})"
|
| 153 |
+
elif self.average_precision is not None:
|
| 154 |
+
line_kwargs["label"] = f"AP = {self.average_precision:0.2f}"
|
| 155 |
+
elif name is not None:
|
| 156 |
+
line_kwargs["label"] = name
|
| 157 |
+
line_kwargs.update(**kwargs)
|
| 158 |
+
|
| 159 |
+
import matplotlib.pyplot as plt
|
| 160 |
+
|
| 161 |
+
if ax is None:
|
| 162 |
+
fig, ax = plt.subplots()
|
| 163 |
+
|
| 164 |
+
(self.line_,) = ax.plot(self.recall, self.precision, **line_kwargs)
|
| 165 |
+
info_pos_label = (
|
| 166 |
+
f" (Positive label: {self.pos_label})" if self.pos_label is not None else ""
|
| 167 |
+
)
|
| 168 |
+
|
| 169 |
+
xlabel = "Recall" + info_pos_label
|
| 170 |
+
ylabel = "Precision" + info_pos_label
|
| 171 |
+
ax.set(xlabel=xlabel, ylabel=ylabel)
|
| 172 |
+
|
| 173 |
+
if "label" in line_kwargs:
|
| 174 |
+
ax.legend(loc="lower left")
|
| 175 |
+
|
| 176 |
+
self.ax_ = ax
|
| 177 |
+
self.figure_ = ax.figure
|
| 178 |
+
return self
|
| 179 |
+
|
| 180 |
+
@classmethod
|
| 181 |
+
def from_estimator(
|
| 182 |
+
cls,
|
| 183 |
+
estimator,
|
| 184 |
+
X,
|
| 185 |
+
y,
|
| 186 |
+
*,
|
| 187 |
+
sample_weight=None,
|
| 188 |
+
pos_label=None,
|
| 189 |
+
response_method="auto",
|
| 190 |
+
name=None,
|
| 191 |
+
ax=None,
|
| 192 |
+
**kwargs,
|
| 193 |
+
):
|
| 194 |
+
"""Plot precision-recall curve given an estimator and some data.
|
| 195 |
+
|
| 196 |
+
Parameters
|
| 197 |
+
----------
|
| 198 |
+
estimator : estimator instance
|
| 199 |
+
Fitted classifier or a fitted :class:`~sklearn.pipeline.Pipeline`
|
| 200 |
+
in which the last estimator is a classifier.
|
| 201 |
+
|
| 202 |
+
X : {array-like, sparse matrix} of shape (n_samples, n_features)
|
| 203 |
+
Input values.
|
| 204 |
+
|
| 205 |
+
y : array-like of shape (n_samples,)
|
| 206 |
+
Target values.
|
| 207 |
+
|
| 208 |
+
sample_weight : array-like of shape (n_samples,), default=None
|
| 209 |
+
Sample weights.
|
| 210 |
+
|
| 211 |
+
pos_label : str or int, default=None
|
| 212 |
+
The class considered as the positive class when computing the
|
| 213 |
+
precision and recall metrics. By default, `estimators.classes_[1]`
|
| 214 |
+
is considered as the positive class.
|
| 215 |
+
|
| 216 |
+
response_method : {'predict_proba', 'decision_function', 'auto'}, \
|
| 217 |
+
default='auto'
|
| 218 |
+
Specifies whether to use :term:`predict_proba` or
|
| 219 |
+
:term:`decision_function` as the target response. If set to 'auto',
|
| 220 |
+
:term:`predict_proba` is tried first and if it does not exist
|
| 221 |
+
:term:`decision_function` is tried next.
|
| 222 |
+
|
| 223 |
+
name : str, default=None
|
| 224 |
+
Name for labeling curve. If `None`, no name is used.
|
| 225 |
+
|
| 226 |
+
ax : matplotlib axes, default=None
|
| 227 |
+
Axes object to plot on. If `None`, a new figure and axes is created.
|
| 228 |
+
|
| 229 |
+
**kwargs : dict
|
| 230 |
+
Keyword arguments to be passed to matplotlib's `plot`.
|
| 231 |
+
|
| 232 |
+
Returns
|
| 233 |
+
-------
|
| 234 |
+
display : :class:`~sklearn.metrics.PrecisionRecallDisplay`
|
| 235 |
+
|
| 236 |
+
See Also
|
| 237 |
+
--------
|
| 238 |
+
PrecisionRecallDisplay.from_predictions : Plot precision-recall curve
|
| 239 |
+
using estimated probabilities or output of decision function.
|
| 240 |
+
|
| 241 |
+
Notes
|
| 242 |
+
-----
|
| 243 |
+
The average precision (cf. :func:`~sklearn.metrics.average_precision`)
|
| 244 |
+
in scikit-learn is computed without any interpolation. To be consistent
|
| 245 |
+
with this metric, the precision-recall curve is plotted without any
|
| 246 |
+
interpolation as well (step-wise style).
|
| 247 |
+
|
| 248 |
+
You can change this style by passing the keyword argument
|
| 249 |
+
`drawstyle="default"`. However, the curve will not be strictly
|
| 250 |
+
consistent with the reported average precision.
|
| 251 |
+
|
| 252 |
+
Examples
|
| 253 |
+
--------
|
| 254 |
+
>>> import matplotlib.pyplot as plt
|
| 255 |
+
>>> from sklearn.datasets import make_classification
|
| 256 |
+
>>> from sklearn.metrics import PrecisionRecallDisplay
|
| 257 |
+
>>> from sklearn.model_selection import train_test_split
|
| 258 |
+
>>> from sklearn.linear_model import LogisticRegression
|
| 259 |
+
>>> X, y = make_classification(random_state=0)
|
| 260 |
+
>>> X_train, X_test, y_train, y_test = train_test_split(
|
| 261 |
+
... X, y, random_state=0)
|
| 262 |
+
>>> clf = LogisticRegression()
|
| 263 |
+
>>> clf.fit(X_train, y_train)
|
| 264 |
+
LogisticRegression()
|
| 265 |
+
>>> PrecisionRecallDisplay.from_estimator(
|
| 266 |
+
... clf, X_test, y_test)
|
| 267 |
+
<...>
|
| 268 |
+
>>> plt.show()
|
| 269 |
+
"""
|
| 270 |
+
method_name = f"{cls.__name__}.from_estimator"
|
| 271 |
+
check_matplotlib_support(method_name)
|
| 272 |
+
if not is_classifier(estimator):
|
| 273 |
+
raise ValueError(f"{method_name} only supports classifiers")
|
| 274 |
+
y_pred, pos_label = _get_response(
|
| 275 |
+
X,
|
| 276 |
+
estimator,
|
| 277 |
+
response_method,
|
| 278 |
+
pos_label=pos_label,
|
| 279 |
+
)
|
| 280 |
+
|
| 281 |
+
name = name if name is not None else estimator.__class__.__name__
|
| 282 |
+
|
| 283 |
+
return cls.from_predictions(
|
| 284 |
+
y,
|
| 285 |
+
y_pred,
|
| 286 |
+
sample_weight=sample_weight,
|
| 287 |
+
name=name,
|
| 288 |
+
pos_label=pos_label,
|
| 289 |
+
ax=ax,
|
| 290 |
+
**kwargs,
|
| 291 |
+
)
|
| 292 |
+
|
| 293 |
+
@classmethod
|
| 294 |
+
def from_predictions(
|
| 295 |
+
cls,
|
| 296 |
+
y_true,
|
| 297 |
+
y_pred,
|
| 298 |
+
*,
|
| 299 |
+
sample_weight=None,
|
| 300 |
+
pos_label=None,
|
| 301 |
+
name=None,
|
| 302 |
+
ax=None,
|
| 303 |
+
**kwargs,
|
| 304 |
+
):
|
| 305 |
+
"""Plot precision-recall curve given binary class predictions.
|
| 306 |
+
|
| 307 |
+
Parameters
|
| 308 |
+
----------
|
| 309 |
+
y_true : array-like of shape (n_samples,)
|
| 310 |
+
True binary labels.
|
| 311 |
+
|
| 312 |
+
y_pred : array-like of shape (n_samples,)
|
| 313 |
+
Estimated probabilities or output of decision function.
|
| 314 |
+
|
| 315 |
+
sample_weight : array-like of shape (n_samples,), default=None
|
| 316 |
+
Sample weights.
|
| 317 |
+
|
| 318 |
+
pos_label : str or int, default=None
|
| 319 |
+
The class considered as the positive class when computing the
|
| 320 |
+
precision and recall metrics.
|
| 321 |
+
|
| 322 |
+
name : str, default=None
|
| 323 |
+
Name for labeling curve. If `None`, name will be set to
|
| 324 |
+
`"Classifier"`.
|
| 325 |
+
|
| 326 |
+
ax : matplotlib axes, default=None
|
| 327 |
+
Axes object to plot on. If `None`, a new figure and axes is created.
|
| 328 |
+
|
| 329 |
+
**kwargs : dict
|
| 330 |
+
Keyword arguments to be passed to matplotlib's `plot`.
|
| 331 |
+
|
| 332 |
+
Returns
|
| 333 |
+
-------
|
| 334 |
+
display : :class:`~sklearn.metrics.PrecisionRecallDisplay`
|
| 335 |
+
|
| 336 |
+
See Also
|
| 337 |
+
--------
|
| 338 |
+
PrecisionRecallDisplay.from_estimator : Plot precision-recall curve
|
| 339 |
+
using an estimator.
|
| 340 |
+
|
| 341 |
+
Notes
|
| 342 |
+
-----
|
| 343 |
+
The average precision (cf. :func:`~sklearn.metrics.average_precision`)
|
| 344 |
+
in scikit-learn is computed without any interpolation. To be consistent
|
| 345 |
+
with this metric, the precision-recall curve is plotted without any
|
| 346 |
+
interpolation as well (step-wise style).
|
| 347 |
+
|
| 348 |
+
You can change this style by passing the keyword argument
|
| 349 |
+
`drawstyle="default"`. However, the curve will not be strictly
|
| 350 |
+
consistent with the reported average precision.
|
| 351 |
+
|
| 352 |
+
Examples
|
| 353 |
+
--------
|
| 354 |
+
>>> import matplotlib.pyplot as plt
|
| 355 |
+
>>> from sklearn.datasets import make_classification
|
| 356 |
+
>>> from sklearn.metrics import PrecisionRecallDisplay
|
| 357 |
+
>>> from sklearn.model_selection import train_test_split
|
| 358 |
+
>>> from sklearn.linear_model import LogisticRegression
|
| 359 |
+
>>> X, y = make_classification(random_state=0)
|
| 360 |
+
>>> X_train, X_test, y_train, y_test = train_test_split(
|
| 361 |
+
... X, y, random_state=0)
|
| 362 |
+
>>> clf = LogisticRegression()
|
| 363 |
+
>>> clf.fit(X_train, y_train)
|
| 364 |
+
LogisticRegression()
|
| 365 |
+
>>> y_pred = clf.predict_proba(X_test)[:, 1]
|
| 366 |
+
>>> PrecisionRecallDisplay.from_predictions(
|
| 367 |
+
... y_test, y_pred)
|
| 368 |
+
<...>
|
| 369 |
+
>>> plt.show()
|
| 370 |
+
"""
|
| 371 |
+
check_matplotlib_support(f"{cls.__name__}.from_predictions")
|
| 372 |
+
|
| 373 |
+
check_consistent_length(y_true, y_pred, sample_weight)
|
| 374 |
+
pos_label = _check_pos_label_consistency(pos_label, y_true)
|
| 375 |
+
|
| 376 |
+
precision, recall, _ = precision_recall_curve(
|
| 377 |
+
y_true, y_pred, pos_label=pos_label, sample_weight=sample_weight
|
| 378 |
+
)
|
| 379 |
+
average_precision = average_precision_score(
|
| 380 |
+
y_true, y_pred, pos_label=pos_label, sample_weight=sample_weight
|
| 381 |
+
)
|
| 382 |
+
|
| 383 |
+
name = name if name is not None else "Classifier"
|
| 384 |
+
|
| 385 |
+
viz = PrecisionRecallDisplay(
|
| 386 |
+
precision=precision,
|
| 387 |
+
recall=recall,
|
| 388 |
+
average_precision=average_precision,
|
| 389 |
+
estimator_name=name,
|
| 390 |
+
pos_label=pos_label,
|
| 391 |
+
)
|
| 392 |
+
|
| 393 |
+
return viz.plot(ax=ax, name=name, **kwargs)
|
videollama2/lib/python3.10/site-packages/sklearn/metrics/_plot/regression.py
ADDED
|
@@ -0,0 +1,406 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import numbers
|
| 2 |
+
|
| 3 |
+
import numpy as np
|
| 4 |
+
|
| 5 |
+
from ...utils import check_matplotlib_support
|
| 6 |
+
from ...utils import check_random_state
|
| 7 |
+
from ...utils import _safe_indexing
|
| 8 |
+
|
| 9 |
+
|
| 10 |
+
class PredictionErrorDisplay:
|
| 11 |
+
"""Visualization of the prediction error of a regression model.
|
| 12 |
+
|
| 13 |
+
This tool can display "residuals vs predicted" or "actual vs predicted"
|
| 14 |
+
using scatter plots to qualitatively assess the behavior of a regressor,
|
| 15 |
+
preferably on held-out data points.
|
| 16 |
+
|
| 17 |
+
See the details in the docstrings of
|
| 18 |
+
:func:`~sklearn.metrics.PredictionErrorDisplay.from_estimator` or
|
| 19 |
+
:func:`~sklearn.metrics.PredictionErrorDisplay.from_predictions` to
|
| 20 |
+
create a visualizer. All parameters are stored as attributes.
|
| 21 |
+
|
| 22 |
+
For general information regarding `scikit-learn` visualization tools, read
|
| 23 |
+
more in the :ref:`Visualization Guide <visualizations>`.
|
| 24 |
+
For details regarding interpreting these plots, refer to the
|
| 25 |
+
:ref:`Model Evaluation Guide <visualization_regression_evaluation>`.
|
| 26 |
+
|
| 27 |
+
.. versionadded:: 1.2
|
| 28 |
+
|
| 29 |
+
Parameters
|
| 30 |
+
----------
|
| 31 |
+
y_true : ndarray of shape (n_samples,)
|
| 32 |
+
True values.
|
| 33 |
+
|
| 34 |
+
y_pred : ndarray of shape (n_samples,)
|
| 35 |
+
Prediction values.
|
| 36 |
+
|
| 37 |
+
Attributes
|
| 38 |
+
----------
|
| 39 |
+
line_ : matplotlib Artist
|
| 40 |
+
Optimal line representing `y_true == y_pred`. Therefore, it is a
|
| 41 |
+
diagonal line for `kind="predictions"` and a horizontal line for
|
| 42 |
+
`kind="residuals"`.
|
| 43 |
+
|
| 44 |
+
errors_lines_ : matplotlib Artist or None
|
| 45 |
+
Residual lines. If `with_errors=False`, then it is set to `None`.
|
| 46 |
+
|
| 47 |
+
scatter_ : matplotlib Artist
|
| 48 |
+
Scatter data points.
|
| 49 |
+
|
| 50 |
+
ax_ : matplotlib Axes
|
| 51 |
+
Axes with the different matplotlib axis.
|
| 52 |
+
|
| 53 |
+
figure_ : matplotlib Figure
|
| 54 |
+
Figure containing the scatter and lines.
|
| 55 |
+
|
| 56 |
+
See Also
|
| 57 |
+
--------
|
| 58 |
+
PredictionErrorDisplay.from_estimator : Prediction error visualization
|
| 59 |
+
given an estimator and some data.
|
| 60 |
+
PredictionErrorDisplay.from_predictions : Prediction error visualization
|
| 61 |
+
given the true and predicted targets.
|
| 62 |
+
|
| 63 |
+
Examples
|
| 64 |
+
--------
|
| 65 |
+
>>> import matplotlib.pyplot as plt
|
| 66 |
+
>>> from sklearn.datasets import load_diabetes
|
| 67 |
+
>>> from sklearn.linear_model import Ridge
|
| 68 |
+
>>> from sklearn.metrics import PredictionErrorDisplay
|
| 69 |
+
>>> X, y = load_diabetes(return_X_y=True)
|
| 70 |
+
>>> ridge = Ridge().fit(X, y)
|
| 71 |
+
>>> y_pred = ridge.predict(X)
|
| 72 |
+
>>> display = PredictionErrorDisplay(y_true=y, y_pred=y_pred)
|
| 73 |
+
>>> display.plot()
|
| 74 |
+
<...>
|
| 75 |
+
>>> plt.show()
|
| 76 |
+
"""
|
| 77 |
+
|
| 78 |
+
def __init__(self, *, y_true, y_pred):
|
| 79 |
+
self.y_true = y_true
|
| 80 |
+
self.y_pred = y_pred
|
| 81 |
+
|
| 82 |
+
def plot(
|
| 83 |
+
self,
|
| 84 |
+
ax=None,
|
| 85 |
+
*,
|
| 86 |
+
kind="residual_vs_predicted",
|
| 87 |
+
scatter_kwargs=None,
|
| 88 |
+
line_kwargs=None,
|
| 89 |
+
):
|
| 90 |
+
"""Plot visualization.
|
| 91 |
+
|
| 92 |
+
Extra keyword arguments will be passed to matplotlib's ``plot``.
|
| 93 |
+
|
| 94 |
+
Parameters
|
| 95 |
+
----------
|
| 96 |
+
ax : matplotlib axes, default=None
|
| 97 |
+
Axes object to plot on. If `None`, a new figure and axes is
|
| 98 |
+
created.
|
| 99 |
+
|
| 100 |
+
kind : {"actual_vs_predicted", "residual_vs_predicted"}, \
|
| 101 |
+
default="residual_vs_predicted"
|
| 102 |
+
The type of plot to draw:
|
| 103 |
+
|
| 104 |
+
- "actual_vs_predicted" draws the the observed values (y-axis) vs.
|
| 105 |
+
the predicted values (x-axis).
|
| 106 |
+
- "residual_vs_predicted" draws the residuals, i.e difference
|
| 107 |
+
between observed and predicted values, (y-axis) vs. the predicted
|
| 108 |
+
values (x-axis).
|
| 109 |
+
|
| 110 |
+
scatter_kwargs : dict, default=None
|
| 111 |
+
Dictionary with keywords passed to the `matplotlib.pyplot.scatter`
|
| 112 |
+
call.
|
| 113 |
+
|
| 114 |
+
line_kwargs : dict, default=None
|
| 115 |
+
Dictionary with keyword passed to the `matplotlib.pyplot.plot`
|
| 116 |
+
call to draw the optimal line.
|
| 117 |
+
|
| 118 |
+
Returns
|
| 119 |
+
-------
|
| 120 |
+
display : :class:`~sklearn.metrics.plot.PredictionErrorDisplay`
|
| 121 |
+
Object that stores computed values.
|
| 122 |
+
"""
|
| 123 |
+
check_matplotlib_support(f"{self.__class__.__name__}.plot")
|
| 124 |
+
|
| 125 |
+
expected_kind = ("actual_vs_predicted", "residual_vs_predicted")
|
| 126 |
+
if kind not in expected_kind:
|
| 127 |
+
raise ValueError(
|
| 128 |
+
f"`kind` must be one of {', '.join(expected_kind)}. "
|
| 129 |
+
f"Got {kind!r} instead."
|
| 130 |
+
)
|
| 131 |
+
|
| 132 |
+
import matplotlib.pyplot as plt
|
| 133 |
+
|
| 134 |
+
if scatter_kwargs is None:
|
| 135 |
+
scatter_kwargs = {}
|
| 136 |
+
if line_kwargs is None:
|
| 137 |
+
line_kwargs = {}
|
| 138 |
+
|
| 139 |
+
default_scatter_kwargs = {"color": "tab:blue", "alpha": 0.8}
|
| 140 |
+
default_line_kwargs = {"color": "black", "alpha": 0.7, "linestyle": "--"}
|
| 141 |
+
|
| 142 |
+
scatter_kwargs = {**default_scatter_kwargs, **scatter_kwargs}
|
| 143 |
+
line_kwargs = {**default_line_kwargs, **line_kwargs}
|
| 144 |
+
|
| 145 |
+
if ax is None:
|
| 146 |
+
_, ax = plt.subplots()
|
| 147 |
+
|
| 148 |
+
if kind == "actual_vs_predicted":
|
| 149 |
+
max_value = max(np.max(self.y_true), np.max(self.y_pred))
|
| 150 |
+
min_value = min(np.min(self.y_true), np.min(self.y_pred))
|
| 151 |
+
self.line_ = ax.plot(
|
| 152 |
+
[min_value, max_value], [min_value, max_value], **line_kwargs
|
| 153 |
+
)[0]
|
| 154 |
+
|
| 155 |
+
x_data, y_data = self.y_pred, self.y_true
|
| 156 |
+
xlabel, ylabel = "Predicted values", "Actual values"
|
| 157 |
+
|
| 158 |
+
self.scatter_ = ax.scatter(x_data, y_data, **scatter_kwargs)
|
| 159 |
+
|
| 160 |
+
# force to have a squared axis
|
| 161 |
+
ax.set_aspect("equal", adjustable="datalim")
|
| 162 |
+
ax.set_xticks(np.linspace(min_value, max_value, num=5))
|
| 163 |
+
ax.set_yticks(np.linspace(min_value, max_value, num=5))
|
| 164 |
+
else: # kind == "residual_vs_predicted"
|
| 165 |
+
self.line_ = ax.plot(
|
| 166 |
+
[np.min(self.y_pred), np.max(self.y_pred)],
|
| 167 |
+
[0, 0],
|
| 168 |
+
**line_kwargs,
|
| 169 |
+
)[0]
|
| 170 |
+
self.scatter_ = ax.scatter(
|
| 171 |
+
self.y_pred, self.y_true - self.y_pred, **scatter_kwargs
|
| 172 |
+
)
|
| 173 |
+
xlabel, ylabel = "Predicted values", "Residuals (actual - predicted)"
|
| 174 |
+
|
| 175 |
+
ax.set(xlabel=xlabel, ylabel=ylabel)
|
| 176 |
+
|
| 177 |
+
self.ax_ = ax
|
| 178 |
+
self.figure_ = ax.figure
|
| 179 |
+
|
| 180 |
+
return self
|
| 181 |
+
|
| 182 |
+
@classmethod
|
| 183 |
+
def from_estimator(
|
| 184 |
+
cls,
|
| 185 |
+
estimator,
|
| 186 |
+
X,
|
| 187 |
+
y,
|
| 188 |
+
*,
|
| 189 |
+
kind="residual_vs_predicted",
|
| 190 |
+
subsample=1_000,
|
| 191 |
+
random_state=None,
|
| 192 |
+
ax=None,
|
| 193 |
+
scatter_kwargs=None,
|
| 194 |
+
line_kwargs=None,
|
| 195 |
+
):
|
| 196 |
+
"""Plot the prediction error given a regressor and some data.
|
| 197 |
+
|
| 198 |
+
For general information regarding `scikit-learn` visualization tools,
|
| 199 |
+
read more in the :ref:`Visualization Guide <visualizations>`.
|
| 200 |
+
For details regarding interpreting these plots, refer to the
|
| 201 |
+
:ref:`Model Evaluation Guide <visualization_regression_evaluation>`.
|
| 202 |
+
|
| 203 |
+
.. versionadded:: 1.2
|
| 204 |
+
|
| 205 |
+
Parameters
|
| 206 |
+
----------
|
| 207 |
+
estimator : estimator instance
|
| 208 |
+
Fitted regressor or a fitted :class:`~sklearn.pipeline.Pipeline`
|
| 209 |
+
in which the last estimator is a regressor.
|
| 210 |
+
|
| 211 |
+
X : {array-like, sparse matrix} of shape (n_samples, n_features)
|
| 212 |
+
Input values.
|
| 213 |
+
|
| 214 |
+
y : array-like of shape (n_samples,)
|
| 215 |
+
Target values.
|
| 216 |
+
|
| 217 |
+
kind : {"actual_vs_predicted", "residual_vs_predicted"}, \
|
| 218 |
+
default="residual_vs_predicted"
|
| 219 |
+
The type of plot to draw:
|
| 220 |
+
|
| 221 |
+
- "actual_vs_predicted" draws the the observed values (y-axis) vs.
|
| 222 |
+
the predicted values (x-axis).
|
| 223 |
+
- "residual_vs_predicted" draws the residuals, i.e difference
|
| 224 |
+
between observed and predicted values, (y-axis) vs. the predicted
|
| 225 |
+
values (x-axis).
|
| 226 |
+
|
| 227 |
+
subsample : float, int or None, default=1_000
|
| 228 |
+
Sampling the samples to be shown on the scatter plot. If `float`,
|
| 229 |
+
it should be between 0 and 1 and represents the proportion of the
|
| 230 |
+
original dataset. If `int`, it represents the number of samples
|
| 231 |
+
display on the scatter plot. If `None`, no subsampling will be
|
| 232 |
+
applied. by default, a 1000 samples or less will be displayed.
|
| 233 |
+
|
| 234 |
+
random_state : int or RandomState, default=None
|
| 235 |
+
Controls the randomness when `subsample` is not `None`.
|
| 236 |
+
See :term:`Glossary <random_state>` for details.
|
| 237 |
+
|
| 238 |
+
ax : matplotlib axes, default=None
|
| 239 |
+
Axes object to plot on. If `None`, a new figure and axes is
|
| 240 |
+
created.
|
| 241 |
+
|
| 242 |
+
scatter_kwargs : dict, default=None
|
| 243 |
+
Dictionary with keywords passed to the `matplotlib.pyplot.scatter`
|
| 244 |
+
call.
|
| 245 |
+
|
| 246 |
+
line_kwargs : dict, default=None
|
| 247 |
+
Dictionary with keyword passed to the `matplotlib.pyplot.plot`
|
| 248 |
+
call to draw the optimal line.
|
| 249 |
+
|
| 250 |
+
Returns
|
| 251 |
+
-------
|
| 252 |
+
display : :class:`~sklearn.metrics.PredictionErrorDisplay`
|
| 253 |
+
Object that stores the computed values.
|
| 254 |
+
|
| 255 |
+
See Also
|
| 256 |
+
--------
|
| 257 |
+
PredictionErrorDisplay : Prediction error visualization for regression.
|
| 258 |
+
PredictionErrorDisplay.from_predictions : Prediction error visualization
|
| 259 |
+
given the true and predicted targets.
|
| 260 |
+
|
| 261 |
+
Examples
|
| 262 |
+
--------
|
| 263 |
+
>>> import matplotlib.pyplot as plt
|
| 264 |
+
>>> from sklearn.datasets import load_diabetes
|
| 265 |
+
>>> from sklearn.linear_model import Ridge
|
| 266 |
+
>>> from sklearn.metrics import PredictionErrorDisplay
|
| 267 |
+
>>> X, y = load_diabetes(return_X_y=True)
|
| 268 |
+
>>> ridge = Ridge().fit(X, y)
|
| 269 |
+
>>> disp = PredictionErrorDisplay.from_estimator(ridge, X, y)
|
| 270 |
+
>>> plt.show()
|
| 271 |
+
"""
|
| 272 |
+
check_matplotlib_support(f"{cls.__name__}.from_estimator")
|
| 273 |
+
|
| 274 |
+
y_pred = estimator.predict(X)
|
| 275 |
+
|
| 276 |
+
return cls.from_predictions(
|
| 277 |
+
y_true=y,
|
| 278 |
+
y_pred=y_pred,
|
| 279 |
+
kind=kind,
|
| 280 |
+
subsample=subsample,
|
| 281 |
+
random_state=random_state,
|
| 282 |
+
ax=ax,
|
| 283 |
+
scatter_kwargs=scatter_kwargs,
|
| 284 |
+
line_kwargs=line_kwargs,
|
| 285 |
+
)
|
| 286 |
+
|
| 287 |
+
@classmethod
|
| 288 |
+
def from_predictions(
|
| 289 |
+
cls,
|
| 290 |
+
y_true,
|
| 291 |
+
y_pred,
|
| 292 |
+
*,
|
| 293 |
+
kind="residual_vs_predicted",
|
| 294 |
+
subsample=1_000,
|
| 295 |
+
random_state=None,
|
| 296 |
+
ax=None,
|
| 297 |
+
scatter_kwargs=None,
|
| 298 |
+
line_kwargs=None,
|
| 299 |
+
):
|
| 300 |
+
"""Plot the prediction error given the true and predicted targets.
|
| 301 |
+
|
| 302 |
+
For general information regarding `scikit-learn` visualization tools,
|
| 303 |
+
read more in the :ref:`Visualization Guide <visualizations>`.
|
| 304 |
+
For details regarding interpreting these plots, refer to the
|
| 305 |
+
:ref:`Model Evaluation Guide <visualization_regression_evaluation>`.
|
| 306 |
+
|
| 307 |
+
.. versionadded:: 1.2
|
| 308 |
+
|
| 309 |
+
Parameters
|
| 310 |
+
----------
|
| 311 |
+
y_true : array-like of shape (n_samples,)
|
| 312 |
+
True target values.
|
| 313 |
+
|
| 314 |
+
y_pred : array-like of shape (n_samples,)
|
| 315 |
+
Predicted target values.
|
| 316 |
+
|
| 317 |
+
kind : {"actual_vs_predicted", "residual_vs_predicted"}, \
|
| 318 |
+
default="residual_vs_predicted"
|
| 319 |
+
The type of plot to draw:
|
| 320 |
+
|
| 321 |
+
- "actual_vs_predicted" draws the the observed values (y-axis) vs.
|
| 322 |
+
the predicted values (x-axis).
|
| 323 |
+
- "residual_vs_predicted" draws the residuals, i.e difference
|
| 324 |
+
between observed and predicted values, (y-axis) vs. the predicted
|
| 325 |
+
values (x-axis).
|
| 326 |
+
|
| 327 |
+
subsample : float, int or None, default=1_000
|
| 328 |
+
Sampling the samples to be shown on the scatter plot. If `float`,
|
| 329 |
+
it should be between 0 and 1 and represents the proportion of the
|
| 330 |
+
original dataset. If `int`, it represents the number of samples
|
| 331 |
+
display on the scatter plot. If `None`, no subsampling will be
|
| 332 |
+
applied. by default, a 1000 samples or less will be displayed.
|
| 333 |
+
|
| 334 |
+
random_state : int or RandomState, default=None
|
| 335 |
+
Controls the randomness when `subsample` is not `None`.
|
| 336 |
+
See :term:`Glossary <random_state>` for details.
|
| 337 |
+
|
| 338 |
+
ax : matplotlib axes, default=None
|
| 339 |
+
Axes object to plot on. If `None`, a new figure and axes is
|
| 340 |
+
created.
|
| 341 |
+
|
| 342 |
+
scatter_kwargs : dict, default=None
|
| 343 |
+
Dictionary with keywords passed to the `matplotlib.pyplot.scatter`
|
| 344 |
+
call.
|
| 345 |
+
|
| 346 |
+
line_kwargs : dict, default=None
|
| 347 |
+
Dictionary with keyword passed to the `matplotlib.pyplot.plot`
|
| 348 |
+
call to draw the optimal line.
|
| 349 |
+
|
| 350 |
+
Returns
|
| 351 |
+
-------
|
| 352 |
+
display : :class:`~sklearn.metrics.PredictionErrorDisplay`
|
| 353 |
+
Object that stores the computed values.
|
| 354 |
+
|
| 355 |
+
See Also
|
| 356 |
+
--------
|
| 357 |
+
PredictionErrorDisplay : Prediction error visualization for regression.
|
| 358 |
+
PredictionErrorDisplay.from_estimator : Prediction error visualization
|
| 359 |
+
given an estimator and some data.
|
| 360 |
+
|
| 361 |
+
Examples
|
| 362 |
+
--------
|
| 363 |
+
>>> import matplotlib.pyplot as plt
|
| 364 |
+
>>> from sklearn.datasets import load_diabetes
|
| 365 |
+
>>> from sklearn.linear_model import Ridge
|
| 366 |
+
>>> from sklearn.metrics import PredictionErrorDisplay
|
| 367 |
+
>>> X, y = load_diabetes(return_X_y=True)
|
| 368 |
+
>>> ridge = Ridge().fit(X, y)
|
| 369 |
+
>>> y_pred = ridge.predict(X)
|
| 370 |
+
>>> disp = PredictionErrorDisplay.from_predictions(y_true=y, y_pred=y_pred)
|
| 371 |
+
>>> plt.show()
|
| 372 |
+
"""
|
| 373 |
+
check_matplotlib_support(f"{cls.__name__}.from_predictions")
|
| 374 |
+
|
| 375 |
+
random_state = check_random_state(random_state)
|
| 376 |
+
|
| 377 |
+
n_samples = len(y_true)
|
| 378 |
+
if isinstance(subsample, numbers.Integral):
|
| 379 |
+
if subsample <= 0:
|
| 380 |
+
raise ValueError(
|
| 381 |
+
f"When an integer, subsample={subsample} should be positive."
|
| 382 |
+
)
|
| 383 |
+
elif isinstance(subsample, numbers.Real):
|
| 384 |
+
if subsample <= 0 or subsample >= 1:
|
| 385 |
+
raise ValueError(
|
| 386 |
+
f"When a floating-point, subsample={subsample} should"
|
| 387 |
+
" be in the (0, 1) range."
|
| 388 |
+
)
|
| 389 |
+
subsample = int(n_samples * subsample)
|
| 390 |
+
|
| 391 |
+
if subsample is not None and subsample < n_samples:
|
| 392 |
+
indices = random_state.choice(np.arange(n_samples), size=subsample)
|
| 393 |
+
y_true = _safe_indexing(y_true, indices, axis=0)
|
| 394 |
+
y_pred = _safe_indexing(y_pred, indices, axis=0)
|
| 395 |
+
|
| 396 |
+
viz = PredictionErrorDisplay(
|
| 397 |
+
y_true=y_true,
|
| 398 |
+
y_pred=y_pred,
|
| 399 |
+
)
|
| 400 |
+
|
| 401 |
+
return viz.plot(
|
| 402 |
+
ax=ax,
|
| 403 |
+
kind=kind,
|
| 404 |
+
scatter_kwargs=scatter_kwargs,
|
| 405 |
+
line_kwargs=line_kwargs,
|
| 406 |
+
)
|
videollama2/lib/python3.10/site-packages/sklearn/metrics/_plot/roc_curve.py
ADDED
|
@@ -0,0 +1,352 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from .base import _get_response
|
| 2 |
+
|
| 3 |
+
from .. import auc
|
| 4 |
+
from .. import roc_curve
|
| 5 |
+
from .._base import _check_pos_label_consistency
|
| 6 |
+
|
| 7 |
+
from ...utils import check_matplotlib_support
|
| 8 |
+
|
| 9 |
+
|
| 10 |
+
class RocCurveDisplay:
|
| 11 |
+
"""ROC Curve visualization.
|
| 12 |
+
|
| 13 |
+
It is recommend to use
|
| 14 |
+
:func:`~sklearn.metrics.RocCurveDisplay.from_estimator` or
|
| 15 |
+
:func:`~sklearn.metrics.RocCurveDisplay.from_predictions` to create
|
| 16 |
+
a :class:`~sklearn.metrics.RocCurveDisplay`. All parameters are
|
| 17 |
+
stored as attributes.
|
| 18 |
+
|
| 19 |
+
Read more in the :ref:`User Guide <visualizations>`.
|
| 20 |
+
|
| 21 |
+
Parameters
|
| 22 |
+
----------
|
| 23 |
+
fpr : ndarray
|
| 24 |
+
False positive rate.
|
| 25 |
+
|
| 26 |
+
tpr : ndarray
|
| 27 |
+
True positive rate.
|
| 28 |
+
|
| 29 |
+
roc_auc : float, default=None
|
| 30 |
+
Area under ROC curve. If None, the roc_auc score is not shown.
|
| 31 |
+
|
| 32 |
+
estimator_name : str, default=None
|
| 33 |
+
Name of estimator. If None, the estimator name is not shown.
|
| 34 |
+
|
| 35 |
+
pos_label : str or int, default=None
|
| 36 |
+
The class considered as the positive class when computing the roc auc
|
| 37 |
+
metrics. By default, `estimators.classes_[1]` is considered
|
| 38 |
+
as the positive class.
|
| 39 |
+
|
| 40 |
+
.. versionadded:: 0.24
|
| 41 |
+
|
| 42 |
+
Attributes
|
| 43 |
+
----------
|
| 44 |
+
line_ : matplotlib Artist
|
| 45 |
+
ROC Curve.
|
| 46 |
+
|
| 47 |
+
ax_ : matplotlib Axes
|
| 48 |
+
Axes with ROC Curve.
|
| 49 |
+
|
| 50 |
+
figure_ : matplotlib Figure
|
| 51 |
+
Figure containing the curve.
|
| 52 |
+
|
| 53 |
+
See Also
|
| 54 |
+
--------
|
| 55 |
+
roc_curve : Compute Receiver operating characteristic (ROC) curve.
|
| 56 |
+
RocCurveDisplay.from_estimator : Plot Receiver Operating Characteristic
|
| 57 |
+
(ROC) curve given an estimator and some data.
|
| 58 |
+
RocCurveDisplay.from_predictions : Plot Receiver Operating Characteristic
|
| 59 |
+
(ROC) curve given the true and predicted values.
|
| 60 |
+
roc_auc_score : Compute the area under the ROC curve.
|
| 61 |
+
|
| 62 |
+
Examples
|
| 63 |
+
--------
|
| 64 |
+
>>> import matplotlib.pyplot as plt
|
| 65 |
+
>>> import numpy as np
|
| 66 |
+
>>> from sklearn import metrics
|
| 67 |
+
>>> y = np.array([0, 0, 1, 1])
|
| 68 |
+
>>> pred = np.array([0.1, 0.4, 0.35, 0.8])
|
| 69 |
+
>>> fpr, tpr, thresholds = metrics.roc_curve(y, pred)
|
| 70 |
+
>>> roc_auc = metrics.auc(fpr, tpr)
|
| 71 |
+
>>> display = metrics.RocCurveDisplay(fpr=fpr, tpr=tpr, roc_auc=roc_auc,
|
| 72 |
+
... estimator_name='example estimator')
|
| 73 |
+
>>> display.plot()
|
| 74 |
+
<...>
|
| 75 |
+
>>> plt.show()
|
| 76 |
+
"""
|
| 77 |
+
|
| 78 |
+
def __init__(self, *, fpr, tpr, roc_auc=None, estimator_name=None, pos_label=None):
|
| 79 |
+
self.estimator_name = estimator_name
|
| 80 |
+
self.fpr = fpr
|
| 81 |
+
self.tpr = tpr
|
| 82 |
+
self.roc_auc = roc_auc
|
| 83 |
+
self.pos_label = pos_label
|
| 84 |
+
|
| 85 |
+
def plot(self, ax=None, *, name=None, **kwargs):
|
| 86 |
+
"""Plot visualization.
|
| 87 |
+
|
| 88 |
+
Extra keyword arguments will be passed to matplotlib's ``plot``.
|
| 89 |
+
|
| 90 |
+
Parameters
|
| 91 |
+
----------
|
| 92 |
+
ax : matplotlib axes, default=None
|
| 93 |
+
Axes object to plot on. If `None`, a new figure and axes is
|
| 94 |
+
created.
|
| 95 |
+
|
| 96 |
+
name : str, default=None
|
| 97 |
+
Name of ROC Curve for labeling. If `None`, use `estimator_name` if
|
| 98 |
+
not `None`, otherwise no labeling is shown.
|
| 99 |
+
|
| 100 |
+
**kwargs : dict
|
| 101 |
+
Keyword arguments to be passed to matplotlib's `plot`.
|
| 102 |
+
|
| 103 |
+
Returns
|
| 104 |
+
-------
|
| 105 |
+
display : :class:`~sklearn.metrics.plot.RocCurveDisplay`
|
| 106 |
+
Object that stores computed values.
|
| 107 |
+
"""
|
| 108 |
+
check_matplotlib_support("RocCurveDisplay.plot")
|
| 109 |
+
|
| 110 |
+
name = self.estimator_name if name is None else name
|
| 111 |
+
|
| 112 |
+
line_kwargs = {}
|
| 113 |
+
if self.roc_auc is not None and name is not None:
|
| 114 |
+
line_kwargs["label"] = f"{name} (AUC = {self.roc_auc:0.2f})"
|
| 115 |
+
elif self.roc_auc is not None:
|
| 116 |
+
line_kwargs["label"] = f"AUC = {self.roc_auc:0.2f}"
|
| 117 |
+
elif name is not None:
|
| 118 |
+
line_kwargs["label"] = name
|
| 119 |
+
|
| 120 |
+
line_kwargs.update(**kwargs)
|
| 121 |
+
|
| 122 |
+
import matplotlib.pyplot as plt
|
| 123 |
+
|
| 124 |
+
if ax is None:
|
| 125 |
+
fig, ax = plt.subplots()
|
| 126 |
+
|
| 127 |
+
(self.line_,) = ax.plot(self.fpr, self.tpr, **line_kwargs)
|
| 128 |
+
info_pos_label = (
|
| 129 |
+
f" (Positive label: {self.pos_label})" if self.pos_label is not None else ""
|
| 130 |
+
)
|
| 131 |
+
|
| 132 |
+
xlabel = "False Positive Rate" + info_pos_label
|
| 133 |
+
ylabel = "True Positive Rate" + info_pos_label
|
| 134 |
+
ax.set(xlabel=xlabel, ylabel=ylabel)
|
| 135 |
+
|
| 136 |
+
if "label" in line_kwargs:
|
| 137 |
+
ax.legend(loc="lower right")
|
| 138 |
+
|
| 139 |
+
self.ax_ = ax
|
| 140 |
+
self.figure_ = ax.figure
|
| 141 |
+
return self
|
| 142 |
+
|
| 143 |
+
@classmethod
|
| 144 |
+
def from_estimator(
|
| 145 |
+
cls,
|
| 146 |
+
estimator,
|
| 147 |
+
X,
|
| 148 |
+
y,
|
| 149 |
+
*,
|
| 150 |
+
sample_weight=None,
|
| 151 |
+
drop_intermediate=True,
|
| 152 |
+
response_method="auto",
|
| 153 |
+
pos_label=None,
|
| 154 |
+
name=None,
|
| 155 |
+
ax=None,
|
| 156 |
+
**kwargs,
|
| 157 |
+
):
|
| 158 |
+
"""Create a ROC Curve display from an estimator.
|
| 159 |
+
|
| 160 |
+
Parameters
|
| 161 |
+
----------
|
| 162 |
+
estimator : estimator instance
|
| 163 |
+
Fitted classifier or a fitted :class:`~sklearn.pipeline.Pipeline`
|
| 164 |
+
in which the last estimator is a classifier.
|
| 165 |
+
|
| 166 |
+
X : {array-like, sparse matrix} of shape (n_samples, n_features)
|
| 167 |
+
Input values.
|
| 168 |
+
|
| 169 |
+
y : array-like of shape (n_samples,)
|
| 170 |
+
Target values.
|
| 171 |
+
|
| 172 |
+
sample_weight : array-like of shape (n_samples,), default=None
|
| 173 |
+
Sample weights.
|
| 174 |
+
|
| 175 |
+
drop_intermediate : bool, default=True
|
| 176 |
+
Whether to drop some suboptimal thresholds which would not appear
|
| 177 |
+
on a plotted ROC curve. This is useful in order to create lighter
|
| 178 |
+
ROC curves.
|
| 179 |
+
|
| 180 |
+
response_method : {'predict_proba', 'decision_function', 'auto'} \
|
| 181 |
+
default='auto'
|
| 182 |
+
Specifies whether to use :term:`predict_proba` or
|
| 183 |
+
:term:`decision_function` as the target response. If set to 'auto',
|
| 184 |
+
:term:`predict_proba` is tried first and if it does not exist
|
| 185 |
+
:term:`decision_function` is tried next.
|
| 186 |
+
|
| 187 |
+
pos_label : str or int, default=None
|
| 188 |
+
The class considered as the positive class when computing the roc auc
|
| 189 |
+
metrics. By default, `estimators.classes_[1]` is considered
|
| 190 |
+
as the positive class.
|
| 191 |
+
|
| 192 |
+
name : str, default=None
|
| 193 |
+
Name of ROC Curve for labeling. If `None`, use the name of the
|
| 194 |
+
estimator.
|
| 195 |
+
|
| 196 |
+
ax : matplotlib axes, default=None
|
| 197 |
+
Axes object to plot on. If `None`, a new figure and axes is created.
|
| 198 |
+
|
| 199 |
+
**kwargs : dict
|
| 200 |
+
Keyword arguments to be passed to matplotlib's `plot`.
|
| 201 |
+
|
| 202 |
+
Returns
|
| 203 |
+
-------
|
| 204 |
+
display : :class:`~sklearn.metrics.plot.RocCurveDisplay`
|
| 205 |
+
The ROC Curve display.
|
| 206 |
+
|
| 207 |
+
See Also
|
| 208 |
+
--------
|
| 209 |
+
roc_curve : Compute Receiver operating characteristic (ROC) curve.
|
| 210 |
+
RocCurveDisplay.from_predictions : ROC Curve visualization given the
|
| 211 |
+
probabilities of scores of a classifier.
|
| 212 |
+
roc_auc_score : Compute the area under the ROC curve.
|
| 213 |
+
|
| 214 |
+
Examples
|
| 215 |
+
--------
|
| 216 |
+
>>> import matplotlib.pyplot as plt
|
| 217 |
+
>>> from sklearn.datasets import make_classification
|
| 218 |
+
>>> from sklearn.metrics import RocCurveDisplay
|
| 219 |
+
>>> from sklearn.model_selection import train_test_split
|
| 220 |
+
>>> from sklearn.svm import SVC
|
| 221 |
+
>>> X, y = make_classification(random_state=0)
|
| 222 |
+
>>> X_train, X_test, y_train, y_test = train_test_split(
|
| 223 |
+
... X, y, random_state=0)
|
| 224 |
+
>>> clf = SVC(random_state=0).fit(X_train, y_train)
|
| 225 |
+
>>> RocCurveDisplay.from_estimator(
|
| 226 |
+
... clf, X_test, y_test)
|
| 227 |
+
<...>
|
| 228 |
+
>>> plt.show()
|
| 229 |
+
"""
|
| 230 |
+
check_matplotlib_support(f"{cls.__name__}.from_estimator")
|
| 231 |
+
|
| 232 |
+
name = estimator.__class__.__name__ if name is None else name
|
| 233 |
+
|
| 234 |
+
y_pred, pos_label = _get_response(
|
| 235 |
+
X,
|
| 236 |
+
estimator,
|
| 237 |
+
response_method=response_method,
|
| 238 |
+
pos_label=pos_label,
|
| 239 |
+
)
|
| 240 |
+
|
| 241 |
+
return cls.from_predictions(
|
| 242 |
+
y_true=y,
|
| 243 |
+
y_pred=y_pred,
|
| 244 |
+
sample_weight=sample_weight,
|
| 245 |
+
drop_intermediate=drop_intermediate,
|
| 246 |
+
name=name,
|
| 247 |
+
ax=ax,
|
| 248 |
+
pos_label=pos_label,
|
| 249 |
+
**kwargs,
|
| 250 |
+
)
|
| 251 |
+
|
| 252 |
+
@classmethod
|
| 253 |
+
def from_predictions(
|
| 254 |
+
cls,
|
| 255 |
+
y_true,
|
| 256 |
+
y_pred,
|
| 257 |
+
*,
|
| 258 |
+
sample_weight=None,
|
| 259 |
+
drop_intermediate=True,
|
| 260 |
+
pos_label=None,
|
| 261 |
+
name=None,
|
| 262 |
+
ax=None,
|
| 263 |
+
**kwargs,
|
| 264 |
+
):
|
| 265 |
+
"""Plot ROC curve given the true and predicted values.
|
| 266 |
+
|
| 267 |
+
Read more in the :ref:`User Guide <visualizations>`.
|
| 268 |
+
|
| 269 |
+
.. versionadded:: 1.0
|
| 270 |
+
|
| 271 |
+
Parameters
|
| 272 |
+
----------
|
| 273 |
+
y_true : array-like of shape (n_samples,)
|
| 274 |
+
True labels.
|
| 275 |
+
|
| 276 |
+
y_pred : array-like of shape (n_samples,)
|
| 277 |
+
Target scores, can either be probability estimates of the positive
|
| 278 |
+
class, confidence values, or non-thresholded measure of decisions
|
| 279 |
+
(as returned by “decision_function” on some classifiers).
|
| 280 |
+
|
| 281 |
+
sample_weight : array-like of shape (n_samples,), default=None
|
| 282 |
+
Sample weights.
|
| 283 |
+
|
| 284 |
+
drop_intermediate : bool, default=True
|
| 285 |
+
Whether to drop some suboptimal thresholds which would not appear
|
| 286 |
+
on a plotted ROC curve. This is useful in order to create lighter
|
| 287 |
+
ROC curves.
|
| 288 |
+
|
| 289 |
+
pos_label : str or int, default=None
|
| 290 |
+
The label of the positive class. When `pos_label=None`, if `y_true`
|
| 291 |
+
is in {-1, 1} or {0, 1}, `pos_label` is set to 1, otherwise an
|
| 292 |
+
error will be raised.
|
| 293 |
+
|
| 294 |
+
name : str, default=None
|
| 295 |
+
Name of ROC curve for labeling. If `None`, name will be set to
|
| 296 |
+
`"Classifier"`.
|
| 297 |
+
|
| 298 |
+
ax : matplotlib axes, default=None
|
| 299 |
+
Axes object to plot on. If `None`, a new figure and axes is
|
| 300 |
+
created.
|
| 301 |
+
|
| 302 |
+
**kwargs : dict
|
| 303 |
+
Additional keywords arguments passed to matplotlib `plot` function.
|
| 304 |
+
|
| 305 |
+
Returns
|
| 306 |
+
-------
|
| 307 |
+
display : :class:`~sklearn.metrics.RocCurveDisplay`
|
| 308 |
+
Object that stores computed values.
|
| 309 |
+
|
| 310 |
+
See Also
|
| 311 |
+
--------
|
| 312 |
+
roc_curve : Compute Receiver operating characteristic (ROC) curve.
|
| 313 |
+
RocCurveDisplay.from_estimator : ROC Curve visualization given an
|
| 314 |
+
estimator and some data.
|
| 315 |
+
roc_auc_score : Compute the area under the ROC curve.
|
| 316 |
+
|
| 317 |
+
Examples
|
| 318 |
+
--------
|
| 319 |
+
>>> import matplotlib.pyplot as plt
|
| 320 |
+
>>> from sklearn.datasets import make_classification
|
| 321 |
+
>>> from sklearn.metrics import RocCurveDisplay
|
| 322 |
+
>>> from sklearn.model_selection import train_test_split
|
| 323 |
+
>>> from sklearn.svm import SVC
|
| 324 |
+
>>> X, y = make_classification(random_state=0)
|
| 325 |
+
>>> X_train, X_test, y_train, y_test = train_test_split(
|
| 326 |
+
... X, y, random_state=0)
|
| 327 |
+
>>> clf = SVC(random_state=0).fit(X_train, y_train)
|
| 328 |
+
>>> y_pred = clf.decision_function(X_test)
|
| 329 |
+
>>> RocCurveDisplay.from_predictions(
|
| 330 |
+
... y_test, y_pred)
|
| 331 |
+
<...>
|
| 332 |
+
>>> plt.show()
|
| 333 |
+
"""
|
| 334 |
+
check_matplotlib_support(f"{cls.__name__}.from_predictions")
|
| 335 |
+
|
| 336 |
+
fpr, tpr, _ = roc_curve(
|
| 337 |
+
y_true,
|
| 338 |
+
y_pred,
|
| 339 |
+
pos_label=pos_label,
|
| 340 |
+
sample_weight=sample_weight,
|
| 341 |
+
drop_intermediate=drop_intermediate,
|
| 342 |
+
)
|
| 343 |
+
roc_auc = auc(fpr, tpr)
|
| 344 |
+
|
| 345 |
+
name = "Classifier" if name is None else name
|
| 346 |
+
pos_label = _check_pos_label_consistency(pos_label, y_true)
|
| 347 |
+
|
| 348 |
+
viz = RocCurveDisplay(
|
| 349 |
+
fpr=fpr, tpr=tpr, roc_auc=roc_auc, estimator_name=name, pos_label=pos_label
|
| 350 |
+
)
|
| 351 |
+
|
| 352 |
+
return viz.plot(ax=ax, name=name, **kwargs)
|
videollama2/lib/python3.10/site-packages/sklearn/metrics/_plot/tests/__init__.py
ADDED
|
File without changes
|
videollama2/lib/python3.10/site-packages/sklearn/metrics/_plot/tests/test_common_curve_display.py
ADDED
|
@@ -0,0 +1,152 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import pytest
|
| 2 |
+
|
| 3 |
+
from sklearn.base import ClassifierMixin, clone
|
| 4 |
+
from sklearn.compose import make_column_transformer
|
| 5 |
+
from sklearn.datasets import load_iris
|
| 6 |
+
from sklearn.exceptions import NotFittedError
|
| 7 |
+
from sklearn.linear_model import LogisticRegression
|
| 8 |
+
from sklearn.pipeline import make_pipeline
|
| 9 |
+
from sklearn.preprocessing import StandardScaler
|
| 10 |
+
from sklearn.tree import DecisionTreeClassifier
|
| 11 |
+
|
| 12 |
+
from sklearn.metrics import (
|
| 13 |
+
DetCurveDisplay,
|
| 14 |
+
PrecisionRecallDisplay,
|
| 15 |
+
RocCurveDisplay,
|
| 16 |
+
)
|
| 17 |
+
|
| 18 |
+
|
| 19 |
+
@pytest.fixture(scope="module")
|
| 20 |
+
def data():
|
| 21 |
+
return load_iris(return_X_y=True)
|
| 22 |
+
|
| 23 |
+
|
| 24 |
+
@pytest.fixture(scope="module")
|
| 25 |
+
def data_binary(data):
|
| 26 |
+
X, y = data
|
| 27 |
+
return X[y < 2], y[y < 2]
|
| 28 |
+
|
| 29 |
+
|
| 30 |
+
@pytest.mark.parametrize(
|
| 31 |
+
"Display", [DetCurveDisplay, PrecisionRecallDisplay, RocCurveDisplay]
|
| 32 |
+
)
|
| 33 |
+
def test_display_curve_error_non_binary(pyplot, data, Display):
|
| 34 |
+
"""Check that a proper error is raised when only binary classification is
|
| 35 |
+
supported."""
|
| 36 |
+
X, y = data
|
| 37 |
+
clf = DecisionTreeClassifier().fit(X, y)
|
| 38 |
+
|
| 39 |
+
msg = (
|
| 40 |
+
"Expected 'estimator' to be a binary classifier, but got DecisionTreeClassifier"
|
| 41 |
+
)
|
| 42 |
+
with pytest.raises(ValueError, match=msg):
|
| 43 |
+
Display.from_estimator(clf, X, y)
|
| 44 |
+
|
| 45 |
+
|
| 46 |
+
@pytest.mark.parametrize(
|
| 47 |
+
"response_method, msg",
|
| 48 |
+
[
|
| 49 |
+
(
|
| 50 |
+
"predict_proba",
|
| 51 |
+
"response method predict_proba is not defined in MyClassifier",
|
| 52 |
+
),
|
| 53 |
+
(
|
| 54 |
+
"decision_function",
|
| 55 |
+
"response method decision_function is not defined in MyClassifier",
|
| 56 |
+
),
|
| 57 |
+
(
|
| 58 |
+
"auto",
|
| 59 |
+
"response method decision_function or predict_proba is not "
|
| 60 |
+
"defined in MyClassifier",
|
| 61 |
+
),
|
| 62 |
+
(
|
| 63 |
+
"bad_method",
|
| 64 |
+
"response_method must be 'predict_proba', 'decision_function' or 'auto'",
|
| 65 |
+
),
|
| 66 |
+
],
|
| 67 |
+
)
|
| 68 |
+
@pytest.mark.parametrize(
|
| 69 |
+
"Display", [DetCurveDisplay, PrecisionRecallDisplay, RocCurveDisplay]
|
| 70 |
+
)
|
| 71 |
+
def test_display_curve_error_no_response(
|
| 72 |
+
pyplot,
|
| 73 |
+
data_binary,
|
| 74 |
+
response_method,
|
| 75 |
+
msg,
|
| 76 |
+
Display,
|
| 77 |
+
):
|
| 78 |
+
"""Check that a proper error is raised when the response method requested
|
| 79 |
+
is not defined for the given trained classifier."""
|
| 80 |
+
X, y = data_binary
|
| 81 |
+
|
| 82 |
+
class MyClassifier(ClassifierMixin):
|
| 83 |
+
def fit(self, X, y):
|
| 84 |
+
self.classes_ = [0, 1]
|
| 85 |
+
return self
|
| 86 |
+
|
| 87 |
+
clf = MyClassifier().fit(X, y)
|
| 88 |
+
|
| 89 |
+
with pytest.raises(ValueError, match=msg):
|
| 90 |
+
Display.from_estimator(clf, X, y, response_method=response_method)
|
| 91 |
+
|
| 92 |
+
|
| 93 |
+
@pytest.mark.parametrize(
|
| 94 |
+
"Display", [DetCurveDisplay, PrecisionRecallDisplay, RocCurveDisplay]
|
| 95 |
+
)
|
| 96 |
+
@pytest.mark.parametrize("constructor_name", ["from_estimator", "from_predictions"])
|
| 97 |
+
def test_display_curve_estimator_name_multiple_calls(
|
| 98 |
+
pyplot,
|
| 99 |
+
data_binary,
|
| 100 |
+
Display,
|
| 101 |
+
constructor_name,
|
| 102 |
+
):
|
| 103 |
+
"""Check that passing `name` when calling `plot` will overwrite the original name
|
| 104 |
+
in the legend."""
|
| 105 |
+
X, y = data_binary
|
| 106 |
+
clf_name = "my hand-crafted name"
|
| 107 |
+
clf = LogisticRegression().fit(X, y)
|
| 108 |
+
y_pred = clf.predict_proba(X)[:, 1]
|
| 109 |
+
|
| 110 |
+
# safe guard for the binary if/else construction
|
| 111 |
+
assert constructor_name in ("from_estimator", "from_predictions")
|
| 112 |
+
|
| 113 |
+
if constructor_name == "from_estimator":
|
| 114 |
+
disp = Display.from_estimator(clf, X, y, name=clf_name)
|
| 115 |
+
else:
|
| 116 |
+
disp = Display.from_predictions(y, y_pred, name=clf_name)
|
| 117 |
+
assert disp.estimator_name == clf_name
|
| 118 |
+
pyplot.close("all")
|
| 119 |
+
disp.plot()
|
| 120 |
+
assert clf_name in disp.line_.get_label()
|
| 121 |
+
pyplot.close("all")
|
| 122 |
+
clf_name = "another_name"
|
| 123 |
+
disp.plot(name=clf_name)
|
| 124 |
+
assert clf_name in disp.line_.get_label()
|
| 125 |
+
|
| 126 |
+
|
| 127 |
+
@pytest.mark.parametrize(
|
| 128 |
+
"clf",
|
| 129 |
+
[
|
| 130 |
+
LogisticRegression(),
|
| 131 |
+
make_pipeline(StandardScaler(), LogisticRegression()),
|
| 132 |
+
make_pipeline(
|
| 133 |
+
make_column_transformer((StandardScaler(), [0, 1])), LogisticRegression()
|
| 134 |
+
),
|
| 135 |
+
],
|
| 136 |
+
)
|
| 137 |
+
@pytest.mark.parametrize(
|
| 138 |
+
"Display", [DetCurveDisplay, PrecisionRecallDisplay, RocCurveDisplay]
|
| 139 |
+
)
|
| 140 |
+
def test_display_curve_not_fitted_errors(pyplot, data_binary, clf, Display):
|
| 141 |
+
"""Check that a proper error is raised when the classifier is not
|
| 142 |
+
fitted."""
|
| 143 |
+
X, y = data_binary
|
| 144 |
+
# clone since we parametrize the test and the classifier will be fitted
|
| 145 |
+
# when testing the second and subsequent plotting function
|
| 146 |
+
model = clone(clf)
|
| 147 |
+
with pytest.raises(NotFittedError):
|
| 148 |
+
Display.from_estimator(model, X, y)
|
| 149 |
+
model.fit(X, y)
|
| 150 |
+
disp = Display.from_estimator(model, X, y)
|
| 151 |
+
assert model.__class__.__name__ in disp.line_.get_label()
|
| 152 |
+
assert disp.estimator_name == model.__class__.__name__
|
videollama2/lib/python3.10/site-packages/sklearn/metrics/_plot/tests/test_confusion_matrix_display.py
ADDED
|
@@ -0,0 +1,407 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from numpy.testing import (
|
| 2 |
+
assert_allclose,
|
| 3 |
+
assert_array_equal,
|
| 4 |
+
)
|
| 5 |
+
import numpy as np
|
| 6 |
+
import pytest
|
| 7 |
+
|
| 8 |
+
from sklearn.datasets import make_classification
|
| 9 |
+
from sklearn.compose import make_column_transformer
|
| 10 |
+
from sklearn.exceptions import NotFittedError
|
| 11 |
+
from sklearn.linear_model import LogisticRegression
|
| 12 |
+
from sklearn.pipeline import make_pipeline
|
| 13 |
+
from sklearn.preprocessing import StandardScaler
|
| 14 |
+
from sklearn.svm import SVC, SVR
|
| 15 |
+
|
| 16 |
+
from sklearn.metrics import ConfusionMatrixDisplay
|
| 17 |
+
from sklearn.metrics import confusion_matrix
|
| 18 |
+
|
| 19 |
+
|
| 20 |
+
# TODO: Remove when https://github.com/numpy/numpy/issues/14397 is resolved
|
| 21 |
+
pytestmark = pytest.mark.filterwarnings(
|
| 22 |
+
"ignore:In future, it will be an error for 'np.bool_':DeprecationWarning:"
|
| 23 |
+
"matplotlib.*"
|
| 24 |
+
)
|
| 25 |
+
|
| 26 |
+
|
| 27 |
+
def test_confusion_matrix_display_validation(pyplot):
|
| 28 |
+
"""Check that we raise the proper error when validating parameters."""
|
| 29 |
+
X, y = make_classification(
|
| 30 |
+
n_samples=100, n_informative=5, n_classes=5, random_state=0
|
| 31 |
+
)
|
| 32 |
+
|
| 33 |
+
with pytest.raises(NotFittedError):
|
| 34 |
+
ConfusionMatrixDisplay.from_estimator(SVC(), X, y)
|
| 35 |
+
|
| 36 |
+
regressor = SVR().fit(X, y)
|
| 37 |
+
y_pred_regressor = regressor.predict(X)
|
| 38 |
+
y_pred_classifier = SVC().fit(X, y).predict(X)
|
| 39 |
+
|
| 40 |
+
err_msg = "ConfusionMatrixDisplay.from_estimator only supports classifiers"
|
| 41 |
+
with pytest.raises(ValueError, match=err_msg):
|
| 42 |
+
ConfusionMatrixDisplay.from_estimator(regressor, X, y)
|
| 43 |
+
|
| 44 |
+
err_msg = "Mix type of y not allowed, got types"
|
| 45 |
+
with pytest.raises(ValueError, match=err_msg):
|
| 46 |
+
# Force `y_true` to be seen as a regression problem
|
| 47 |
+
ConfusionMatrixDisplay.from_predictions(y + 0.5, y_pred_classifier)
|
| 48 |
+
with pytest.raises(ValueError, match=err_msg):
|
| 49 |
+
ConfusionMatrixDisplay.from_predictions(y, y_pred_regressor)
|
| 50 |
+
|
| 51 |
+
err_msg = "Found input variables with inconsistent numbers of samples"
|
| 52 |
+
with pytest.raises(ValueError, match=err_msg):
|
| 53 |
+
ConfusionMatrixDisplay.from_predictions(y, y_pred_classifier[::2])
|
| 54 |
+
|
| 55 |
+
|
| 56 |
+
@pytest.mark.parametrize("constructor_name", ["from_estimator", "from_predictions"])
|
| 57 |
+
def test_confusion_matrix_display_invalid_option(pyplot, constructor_name):
|
| 58 |
+
"""Check the error raise if an invalid parameter value is passed."""
|
| 59 |
+
X, y = make_classification(
|
| 60 |
+
n_samples=100, n_informative=5, n_classes=5, random_state=0
|
| 61 |
+
)
|
| 62 |
+
classifier = SVC().fit(X, y)
|
| 63 |
+
y_pred = classifier.predict(X)
|
| 64 |
+
|
| 65 |
+
# safe guard for the binary if/else construction
|
| 66 |
+
assert constructor_name in ("from_estimator", "from_predictions")
|
| 67 |
+
extra_params = {"normalize": "invalid"}
|
| 68 |
+
|
| 69 |
+
err_msg = r"normalize must be one of \{'true', 'pred', 'all', None\}"
|
| 70 |
+
with pytest.raises(ValueError, match=err_msg):
|
| 71 |
+
if constructor_name == "from_estimator":
|
| 72 |
+
ConfusionMatrixDisplay.from_estimator(classifier, X, y, **extra_params)
|
| 73 |
+
else:
|
| 74 |
+
ConfusionMatrixDisplay.from_predictions(y, y_pred, **extra_params)
|
| 75 |
+
|
| 76 |
+
|
| 77 |
+
@pytest.mark.parametrize("constructor_name", ["from_estimator", "from_predictions"])
|
| 78 |
+
@pytest.mark.parametrize("with_labels", [True, False])
|
| 79 |
+
@pytest.mark.parametrize("with_display_labels", [True, False])
|
| 80 |
+
def test_confusion_matrix_display_custom_labels(
|
| 81 |
+
pyplot, constructor_name, with_labels, with_display_labels
|
| 82 |
+
):
|
| 83 |
+
"""Check the resulting plot when labels are given."""
|
| 84 |
+
n_classes = 5
|
| 85 |
+
X, y = make_classification(
|
| 86 |
+
n_samples=100, n_informative=5, n_classes=n_classes, random_state=0
|
| 87 |
+
)
|
| 88 |
+
classifier = SVC().fit(X, y)
|
| 89 |
+
y_pred = classifier.predict(X)
|
| 90 |
+
|
| 91 |
+
# safe guard for the binary if/else construction
|
| 92 |
+
assert constructor_name in ("from_estimator", "from_predictions")
|
| 93 |
+
|
| 94 |
+
ax = pyplot.gca()
|
| 95 |
+
labels = [2, 1, 0, 3, 4] if with_labels else None
|
| 96 |
+
display_labels = ["b", "d", "a", "e", "f"] if with_display_labels else None
|
| 97 |
+
|
| 98 |
+
cm = confusion_matrix(y, y_pred, labels=labels)
|
| 99 |
+
common_kwargs = {
|
| 100 |
+
"ax": ax,
|
| 101 |
+
"display_labels": display_labels,
|
| 102 |
+
"labels": labels,
|
| 103 |
+
}
|
| 104 |
+
if constructor_name == "from_estimator":
|
| 105 |
+
disp = ConfusionMatrixDisplay.from_estimator(classifier, X, y, **common_kwargs)
|
| 106 |
+
else:
|
| 107 |
+
disp = ConfusionMatrixDisplay.from_predictions(y, y_pred, **common_kwargs)
|
| 108 |
+
assert_allclose(disp.confusion_matrix, cm)
|
| 109 |
+
|
| 110 |
+
if with_display_labels:
|
| 111 |
+
expected_display_labels = display_labels
|
| 112 |
+
elif with_labels:
|
| 113 |
+
expected_display_labels = labels
|
| 114 |
+
else:
|
| 115 |
+
expected_display_labels = list(range(n_classes))
|
| 116 |
+
|
| 117 |
+
expected_display_labels_str = [str(name) for name in expected_display_labels]
|
| 118 |
+
|
| 119 |
+
x_ticks = [tick.get_text() for tick in disp.ax_.get_xticklabels()]
|
| 120 |
+
y_ticks = [tick.get_text() for tick in disp.ax_.get_yticklabels()]
|
| 121 |
+
|
| 122 |
+
assert_array_equal(disp.display_labels, expected_display_labels)
|
| 123 |
+
assert_array_equal(x_ticks, expected_display_labels_str)
|
| 124 |
+
assert_array_equal(y_ticks, expected_display_labels_str)
|
| 125 |
+
|
| 126 |
+
|
| 127 |
+
@pytest.mark.parametrize("constructor_name", ["from_estimator", "from_predictions"])
|
| 128 |
+
@pytest.mark.parametrize("normalize", ["true", "pred", "all", None])
|
| 129 |
+
@pytest.mark.parametrize("include_values", [True, False])
|
| 130 |
+
def test_confusion_matrix_display_plotting(
|
| 131 |
+
pyplot,
|
| 132 |
+
constructor_name,
|
| 133 |
+
normalize,
|
| 134 |
+
include_values,
|
| 135 |
+
):
|
| 136 |
+
"""Check the overall plotting rendering."""
|
| 137 |
+
n_classes = 5
|
| 138 |
+
X, y = make_classification(
|
| 139 |
+
n_samples=100, n_informative=5, n_classes=n_classes, random_state=0
|
| 140 |
+
)
|
| 141 |
+
classifier = SVC().fit(X, y)
|
| 142 |
+
y_pred = classifier.predict(X)
|
| 143 |
+
|
| 144 |
+
# safe guard for the binary if/else construction
|
| 145 |
+
assert constructor_name in ("from_estimator", "from_predictions")
|
| 146 |
+
|
| 147 |
+
ax = pyplot.gca()
|
| 148 |
+
cmap = "plasma"
|
| 149 |
+
|
| 150 |
+
cm = confusion_matrix(y, y_pred)
|
| 151 |
+
common_kwargs = {
|
| 152 |
+
"normalize": normalize,
|
| 153 |
+
"cmap": cmap,
|
| 154 |
+
"ax": ax,
|
| 155 |
+
"include_values": include_values,
|
| 156 |
+
}
|
| 157 |
+
if constructor_name == "from_estimator":
|
| 158 |
+
disp = ConfusionMatrixDisplay.from_estimator(classifier, X, y, **common_kwargs)
|
| 159 |
+
else:
|
| 160 |
+
disp = ConfusionMatrixDisplay.from_predictions(y, y_pred, **common_kwargs)
|
| 161 |
+
|
| 162 |
+
assert disp.ax_ == ax
|
| 163 |
+
|
| 164 |
+
if normalize == "true":
|
| 165 |
+
cm = cm / cm.sum(axis=1, keepdims=True)
|
| 166 |
+
elif normalize == "pred":
|
| 167 |
+
cm = cm / cm.sum(axis=0, keepdims=True)
|
| 168 |
+
elif normalize == "all":
|
| 169 |
+
cm = cm / cm.sum()
|
| 170 |
+
|
| 171 |
+
assert_allclose(disp.confusion_matrix, cm)
|
| 172 |
+
import matplotlib as mpl
|
| 173 |
+
|
| 174 |
+
assert isinstance(disp.im_, mpl.image.AxesImage)
|
| 175 |
+
assert disp.im_.get_cmap().name == cmap
|
| 176 |
+
assert isinstance(disp.ax_, pyplot.Axes)
|
| 177 |
+
assert isinstance(disp.figure_, pyplot.Figure)
|
| 178 |
+
|
| 179 |
+
assert disp.ax_.get_ylabel() == "True label"
|
| 180 |
+
assert disp.ax_.get_xlabel() == "Predicted label"
|
| 181 |
+
|
| 182 |
+
x_ticks = [tick.get_text() for tick in disp.ax_.get_xticklabels()]
|
| 183 |
+
y_ticks = [tick.get_text() for tick in disp.ax_.get_yticklabels()]
|
| 184 |
+
|
| 185 |
+
expected_display_labels = list(range(n_classes))
|
| 186 |
+
|
| 187 |
+
expected_display_labels_str = [str(name) for name in expected_display_labels]
|
| 188 |
+
|
| 189 |
+
assert_array_equal(disp.display_labels, expected_display_labels)
|
| 190 |
+
assert_array_equal(x_ticks, expected_display_labels_str)
|
| 191 |
+
assert_array_equal(y_ticks, expected_display_labels_str)
|
| 192 |
+
|
| 193 |
+
image_data = disp.im_.get_array().data
|
| 194 |
+
assert_allclose(image_data, cm)
|
| 195 |
+
|
| 196 |
+
if include_values:
|
| 197 |
+
assert disp.text_.shape == (n_classes, n_classes)
|
| 198 |
+
fmt = ".2g"
|
| 199 |
+
expected_text = np.array([format(v, fmt) for v in cm.ravel(order="C")])
|
| 200 |
+
text_text = np.array([t.get_text() for t in disp.text_.ravel(order="C")])
|
| 201 |
+
assert_array_equal(expected_text, text_text)
|
| 202 |
+
else:
|
| 203 |
+
assert disp.text_ is None
|
| 204 |
+
|
| 205 |
+
|
| 206 |
+
@pytest.mark.parametrize("constructor_name", ["from_estimator", "from_predictions"])
|
| 207 |
+
def test_confusion_matrix_display(pyplot, constructor_name):
|
| 208 |
+
"""Check the behaviour of the default constructor without using the class
|
| 209 |
+
methods."""
|
| 210 |
+
n_classes = 5
|
| 211 |
+
X, y = make_classification(
|
| 212 |
+
n_samples=100, n_informative=5, n_classes=n_classes, random_state=0
|
| 213 |
+
)
|
| 214 |
+
classifier = SVC().fit(X, y)
|
| 215 |
+
y_pred = classifier.predict(X)
|
| 216 |
+
|
| 217 |
+
# safe guard for the binary if/else construction
|
| 218 |
+
assert constructor_name in ("from_estimator", "from_predictions")
|
| 219 |
+
|
| 220 |
+
cm = confusion_matrix(y, y_pred)
|
| 221 |
+
common_kwargs = {
|
| 222 |
+
"normalize": None,
|
| 223 |
+
"include_values": True,
|
| 224 |
+
"cmap": "viridis",
|
| 225 |
+
"xticks_rotation": 45.0,
|
| 226 |
+
}
|
| 227 |
+
if constructor_name == "from_estimator":
|
| 228 |
+
disp = ConfusionMatrixDisplay.from_estimator(classifier, X, y, **common_kwargs)
|
| 229 |
+
else:
|
| 230 |
+
disp = ConfusionMatrixDisplay.from_predictions(y, y_pred, **common_kwargs)
|
| 231 |
+
|
| 232 |
+
assert_allclose(disp.confusion_matrix, cm)
|
| 233 |
+
assert disp.text_.shape == (n_classes, n_classes)
|
| 234 |
+
|
| 235 |
+
rotations = [tick.get_rotation() for tick in disp.ax_.get_xticklabels()]
|
| 236 |
+
assert_allclose(rotations, 45.0)
|
| 237 |
+
|
| 238 |
+
image_data = disp.im_.get_array().data
|
| 239 |
+
assert_allclose(image_data, cm)
|
| 240 |
+
|
| 241 |
+
disp.plot(cmap="plasma")
|
| 242 |
+
assert disp.im_.get_cmap().name == "plasma"
|
| 243 |
+
|
| 244 |
+
disp.plot(include_values=False)
|
| 245 |
+
assert disp.text_ is None
|
| 246 |
+
|
| 247 |
+
disp.plot(xticks_rotation=90.0)
|
| 248 |
+
rotations = [tick.get_rotation() for tick in disp.ax_.get_xticklabels()]
|
| 249 |
+
assert_allclose(rotations, 90.0)
|
| 250 |
+
|
| 251 |
+
disp.plot(values_format="e")
|
| 252 |
+
expected_text = np.array([format(v, "e") for v in cm.ravel(order="C")])
|
| 253 |
+
text_text = np.array([t.get_text() for t in disp.text_.ravel(order="C")])
|
| 254 |
+
assert_array_equal(expected_text, text_text)
|
| 255 |
+
|
| 256 |
+
|
| 257 |
+
def test_confusion_matrix_contrast(pyplot):
|
| 258 |
+
"""Check that the text color is appropriate depending on background."""
|
| 259 |
+
|
| 260 |
+
cm = np.eye(2) / 2
|
| 261 |
+
disp = ConfusionMatrixDisplay(cm, display_labels=[0, 1])
|
| 262 |
+
|
| 263 |
+
disp.plot(cmap=pyplot.cm.gray)
|
| 264 |
+
# diagonal text is black
|
| 265 |
+
assert_allclose(disp.text_[0, 0].get_color(), [0.0, 0.0, 0.0, 1.0])
|
| 266 |
+
assert_allclose(disp.text_[1, 1].get_color(), [0.0, 0.0, 0.0, 1.0])
|
| 267 |
+
|
| 268 |
+
# off-diagonal text is white
|
| 269 |
+
assert_allclose(disp.text_[0, 1].get_color(), [1.0, 1.0, 1.0, 1.0])
|
| 270 |
+
assert_allclose(disp.text_[1, 0].get_color(), [1.0, 1.0, 1.0, 1.0])
|
| 271 |
+
|
| 272 |
+
disp.plot(cmap=pyplot.cm.gray_r)
|
| 273 |
+
# diagonal text is white
|
| 274 |
+
assert_allclose(disp.text_[0, 1].get_color(), [0.0, 0.0, 0.0, 1.0])
|
| 275 |
+
assert_allclose(disp.text_[1, 0].get_color(), [0.0, 0.0, 0.0, 1.0])
|
| 276 |
+
|
| 277 |
+
# off-diagonal text is black
|
| 278 |
+
assert_allclose(disp.text_[0, 0].get_color(), [1.0, 1.0, 1.0, 1.0])
|
| 279 |
+
assert_allclose(disp.text_[1, 1].get_color(), [1.0, 1.0, 1.0, 1.0])
|
| 280 |
+
|
| 281 |
+
# Regression test for #15920
|
| 282 |
+
cm = np.array([[19, 34], [32, 58]])
|
| 283 |
+
disp = ConfusionMatrixDisplay(cm, display_labels=[0, 1])
|
| 284 |
+
|
| 285 |
+
disp.plot(cmap=pyplot.cm.Blues)
|
| 286 |
+
min_color = pyplot.cm.Blues(0)
|
| 287 |
+
max_color = pyplot.cm.Blues(255)
|
| 288 |
+
assert_allclose(disp.text_[0, 0].get_color(), max_color)
|
| 289 |
+
assert_allclose(disp.text_[0, 1].get_color(), max_color)
|
| 290 |
+
assert_allclose(disp.text_[1, 0].get_color(), max_color)
|
| 291 |
+
assert_allclose(disp.text_[1, 1].get_color(), min_color)
|
| 292 |
+
|
| 293 |
+
|
| 294 |
+
@pytest.mark.parametrize(
|
| 295 |
+
"clf",
|
| 296 |
+
[
|
| 297 |
+
LogisticRegression(),
|
| 298 |
+
make_pipeline(StandardScaler(), LogisticRegression()),
|
| 299 |
+
make_pipeline(
|
| 300 |
+
make_column_transformer((StandardScaler(), [0, 1])),
|
| 301 |
+
LogisticRegression(),
|
| 302 |
+
),
|
| 303 |
+
],
|
| 304 |
+
ids=["clf", "pipeline-clf", "pipeline-column_transformer-clf"],
|
| 305 |
+
)
|
| 306 |
+
def test_confusion_matrix_pipeline(pyplot, clf):
|
| 307 |
+
"""Check the behaviour of the plotting with more complex pipeline."""
|
| 308 |
+
n_classes = 5
|
| 309 |
+
X, y = make_classification(
|
| 310 |
+
n_samples=100, n_informative=5, n_classes=n_classes, random_state=0
|
| 311 |
+
)
|
| 312 |
+
with pytest.raises(NotFittedError):
|
| 313 |
+
ConfusionMatrixDisplay.from_estimator(clf, X, y)
|
| 314 |
+
clf.fit(X, y)
|
| 315 |
+
y_pred = clf.predict(X)
|
| 316 |
+
|
| 317 |
+
disp = ConfusionMatrixDisplay.from_estimator(clf, X, y)
|
| 318 |
+
cm = confusion_matrix(y, y_pred)
|
| 319 |
+
|
| 320 |
+
assert_allclose(disp.confusion_matrix, cm)
|
| 321 |
+
assert disp.text_.shape == (n_classes, n_classes)
|
| 322 |
+
|
| 323 |
+
|
| 324 |
+
@pytest.mark.parametrize("constructor_name", ["from_estimator", "from_predictions"])
|
| 325 |
+
def test_confusion_matrix_with_unknown_labels(pyplot, constructor_name):
|
| 326 |
+
"""Check that when labels=None, the unique values in `y_pred` and `y_true`
|
| 327 |
+
will be used.
|
| 328 |
+
Non-regression test for:
|
| 329 |
+
https://github.com/scikit-learn/scikit-learn/pull/18405
|
| 330 |
+
"""
|
| 331 |
+
n_classes = 5
|
| 332 |
+
X, y = make_classification(
|
| 333 |
+
n_samples=100, n_informative=5, n_classes=n_classes, random_state=0
|
| 334 |
+
)
|
| 335 |
+
classifier = SVC().fit(X, y)
|
| 336 |
+
y_pred = classifier.predict(X)
|
| 337 |
+
# create unseen labels in `y_true` not seen during fitting and not present
|
| 338 |
+
# in 'classifier.classes_'
|
| 339 |
+
y = y + 1
|
| 340 |
+
|
| 341 |
+
# safe guard for the binary if/else construction
|
| 342 |
+
assert constructor_name in ("from_estimator", "from_predictions")
|
| 343 |
+
|
| 344 |
+
common_kwargs = {"labels": None}
|
| 345 |
+
if constructor_name == "from_estimator":
|
| 346 |
+
disp = ConfusionMatrixDisplay.from_estimator(classifier, X, y, **common_kwargs)
|
| 347 |
+
else:
|
| 348 |
+
disp = ConfusionMatrixDisplay.from_predictions(y, y_pred, **common_kwargs)
|
| 349 |
+
|
| 350 |
+
display_labels = [tick.get_text() for tick in disp.ax_.get_xticklabels()]
|
| 351 |
+
expected_labels = [str(i) for i in range(n_classes + 1)]
|
| 352 |
+
assert_array_equal(expected_labels, display_labels)
|
| 353 |
+
|
| 354 |
+
|
| 355 |
+
def test_colormap_max(pyplot):
|
| 356 |
+
"""Check that the max color is used for the color of the text."""
|
| 357 |
+
|
| 358 |
+
from matplotlib import cm
|
| 359 |
+
|
| 360 |
+
gray = cm.get_cmap("gray", 1024)
|
| 361 |
+
confusion_matrix = np.array([[1.0, 0.0], [0.0, 1.0]])
|
| 362 |
+
|
| 363 |
+
disp = ConfusionMatrixDisplay(confusion_matrix)
|
| 364 |
+
disp.plot(cmap=gray)
|
| 365 |
+
|
| 366 |
+
color = disp.text_[1, 0].get_color()
|
| 367 |
+
assert_allclose(color, [1.0, 1.0, 1.0, 1.0])
|
| 368 |
+
|
| 369 |
+
|
| 370 |
+
def test_im_kw_adjust_vmin_vmax(pyplot):
|
| 371 |
+
"""Check that im_kw passes kwargs to imshow"""
|
| 372 |
+
|
| 373 |
+
confusion_matrix = np.array([[0.48, 0.04], [0.08, 0.4]])
|
| 374 |
+
disp = ConfusionMatrixDisplay(confusion_matrix)
|
| 375 |
+
disp.plot(im_kw=dict(vmin=0.0, vmax=0.8))
|
| 376 |
+
|
| 377 |
+
clim = disp.im_.get_clim()
|
| 378 |
+
assert clim[0] == pytest.approx(0.0)
|
| 379 |
+
assert clim[1] == pytest.approx(0.8)
|
| 380 |
+
|
| 381 |
+
|
| 382 |
+
def test_confusion_matrix_text_kw(pyplot):
|
| 383 |
+
"""Check that text_kw is passed to the text call."""
|
| 384 |
+
font_size = 15.0
|
| 385 |
+
X, y = make_classification(random_state=0)
|
| 386 |
+
classifier = SVC().fit(X, y)
|
| 387 |
+
|
| 388 |
+
# from_estimator passes the font size
|
| 389 |
+
disp = ConfusionMatrixDisplay.from_estimator(
|
| 390 |
+
classifier, X, y, text_kw={"fontsize": font_size}
|
| 391 |
+
)
|
| 392 |
+
for text in disp.text_.reshape(-1):
|
| 393 |
+
assert text.get_fontsize() == font_size
|
| 394 |
+
|
| 395 |
+
# plot adjusts plot to new font size
|
| 396 |
+
new_font_size = 20.0
|
| 397 |
+
disp.plot(text_kw={"fontsize": new_font_size})
|
| 398 |
+
for text in disp.text_.reshape(-1):
|
| 399 |
+
assert text.get_fontsize() == new_font_size
|
| 400 |
+
|
| 401 |
+
# from_predictions passes the font size
|
| 402 |
+
y_pred = classifier.predict(X)
|
| 403 |
+
disp = ConfusionMatrixDisplay.from_predictions(
|
| 404 |
+
y, y_pred, text_kw={"fontsize": font_size}
|
| 405 |
+
)
|
| 406 |
+
for text in disp.text_.reshape(-1):
|
| 407 |
+
assert text.get_fontsize() == font_size
|
videollama2/lib/python3.10/site-packages/sklearn/metrics/_plot/tests/test_precision_recall_display.py
ADDED
|
@@ -0,0 +1,292 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import numpy as np
|
| 2 |
+
import pytest
|
| 3 |
+
|
| 4 |
+
from sklearn.compose import make_column_transformer
|
| 5 |
+
from sklearn.datasets import load_breast_cancer, make_classification
|
| 6 |
+
from sklearn.exceptions import NotFittedError
|
| 7 |
+
from sklearn.linear_model import LogisticRegression
|
| 8 |
+
from sklearn.metrics import average_precision_score, precision_recall_curve
|
| 9 |
+
from sklearn.model_selection import train_test_split
|
| 10 |
+
from sklearn.pipeline import make_pipeline
|
| 11 |
+
from sklearn.preprocessing import StandardScaler
|
| 12 |
+
from sklearn.svm import SVC, SVR
|
| 13 |
+
from sklearn.utils import shuffle
|
| 14 |
+
|
| 15 |
+
from sklearn.metrics import PrecisionRecallDisplay
|
| 16 |
+
|
| 17 |
+
# TODO: Remove when https://github.com/numpy/numpy/issues/14397 is resolved
|
| 18 |
+
pytestmark = pytest.mark.filterwarnings(
|
| 19 |
+
"ignore:In future, it will be an error for 'np.bool_':DeprecationWarning:"
|
| 20 |
+
"matplotlib.*"
|
| 21 |
+
)
|
| 22 |
+
|
| 23 |
+
|
| 24 |
+
def test_precision_recall_display_validation(pyplot):
|
| 25 |
+
"""Check that we raise the proper error when validating parameters."""
|
| 26 |
+
X, y = make_classification(
|
| 27 |
+
n_samples=100, n_informative=5, n_classes=5, random_state=0
|
| 28 |
+
)
|
| 29 |
+
|
| 30 |
+
with pytest.raises(NotFittedError):
|
| 31 |
+
PrecisionRecallDisplay.from_estimator(SVC(), X, y)
|
| 32 |
+
|
| 33 |
+
regressor = SVR().fit(X, y)
|
| 34 |
+
y_pred_regressor = regressor.predict(X)
|
| 35 |
+
classifier = SVC(probability=True).fit(X, y)
|
| 36 |
+
y_pred_classifier = classifier.predict_proba(X)[:, -1]
|
| 37 |
+
|
| 38 |
+
err_msg = "PrecisionRecallDisplay.from_estimator only supports classifiers"
|
| 39 |
+
with pytest.raises(ValueError, match=err_msg):
|
| 40 |
+
PrecisionRecallDisplay.from_estimator(regressor, X, y)
|
| 41 |
+
|
| 42 |
+
err_msg = "Expected 'estimator' to be a binary classifier, but got SVC"
|
| 43 |
+
with pytest.raises(ValueError, match=err_msg):
|
| 44 |
+
PrecisionRecallDisplay.from_estimator(classifier, X, y)
|
| 45 |
+
|
| 46 |
+
err_msg = "{} format is not supported"
|
| 47 |
+
with pytest.raises(ValueError, match=err_msg.format("continuous")):
|
| 48 |
+
# Force `y_true` to be seen as a regression problem
|
| 49 |
+
PrecisionRecallDisplay.from_predictions(y + 0.5, y_pred_classifier, pos_label=1)
|
| 50 |
+
with pytest.raises(ValueError, match=err_msg.format("multiclass")):
|
| 51 |
+
PrecisionRecallDisplay.from_predictions(y, y_pred_regressor, pos_label=1)
|
| 52 |
+
|
| 53 |
+
err_msg = "Found input variables with inconsistent numbers of samples"
|
| 54 |
+
with pytest.raises(ValueError, match=err_msg):
|
| 55 |
+
PrecisionRecallDisplay.from_predictions(y, y_pred_classifier[::2])
|
| 56 |
+
|
| 57 |
+
X, y = make_classification(n_classes=2, n_samples=50, random_state=0)
|
| 58 |
+
y += 10
|
| 59 |
+
classifier.fit(X, y)
|
| 60 |
+
y_pred_classifier = classifier.predict_proba(X)[:, -1]
|
| 61 |
+
err_msg = r"y_true takes value in {10, 11} and pos_label is not specified"
|
| 62 |
+
with pytest.raises(ValueError, match=err_msg):
|
| 63 |
+
PrecisionRecallDisplay.from_predictions(y, y_pred_classifier)
|
| 64 |
+
|
| 65 |
+
|
| 66 |
+
@pytest.mark.parametrize("constructor_name", ["from_estimator", "from_predictions"])
|
| 67 |
+
@pytest.mark.parametrize("response_method", ["predict_proba", "decision_function"])
|
| 68 |
+
def test_precision_recall_display_plotting(pyplot, constructor_name, response_method):
|
| 69 |
+
"""Check the overall plotting rendering."""
|
| 70 |
+
X, y = make_classification(n_classes=2, n_samples=50, random_state=0)
|
| 71 |
+
pos_label = 1
|
| 72 |
+
|
| 73 |
+
classifier = LogisticRegression().fit(X, y)
|
| 74 |
+
classifier.fit(X, y)
|
| 75 |
+
|
| 76 |
+
y_pred = getattr(classifier, response_method)(X)
|
| 77 |
+
y_pred = y_pred if y_pred.ndim == 1 else y_pred[:, pos_label]
|
| 78 |
+
|
| 79 |
+
# safe guard for the binary if/else construction
|
| 80 |
+
assert constructor_name in ("from_estimator", "from_predictions")
|
| 81 |
+
|
| 82 |
+
if constructor_name == "from_estimator":
|
| 83 |
+
display = PrecisionRecallDisplay.from_estimator(
|
| 84 |
+
classifier, X, y, response_method=response_method
|
| 85 |
+
)
|
| 86 |
+
else:
|
| 87 |
+
display = PrecisionRecallDisplay.from_predictions(
|
| 88 |
+
y, y_pred, pos_label=pos_label
|
| 89 |
+
)
|
| 90 |
+
|
| 91 |
+
precision, recall, _ = precision_recall_curve(y, y_pred, pos_label=pos_label)
|
| 92 |
+
average_precision = average_precision_score(y, y_pred, pos_label=pos_label)
|
| 93 |
+
|
| 94 |
+
np.testing.assert_allclose(display.precision, precision)
|
| 95 |
+
np.testing.assert_allclose(display.recall, recall)
|
| 96 |
+
assert display.average_precision == pytest.approx(average_precision)
|
| 97 |
+
|
| 98 |
+
import matplotlib as mpl
|
| 99 |
+
|
| 100 |
+
assert isinstance(display.line_, mpl.lines.Line2D)
|
| 101 |
+
assert isinstance(display.ax_, mpl.axes.Axes)
|
| 102 |
+
assert isinstance(display.figure_, mpl.figure.Figure)
|
| 103 |
+
|
| 104 |
+
assert display.ax_.get_xlabel() == "Recall (Positive label: 1)"
|
| 105 |
+
assert display.ax_.get_ylabel() == "Precision (Positive label: 1)"
|
| 106 |
+
|
| 107 |
+
# plotting passing some new parameters
|
| 108 |
+
display.plot(alpha=0.8, name="MySpecialEstimator")
|
| 109 |
+
expected_label = f"MySpecialEstimator (AP = {average_precision:0.2f})"
|
| 110 |
+
assert display.line_.get_label() == expected_label
|
| 111 |
+
assert display.line_.get_alpha() == pytest.approx(0.8)
|
| 112 |
+
|
| 113 |
+
|
| 114 |
+
@pytest.mark.parametrize(
|
| 115 |
+
"constructor_name, default_label",
|
| 116 |
+
[
|
| 117 |
+
("from_estimator", "LogisticRegression (AP = {:.2f})"),
|
| 118 |
+
("from_predictions", "Classifier (AP = {:.2f})"),
|
| 119 |
+
],
|
| 120 |
+
)
|
| 121 |
+
def test_precision_recall_display_name(pyplot, constructor_name, default_label):
|
| 122 |
+
"""Check the behaviour of the name parameters"""
|
| 123 |
+
X, y = make_classification(n_classes=2, n_samples=100, random_state=0)
|
| 124 |
+
pos_label = 1
|
| 125 |
+
|
| 126 |
+
classifier = LogisticRegression().fit(X, y)
|
| 127 |
+
classifier.fit(X, y)
|
| 128 |
+
|
| 129 |
+
y_pred = classifier.predict_proba(X)[:, pos_label]
|
| 130 |
+
|
| 131 |
+
# safe guard for the binary if/else construction
|
| 132 |
+
assert constructor_name in ("from_estimator", "from_predictions")
|
| 133 |
+
|
| 134 |
+
if constructor_name == "from_estimator":
|
| 135 |
+
display = PrecisionRecallDisplay.from_estimator(classifier, X, y)
|
| 136 |
+
else:
|
| 137 |
+
display = PrecisionRecallDisplay.from_predictions(
|
| 138 |
+
y, y_pred, pos_label=pos_label
|
| 139 |
+
)
|
| 140 |
+
|
| 141 |
+
average_precision = average_precision_score(y, y_pred, pos_label=pos_label)
|
| 142 |
+
|
| 143 |
+
# check that the default name is used
|
| 144 |
+
assert display.line_.get_label() == default_label.format(average_precision)
|
| 145 |
+
|
| 146 |
+
# check that the name can be set
|
| 147 |
+
display.plot(name="MySpecialEstimator")
|
| 148 |
+
assert (
|
| 149 |
+
display.line_.get_label()
|
| 150 |
+
== f"MySpecialEstimator (AP = {average_precision:.2f})"
|
| 151 |
+
)
|
| 152 |
+
|
| 153 |
+
|
| 154 |
+
@pytest.mark.parametrize(
|
| 155 |
+
"clf",
|
| 156 |
+
[
|
| 157 |
+
make_pipeline(StandardScaler(), LogisticRegression()),
|
| 158 |
+
make_pipeline(
|
| 159 |
+
make_column_transformer((StandardScaler(), [0, 1])), LogisticRegression()
|
| 160 |
+
),
|
| 161 |
+
],
|
| 162 |
+
)
|
| 163 |
+
def test_precision_recall_display_pipeline(pyplot, clf):
|
| 164 |
+
X, y = make_classification(n_classes=2, n_samples=50, random_state=0)
|
| 165 |
+
with pytest.raises(NotFittedError):
|
| 166 |
+
PrecisionRecallDisplay.from_estimator(clf, X, y)
|
| 167 |
+
clf.fit(X, y)
|
| 168 |
+
display = PrecisionRecallDisplay.from_estimator(clf, X, y)
|
| 169 |
+
assert display.estimator_name == clf.__class__.__name__
|
| 170 |
+
|
| 171 |
+
|
| 172 |
+
def test_precision_recall_display_string_labels(pyplot):
|
| 173 |
+
# regression test #15738
|
| 174 |
+
cancer = load_breast_cancer()
|
| 175 |
+
X, y = cancer.data, cancer.target_names[cancer.target]
|
| 176 |
+
|
| 177 |
+
lr = make_pipeline(StandardScaler(), LogisticRegression())
|
| 178 |
+
lr.fit(X, y)
|
| 179 |
+
for klass in cancer.target_names:
|
| 180 |
+
assert klass in lr.classes_
|
| 181 |
+
display = PrecisionRecallDisplay.from_estimator(lr, X, y)
|
| 182 |
+
|
| 183 |
+
y_pred = lr.predict_proba(X)[:, 1]
|
| 184 |
+
avg_prec = average_precision_score(y, y_pred, pos_label=lr.classes_[1])
|
| 185 |
+
|
| 186 |
+
assert display.average_precision == pytest.approx(avg_prec)
|
| 187 |
+
assert display.estimator_name == lr.__class__.__name__
|
| 188 |
+
|
| 189 |
+
err_msg = r"y_true takes value in {'benign', 'malignant'}"
|
| 190 |
+
with pytest.raises(ValueError, match=err_msg):
|
| 191 |
+
PrecisionRecallDisplay.from_predictions(y, y_pred)
|
| 192 |
+
|
| 193 |
+
display = PrecisionRecallDisplay.from_predictions(
|
| 194 |
+
y, y_pred, pos_label=lr.classes_[1]
|
| 195 |
+
)
|
| 196 |
+
assert display.average_precision == pytest.approx(avg_prec)
|
| 197 |
+
|
| 198 |
+
|
| 199 |
+
@pytest.mark.parametrize(
|
| 200 |
+
"average_precision, estimator_name, expected_label",
|
| 201 |
+
[
|
| 202 |
+
(0.9, None, "AP = 0.90"),
|
| 203 |
+
(None, "my_est", "my_est"),
|
| 204 |
+
(0.8, "my_est2", "my_est2 (AP = 0.80)"),
|
| 205 |
+
],
|
| 206 |
+
)
|
| 207 |
+
def test_default_labels(pyplot, average_precision, estimator_name, expected_label):
|
| 208 |
+
"""Check the default labels used in the display."""
|
| 209 |
+
precision = np.array([1, 0.5, 0])
|
| 210 |
+
recall = np.array([0, 0.5, 1])
|
| 211 |
+
display = PrecisionRecallDisplay(
|
| 212 |
+
precision,
|
| 213 |
+
recall,
|
| 214 |
+
average_precision=average_precision,
|
| 215 |
+
estimator_name=estimator_name,
|
| 216 |
+
)
|
| 217 |
+
display.plot()
|
| 218 |
+
assert display.line_.get_label() == expected_label
|
| 219 |
+
|
| 220 |
+
|
| 221 |
+
@pytest.mark.parametrize("constructor_name", ["from_estimator", "from_predictions"])
|
| 222 |
+
@pytest.mark.parametrize("response_method", ["predict_proba", "decision_function"])
|
| 223 |
+
def test_plot_precision_recall_pos_label(pyplot, constructor_name, response_method):
|
| 224 |
+
# check that we can provide the positive label and display the proper
|
| 225 |
+
# statistics
|
| 226 |
+
X, y = load_breast_cancer(return_X_y=True)
|
| 227 |
+
# create an highly imbalanced version of the breast cancer dataset
|
| 228 |
+
idx_positive = np.flatnonzero(y == 1)
|
| 229 |
+
idx_negative = np.flatnonzero(y == 0)
|
| 230 |
+
idx_selected = np.hstack([idx_negative, idx_positive[:25]])
|
| 231 |
+
X, y = X[idx_selected], y[idx_selected]
|
| 232 |
+
X, y = shuffle(X, y, random_state=42)
|
| 233 |
+
# only use 2 features to make the problem even harder
|
| 234 |
+
X = X[:, :2]
|
| 235 |
+
y = np.array(["cancer" if c == 1 else "not cancer" for c in y], dtype=object)
|
| 236 |
+
X_train, X_test, y_train, y_test = train_test_split(
|
| 237 |
+
X,
|
| 238 |
+
y,
|
| 239 |
+
stratify=y,
|
| 240 |
+
random_state=0,
|
| 241 |
+
)
|
| 242 |
+
|
| 243 |
+
classifier = LogisticRegression()
|
| 244 |
+
classifier.fit(X_train, y_train)
|
| 245 |
+
|
| 246 |
+
# sanity check to be sure the positive class is classes_[0] and that we
|
| 247 |
+
# are betrayed by the class imbalance
|
| 248 |
+
assert classifier.classes_.tolist() == ["cancer", "not cancer"]
|
| 249 |
+
|
| 250 |
+
y_pred = getattr(classifier, response_method)(X_test)
|
| 251 |
+
# we select the corresponding probability columns or reverse the decision
|
| 252 |
+
# function otherwise
|
| 253 |
+
y_pred_cancer = -1 * y_pred if y_pred.ndim == 1 else y_pred[:, 0]
|
| 254 |
+
y_pred_not_cancer = y_pred if y_pred.ndim == 1 else y_pred[:, 1]
|
| 255 |
+
|
| 256 |
+
if constructor_name == "from_estimator":
|
| 257 |
+
display = PrecisionRecallDisplay.from_estimator(
|
| 258 |
+
classifier,
|
| 259 |
+
X_test,
|
| 260 |
+
y_test,
|
| 261 |
+
pos_label="cancer",
|
| 262 |
+
response_method=response_method,
|
| 263 |
+
)
|
| 264 |
+
else:
|
| 265 |
+
display = PrecisionRecallDisplay.from_predictions(
|
| 266 |
+
y_test,
|
| 267 |
+
y_pred_cancer,
|
| 268 |
+
pos_label="cancer",
|
| 269 |
+
)
|
| 270 |
+
# we should obtain the statistics of the "cancer" class
|
| 271 |
+
avg_prec_limit = 0.65
|
| 272 |
+
assert display.average_precision < avg_prec_limit
|
| 273 |
+
assert -np.trapz(display.precision, display.recall) < avg_prec_limit
|
| 274 |
+
|
| 275 |
+
# otherwise we should obtain the statistics of the "not cancer" class
|
| 276 |
+
if constructor_name == "from_estimator":
|
| 277 |
+
display = PrecisionRecallDisplay.from_estimator(
|
| 278 |
+
classifier,
|
| 279 |
+
X_test,
|
| 280 |
+
y_test,
|
| 281 |
+
response_method=response_method,
|
| 282 |
+
pos_label="not cancer",
|
| 283 |
+
)
|
| 284 |
+
else:
|
| 285 |
+
display = PrecisionRecallDisplay.from_predictions(
|
| 286 |
+
y_test,
|
| 287 |
+
y_pred_not_cancer,
|
| 288 |
+
pos_label="not cancer",
|
| 289 |
+
)
|
| 290 |
+
avg_prec_limit = 0.95
|
| 291 |
+
assert display.average_precision > avg_prec_limit
|
| 292 |
+
assert -np.trapz(display.precision, display.recall) > avg_prec_limit
|
videollama2/lib/python3.10/site-packages/sklearn/metrics/_plot/tests/test_predict_error_display.py
ADDED
|
@@ -0,0 +1,163 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import pytest
|
| 2 |
+
|
| 3 |
+
from numpy.testing import assert_allclose
|
| 4 |
+
|
| 5 |
+
from sklearn.datasets import load_diabetes
|
| 6 |
+
from sklearn.exceptions import NotFittedError
|
| 7 |
+
from sklearn.linear_model import Ridge
|
| 8 |
+
|
| 9 |
+
from sklearn.metrics import PredictionErrorDisplay
|
| 10 |
+
|
| 11 |
+
X, y = load_diabetes(return_X_y=True)
|
| 12 |
+
|
| 13 |
+
|
| 14 |
+
@pytest.fixture
|
| 15 |
+
def regressor_fitted():
|
| 16 |
+
return Ridge().fit(X, y)
|
| 17 |
+
|
| 18 |
+
|
| 19 |
+
@pytest.mark.parametrize(
|
| 20 |
+
"regressor, params, err_type, err_msg",
|
| 21 |
+
[
|
| 22 |
+
(
|
| 23 |
+
Ridge().fit(X, y),
|
| 24 |
+
{"subsample": -1},
|
| 25 |
+
ValueError,
|
| 26 |
+
"When an integer, subsample=-1 should be",
|
| 27 |
+
),
|
| 28 |
+
(
|
| 29 |
+
Ridge().fit(X, y),
|
| 30 |
+
{"subsample": 20.0},
|
| 31 |
+
ValueError,
|
| 32 |
+
"When a floating-point, subsample=20.0 should be",
|
| 33 |
+
),
|
| 34 |
+
(
|
| 35 |
+
Ridge().fit(X, y),
|
| 36 |
+
{"subsample": -20.0},
|
| 37 |
+
ValueError,
|
| 38 |
+
"When a floating-point, subsample=-20.0 should be",
|
| 39 |
+
),
|
| 40 |
+
(
|
| 41 |
+
Ridge().fit(X, y),
|
| 42 |
+
{"kind": "xxx"},
|
| 43 |
+
ValueError,
|
| 44 |
+
"`kind` must be one of",
|
| 45 |
+
),
|
| 46 |
+
],
|
| 47 |
+
)
|
| 48 |
+
@pytest.mark.parametrize("class_method", ["from_estimator", "from_predictions"])
|
| 49 |
+
def test_prediction_error_display_raise_error(
|
| 50 |
+
pyplot, class_method, regressor, params, err_type, err_msg
|
| 51 |
+
):
|
| 52 |
+
"""Check that we raise the proper error when making the parameters
|
| 53 |
+
# validation."""
|
| 54 |
+
with pytest.raises(err_type, match=err_msg):
|
| 55 |
+
if class_method == "from_estimator":
|
| 56 |
+
PredictionErrorDisplay.from_estimator(regressor, X, y, **params)
|
| 57 |
+
else:
|
| 58 |
+
y_pred = regressor.predict(X)
|
| 59 |
+
PredictionErrorDisplay.from_predictions(y_true=y, y_pred=y_pred, **params)
|
| 60 |
+
|
| 61 |
+
|
| 62 |
+
def test_from_estimator_not_fitted(pyplot):
|
| 63 |
+
"""Check that we raise a `NotFittedError` when the passed regressor is not
|
| 64 |
+
fit."""
|
| 65 |
+
regressor = Ridge()
|
| 66 |
+
with pytest.raises(NotFittedError, match="is not fitted yet."):
|
| 67 |
+
PredictionErrorDisplay.from_estimator(regressor, X, y)
|
| 68 |
+
|
| 69 |
+
|
| 70 |
+
@pytest.mark.parametrize("class_method", ["from_estimator", "from_predictions"])
|
| 71 |
+
@pytest.mark.parametrize("kind", ["actual_vs_predicted", "residual_vs_predicted"])
|
| 72 |
+
def test_prediction_error_display(pyplot, regressor_fitted, class_method, kind):
|
| 73 |
+
"""Check the default behaviour of the display."""
|
| 74 |
+
if class_method == "from_estimator":
|
| 75 |
+
display = PredictionErrorDisplay.from_estimator(
|
| 76 |
+
regressor_fitted, X, y, kind=kind
|
| 77 |
+
)
|
| 78 |
+
else:
|
| 79 |
+
y_pred = regressor_fitted.predict(X)
|
| 80 |
+
display = PredictionErrorDisplay.from_predictions(
|
| 81 |
+
y_true=y, y_pred=y_pred, kind=kind
|
| 82 |
+
)
|
| 83 |
+
|
| 84 |
+
if kind == "actual_vs_predicted":
|
| 85 |
+
assert_allclose(display.line_.get_xdata(), display.line_.get_ydata())
|
| 86 |
+
assert display.ax_.get_xlabel() == "Predicted values"
|
| 87 |
+
assert display.ax_.get_ylabel() == "Actual values"
|
| 88 |
+
assert display.line_ is not None
|
| 89 |
+
else:
|
| 90 |
+
assert display.ax_.get_xlabel() == "Predicted values"
|
| 91 |
+
assert display.ax_.get_ylabel() == "Residuals (actual - predicted)"
|
| 92 |
+
assert display.line_ is not None
|
| 93 |
+
|
| 94 |
+
assert display.ax_.get_legend() is None
|
| 95 |
+
|
| 96 |
+
|
| 97 |
+
@pytest.mark.parametrize("class_method", ["from_estimator", "from_predictions"])
|
| 98 |
+
@pytest.mark.parametrize(
|
| 99 |
+
"subsample, expected_size",
|
| 100 |
+
[(5, 5), (0.1, int(X.shape[0] * 0.1)), (None, X.shape[0])],
|
| 101 |
+
)
|
| 102 |
+
def test_plot_prediction_error_subsample(
|
| 103 |
+
pyplot, regressor_fitted, class_method, subsample, expected_size
|
| 104 |
+
):
|
| 105 |
+
"""Check the behaviour of `subsample`."""
|
| 106 |
+
if class_method == "from_estimator":
|
| 107 |
+
display = PredictionErrorDisplay.from_estimator(
|
| 108 |
+
regressor_fitted, X, y, subsample=subsample
|
| 109 |
+
)
|
| 110 |
+
else:
|
| 111 |
+
y_pred = regressor_fitted.predict(X)
|
| 112 |
+
display = PredictionErrorDisplay.from_predictions(
|
| 113 |
+
y_true=y, y_pred=y_pred, subsample=subsample
|
| 114 |
+
)
|
| 115 |
+
assert len(display.scatter_.get_offsets()) == expected_size
|
| 116 |
+
|
| 117 |
+
|
| 118 |
+
@pytest.mark.parametrize("class_method", ["from_estimator", "from_predictions"])
|
| 119 |
+
def test_plot_prediction_error_ax(pyplot, regressor_fitted, class_method):
|
| 120 |
+
"""Check that we can pass an axis to the display."""
|
| 121 |
+
_, ax = pyplot.subplots()
|
| 122 |
+
if class_method == "from_estimator":
|
| 123 |
+
display = PredictionErrorDisplay.from_estimator(regressor_fitted, X, y, ax=ax)
|
| 124 |
+
else:
|
| 125 |
+
y_pred = regressor_fitted.predict(X)
|
| 126 |
+
display = PredictionErrorDisplay.from_predictions(
|
| 127 |
+
y_true=y, y_pred=y_pred, ax=ax
|
| 128 |
+
)
|
| 129 |
+
assert display.ax_ is ax
|
| 130 |
+
|
| 131 |
+
|
| 132 |
+
@pytest.mark.parametrize("class_method", ["from_estimator", "from_predictions"])
|
| 133 |
+
def test_prediction_error_custom_artist(pyplot, regressor_fitted, class_method):
|
| 134 |
+
"""Check that we can tune the style of the lines."""
|
| 135 |
+
extra_params = {
|
| 136 |
+
"kind": "actual_vs_predicted",
|
| 137 |
+
"scatter_kwargs": {"color": "red"},
|
| 138 |
+
"line_kwargs": {"color": "black"},
|
| 139 |
+
}
|
| 140 |
+
if class_method == "from_estimator":
|
| 141 |
+
display = PredictionErrorDisplay.from_estimator(
|
| 142 |
+
regressor_fitted, X, y, **extra_params
|
| 143 |
+
)
|
| 144 |
+
else:
|
| 145 |
+
y_pred = regressor_fitted.predict(X)
|
| 146 |
+
display = PredictionErrorDisplay.from_predictions(
|
| 147 |
+
y_true=y, y_pred=y_pred, **extra_params
|
| 148 |
+
)
|
| 149 |
+
|
| 150 |
+
assert display.line_.get_color() == "black"
|
| 151 |
+
assert_allclose(display.scatter_.get_edgecolor(), [[1.0, 0.0, 0.0, 0.8]])
|
| 152 |
+
|
| 153 |
+
# create a display with the default values
|
| 154 |
+
if class_method == "from_estimator":
|
| 155 |
+
display = PredictionErrorDisplay.from_estimator(regressor_fitted, X, y)
|
| 156 |
+
else:
|
| 157 |
+
y_pred = regressor_fitted.predict(X)
|
| 158 |
+
display = PredictionErrorDisplay.from_predictions(y_true=y, y_pred=y_pred)
|
| 159 |
+
pyplot.close("all")
|
| 160 |
+
|
| 161 |
+
display.plot(**extra_params)
|
| 162 |
+
assert display.line_.get_color() == "black"
|
| 163 |
+
assert_allclose(display.scatter_.get_edgecolor(), [[1.0, 0.0, 0.0, 0.8]])
|
videollama2/lib/python3.10/site-packages/sklearn/metrics/_regression.py
ADDED
|
@@ -0,0 +1,1497 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""Metrics to assess performance on regression task.
|
| 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 |
+
# Karan Desai <karandesai281196@gmail.com>
|
| 18 |
+
# Noel Dawe <noel@dawe.me>
|
| 19 |
+
# Manoj Kumar <manojkumarsivaraj334@gmail.com>
|
| 20 |
+
# Michael Eickenberg <michael.eickenberg@gmail.com>
|
| 21 |
+
# Konstantin Shmelkov <konstantin.shmelkov@polytechnique.edu>
|
| 22 |
+
# Christian Lorentzen <lorentzen.ch@gmail.com>
|
| 23 |
+
# Ashutosh Hathidara <ashutoshhathidara98@gmail.com>
|
| 24 |
+
# Uttam kumar <bajiraouttamsinha@gmail.com>
|
| 25 |
+
# Sylvain Marie <sylvain.marie@se.com>
|
| 26 |
+
# Ohad Michel <ohadmich@gmail.com>
|
| 27 |
+
# License: BSD 3 clause
|
| 28 |
+
|
| 29 |
+
import numbers
|
| 30 |
+
import warnings
|
| 31 |
+
|
| 32 |
+
import numpy as np
|
| 33 |
+
from scipy.special import xlogy
|
| 34 |
+
|
| 35 |
+
from ..exceptions import UndefinedMetricWarning
|
| 36 |
+
from ..utils.validation import (
|
| 37 |
+
check_array,
|
| 38 |
+
check_consistent_length,
|
| 39 |
+
check_scalar,
|
| 40 |
+
_num_samples,
|
| 41 |
+
column_or_1d,
|
| 42 |
+
_check_sample_weight,
|
| 43 |
+
)
|
| 44 |
+
from ..utils.stats import _weighted_percentile
|
| 45 |
+
|
| 46 |
+
|
| 47 |
+
__ALL__ = [
|
| 48 |
+
"max_error",
|
| 49 |
+
"mean_absolute_error",
|
| 50 |
+
"mean_squared_error",
|
| 51 |
+
"mean_squared_log_error",
|
| 52 |
+
"median_absolute_error",
|
| 53 |
+
"mean_absolute_percentage_error",
|
| 54 |
+
"mean_pinball_loss",
|
| 55 |
+
"r2_score",
|
| 56 |
+
"explained_variance_score",
|
| 57 |
+
"mean_tweedie_deviance",
|
| 58 |
+
"mean_poisson_deviance",
|
| 59 |
+
"mean_gamma_deviance",
|
| 60 |
+
"d2_tweedie_score",
|
| 61 |
+
"d2_pinball_score",
|
| 62 |
+
"d2_absolute_error_score",
|
| 63 |
+
]
|
| 64 |
+
|
| 65 |
+
|
| 66 |
+
def _check_reg_targets(y_true, y_pred, multioutput, dtype="numeric"):
|
| 67 |
+
"""Check that y_true and y_pred belong to the same regression task.
|
| 68 |
+
|
| 69 |
+
Parameters
|
| 70 |
+
----------
|
| 71 |
+
y_true : array-like
|
| 72 |
+
|
| 73 |
+
y_pred : array-like
|
| 74 |
+
|
| 75 |
+
multioutput : array-like or string in ['raw_values', uniform_average',
|
| 76 |
+
'variance_weighted'] or None
|
| 77 |
+
None is accepted due to backward compatibility of r2_score().
|
| 78 |
+
|
| 79 |
+
dtype : str or list, default="numeric"
|
| 80 |
+
the dtype argument passed to check_array.
|
| 81 |
+
|
| 82 |
+
Returns
|
| 83 |
+
-------
|
| 84 |
+
type_true : one of {'continuous', continuous-multioutput'}
|
| 85 |
+
The type of the true target data, as output by
|
| 86 |
+
'utils.multiclass.type_of_target'.
|
| 87 |
+
|
| 88 |
+
y_true : array-like of shape (n_samples, n_outputs)
|
| 89 |
+
Ground truth (correct) target values.
|
| 90 |
+
|
| 91 |
+
y_pred : array-like of shape (n_samples, n_outputs)
|
| 92 |
+
Estimated target values.
|
| 93 |
+
|
| 94 |
+
multioutput : array-like of shape (n_outputs) or string in ['raw_values',
|
| 95 |
+
uniform_average', 'variance_weighted'] or None
|
| 96 |
+
Custom output weights if ``multioutput`` is array-like or
|
| 97 |
+
just the corresponding argument if ``multioutput`` is a
|
| 98 |
+
correct keyword.
|
| 99 |
+
"""
|
| 100 |
+
check_consistent_length(y_true, y_pred)
|
| 101 |
+
y_true = check_array(y_true, ensure_2d=False, dtype=dtype)
|
| 102 |
+
y_pred = check_array(y_pred, ensure_2d=False, dtype=dtype)
|
| 103 |
+
|
| 104 |
+
if y_true.ndim == 1:
|
| 105 |
+
y_true = y_true.reshape((-1, 1))
|
| 106 |
+
|
| 107 |
+
if y_pred.ndim == 1:
|
| 108 |
+
y_pred = y_pred.reshape((-1, 1))
|
| 109 |
+
|
| 110 |
+
if y_true.shape[1] != y_pred.shape[1]:
|
| 111 |
+
raise ValueError(
|
| 112 |
+
"y_true and y_pred have different number of output ({0}!={1})".format(
|
| 113 |
+
y_true.shape[1], y_pred.shape[1]
|
| 114 |
+
)
|
| 115 |
+
)
|
| 116 |
+
|
| 117 |
+
n_outputs = y_true.shape[1]
|
| 118 |
+
allowed_multioutput_str = ("raw_values", "uniform_average", "variance_weighted")
|
| 119 |
+
if isinstance(multioutput, str):
|
| 120 |
+
if multioutput not in allowed_multioutput_str:
|
| 121 |
+
raise ValueError(
|
| 122 |
+
"Allowed 'multioutput' string values are {}. "
|
| 123 |
+
"You provided multioutput={!r}".format(
|
| 124 |
+
allowed_multioutput_str, multioutput
|
| 125 |
+
)
|
| 126 |
+
)
|
| 127 |
+
elif multioutput is not None:
|
| 128 |
+
multioutput = check_array(multioutput, ensure_2d=False)
|
| 129 |
+
if n_outputs == 1:
|
| 130 |
+
raise ValueError("Custom weights are useful only in multi-output cases.")
|
| 131 |
+
elif n_outputs != len(multioutput):
|
| 132 |
+
raise ValueError(
|
| 133 |
+
"There must be equally many custom weights (%d) as outputs (%d)."
|
| 134 |
+
% (len(multioutput), n_outputs)
|
| 135 |
+
)
|
| 136 |
+
y_type = "continuous" if n_outputs == 1 else "continuous-multioutput"
|
| 137 |
+
|
| 138 |
+
return y_type, y_true, y_pred, multioutput
|
| 139 |
+
|
| 140 |
+
|
| 141 |
+
def mean_absolute_error(
|
| 142 |
+
y_true, y_pred, *, sample_weight=None, multioutput="uniform_average"
|
| 143 |
+
):
|
| 144 |
+
"""Mean absolute error regression loss.
|
| 145 |
+
|
| 146 |
+
Read more in the :ref:`User Guide <mean_absolute_error>`.
|
| 147 |
+
|
| 148 |
+
Parameters
|
| 149 |
+
----------
|
| 150 |
+
y_true : array-like of shape (n_samples,) or (n_samples, n_outputs)
|
| 151 |
+
Ground truth (correct) target values.
|
| 152 |
+
|
| 153 |
+
y_pred : array-like of shape (n_samples,) or (n_samples, n_outputs)
|
| 154 |
+
Estimated target values.
|
| 155 |
+
|
| 156 |
+
sample_weight : array-like of shape (n_samples,), default=None
|
| 157 |
+
Sample weights.
|
| 158 |
+
|
| 159 |
+
multioutput : {'raw_values', 'uniform_average'} or array-like of shape \
|
| 160 |
+
(n_outputs,), default='uniform_average'
|
| 161 |
+
Defines aggregating of multiple output values.
|
| 162 |
+
Array-like value defines weights used to average errors.
|
| 163 |
+
|
| 164 |
+
'raw_values' :
|
| 165 |
+
Returns a full set of errors in case of multioutput input.
|
| 166 |
+
|
| 167 |
+
'uniform_average' :
|
| 168 |
+
Errors of all outputs are averaged with uniform weight.
|
| 169 |
+
|
| 170 |
+
Returns
|
| 171 |
+
-------
|
| 172 |
+
loss : float or ndarray of floats
|
| 173 |
+
If multioutput is 'raw_values', then mean absolute error is returned
|
| 174 |
+
for each output separately.
|
| 175 |
+
If multioutput is 'uniform_average' or an ndarray of weights, then the
|
| 176 |
+
weighted average of all output errors is returned.
|
| 177 |
+
|
| 178 |
+
MAE output is non-negative floating point. The best value is 0.0.
|
| 179 |
+
|
| 180 |
+
Examples
|
| 181 |
+
--------
|
| 182 |
+
>>> from sklearn.metrics import mean_absolute_error
|
| 183 |
+
>>> y_true = [3, -0.5, 2, 7]
|
| 184 |
+
>>> y_pred = [2.5, 0.0, 2, 8]
|
| 185 |
+
>>> mean_absolute_error(y_true, y_pred)
|
| 186 |
+
0.5
|
| 187 |
+
>>> y_true = [[0.5, 1], [-1, 1], [7, -6]]
|
| 188 |
+
>>> y_pred = [[0, 2], [-1, 2], [8, -5]]
|
| 189 |
+
>>> mean_absolute_error(y_true, y_pred)
|
| 190 |
+
0.75
|
| 191 |
+
>>> mean_absolute_error(y_true, y_pred, multioutput='raw_values')
|
| 192 |
+
array([0.5, 1. ])
|
| 193 |
+
>>> mean_absolute_error(y_true, y_pred, multioutput=[0.3, 0.7])
|
| 194 |
+
0.85...
|
| 195 |
+
"""
|
| 196 |
+
y_type, y_true, y_pred, multioutput = _check_reg_targets(
|
| 197 |
+
y_true, y_pred, multioutput
|
| 198 |
+
)
|
| 199 |
+
check_consistent_length(y_true, y_pred, sample_weight)
|
| 200 |
+
output_errors = np.average(np.abs(y_pred - y_true), weights=sample_weight, axis=0)
|
| 201 |
+
if isinstance(multioutput, str):
|
| 202 |
+
if multioutput == "raw_values":
|
| 203 |
+
return output_errors
|
| 204 |
+
elif multioutput == "uniform_average":
|
| 205 |
+
# pass None as weights to np.average: uniform mean
|
| 206 |
+
multioutput = None
|
| 207 |
+
|
| 208 |
+
return np.average(output_errors, weights=multioutput)
|
| 209 |
+
|
| 210 |
+
|
| 211 |
+
def mean_pinball_loss(
|
| 212 |
+
y_true, y_pred, *, sample_weight=None, alpha=0.5, multioutput="uniform_average"
|
| 213 |
+
):
|
| 214 |
+
"""Pinball loss for quantile regression.
|
| 215 |
+
|
| 216 |
+
Read more in the :ref:`User Guide <pinball_loss>`.
|
| 217 |
+
|
| 218 |
+
Parameters
|
| 219 |
+
----------
|
| 220 |
+
y_true : array-like of shape (n_samples,) or (n_samples, n_outputs)
|
| 221 |
+
Ground truth (correct) target values.
|
| 222 |
+
|
| 223 |
+
y_pred : array-like of shape (n_samples,) or (n_samples, n_outputs)
|
| 224 |
+
Estimated target values.
|
| 225 |
+
|
| 226 |
+
sample_weight : array-like of shape (n_samples,), default=None
|
| 227 |
+
Sample weights.
|
| 228 |
+
|
| 229 |
+
alpha : float, slope of the pinball loss, default=0.5,
|
| 230 |
+
This loss is equivalent to :ref:`mean_absolute_error` when `alpha=0.5`,
|
| 231 |
+
`alpha=0.95` is minimized by estimators of the 95th percentile.
|
| 232 |
+
|
| 233 |
+
multioutput : {'raw_values', 'uniform_average'} or array-like of shape \
|
| 234 |
+
(n_outputs,), default='uniform_average'
|
| 235 |
+
Defines aggregating of multiple output values.
|
| 236 |
+
Array-like value defines weights used to average errors.
|
| 237 |
+
|
| 238 |
+
'raw_values' :
|
| 239 |
+
Returns a full set of errors in case of multioutput input.
|
| 240 |
+
|
| 241 |
+
'uniform_average' :
|
| 242 |
+
Errors of all outputs are averaged with uniform weight.
|
| 243 |
+
|
| 244 |
+
Returns
|
| 245 |
+
-------
|
| 246 |
+
loss : float or ndarray of floats
|
| 247 |
+
If multioutput is 'raw_values', then mean absolute error is returned
|
| 248 |
+
for each output separately.
|
| 249 |
+
If multioutput is 'uniform_average' or an ndarray of weights, then the
|
| 250 |
+
weighted average of all output errors is returned.
|
| 251 |
+
|
| 252 |
+
The pinball loss output is a non-negative floating point. The best
|
| 253 |
+
value is 0.0.
|
| 254 |
+
|
| 255 |
+
Examples
|
| 256 |
+
--------
|
| 257 |
+
>>> from sklearn.metrics import mean_pinball_loss
|
| 258 |
+
>>> y_true = [1, 2, 3]
|
| 259 |
+
>>> mean_pinball_loss(y_true, [0, 2, 3], alpha=0.1)
|
| 260 |
+
0.03...
|
| 261 |
+
>>> mean_pinball_loss(y_true, [1, 2, 4], alpha=0.1)
|
| 262 |
+
0.3...
|
| 263 |
+
>>> mean_pinball_loss(y_true, [0, 2, 3], alpha=0.9)
|
| 264 |
+
0.3...
|
| 265 |
+
>>> mean_pinball_loss(y_true, [1, 2, 4], alpha=0.9)
|
| 266 |
+
0.03...
|
| 267 |
+
>>> mean_pinball_loss(y_true, y_true, alpha=0.1)
|
| 268 |
+
0.0
|
| 269 |
+
>>> mean_pinball_loss(y_true, y_true, alpha=0.9)
|
| 270 |
+
0.0
|
| 271 |
+
"""
|
| 272 |
+
y_type, y_true, y_pred, multioutput = _check_reg_targets(
|
| 273 |
+
y_true, y_pred, multioutput
|
| 274 |
+
)
|
| 275 |
+
check_consistent_length(y_true, y_pred, sample_weight)
|
| 276 |
+
diff = y_true - y_pred
|
| 277 |
+
sign = (diff >= 0).astype(diff.dtype)
|
| 278 |
+
loss = alpha * sign * diff - (1 - alpha) * (1 - sign) * diff
|
| 279 |
+
output_errors = np.average(loss, weights=sample_weight, axis=0)
|
| 280 |
+
if isinstance(multioutput, str):
|
| 281 |
+
if multioutput == "raw_values":
|
| 282 |
+
return output_errors
|
| 283 |
+
elif multioutput == "uniform_average":
|
| 284 |
+
# pass None as weights to np.average: uniform mean
|
| 285 |
+
multioutput = None
|
| 286 |
+
else:
|
| 287 |
+
raise ValueError(
|
| 288 |
+
"multioutput is expected to be 'raw_values' "
|
| 289 |
+
"or 'uniform_average' but we got %r"
|
| 290 |
+
" instead." % multioutput
|
| 291 |
+
)
|
| 292 |
+
|
| 293 |
+
return np.average(output_errors, weights=multioutput)
|
| 294 |
+
|
| 295 |
+
|
| 296 |
+
def mean_absolute_percentage_error(
|
| 297 |
+
y_true, y_pred, *, sample_weight=None, multioutput="uniform_average"
|
| 298 |
+
):
|
| 299 |
+
"""Mean absolute percentage error (MAPE) regression loss.
|
| 300 |
+
|
| 301 |
+
Note here that the output is not a percentage in the range [0, 100]
|
| 302 |
+
and a value of 100 does not mean 100% but 1e2. Furthermore, the output
|
| 303 |
+
can be arbitrarily high when `y_true` is small (which is specific to the
|
| 304 |
+
metric) or when `abs(y_true - y_pred)` is large (which is common for most
|
| 305 |
+
regression metrics). Read more in the
|
| 306 |
+
:ref:`User Guide <mean_absolute_percentage_error>`.
|
| 307 |
+
|
| 308 |
+
.. versionadded:: 0.24
|
| 309 |
+
|
| 310 |
+
Parameters
|
| 311 |
+
----------
|
| 312 |
+
y_true : array-like of shape (n_samples,) or (n_samples, n_outputs)
|
| 313 |
+
Ground truth (correct) target values.
|
| 314 |
+
|
| 315 |
+
y_pred : array-like of shape (n_samples,) or (n_samples, n_outputs)
|
| 316 |
+
Estimated target values.
|
| 317 |
+
|
| 318 |
+
sample_weight : array-like of shape (n_samples,), default=None
|
| 319 |
+
Sample weights.
|
| 320 |
+
|
| 321 |
+
multioutput : {'raw_values', 'uniform_average'} or array-like
|
| 322 |
+
Defines aggregating of multiple output values.
|
| 323 |
+
Array-like value defines weights used to average errors.
|
| 324 |
+
If input is list then the shape must be (n_outputs,).
|
| 325 |
+
|
| 326 |
+
'raw_values' :
|
| 327 |
+
Returns a full set of errors in case of multioutput input.
|
| 328 |
+
|
| 329 |
+
'uniform_average' :
|
| 330 |
+
Errors of all outputs are averaged with uniform weight.
|
| 331 |
+
|
| 332 |
+
Returns
|
| 333 |
+
-------
|
| 334 |
+
loss : float or ndarray of floats
|
| 335 |
+
If multioutput is 'raw_values', then mean absolute percentage error
|
| 336 |
+
is returned for each output separately.
|
| 337 |
+
If multioutput is 'uniform_average' or an ndarray of weights, then the
|
| 338 |
+
weighted average of all output errors is returned.
|
| 339 |
+
|
| 340 |
+
MAPE output is non-negative floating point. The best value is 0.0.
|
| 341 |
+
But note that bad predictions can lead to arbitrarily large
|
| 342 |
+
MAPE values, especially if some `y_true` values are very close to zero.
|
| 343 |
+
Note that we return a large value instead of `inf` when `y_true` is zero.
|
| 344 |
+
|
| 345 |
+
Examples
|
| 346 |
+
--------
|
| 347 |
+
>>> from sklearn.metrics import mean_absolute_percentage_error
|
| 348 |
+
>>> y_true = [3, -0.5, 2, 7]
|
| 349 |
+
>>> y_pred = [2.5, 0.0, 2, 8]
|
| 350 |
+
>>> mean_absolute_percentage_error(y_true, y_pred)
|
| 351 |
+
0.3273...
|
| 352 |
+
>>> y_true = [[0.5, 1], [-1, 1], [7, -6]]
|
| 353 |
+
>>> y_pred = [[0, 2], [-1, 2], [8, -5]]
|
| 354 |
+
>>> mean_absolute_percentage_error(y_true, y_pred)
|
| 355 |
+
0.5515...
|
| 356 |
+
>>> mean_absolute_percentage_error(y_true, y_pred, multioutput=[0.3, 0.7])
|
| 357 |
+
0.6198...
|
| 358 |
+
>>> # the value when some element of the y_true is zero is arbitrarily high because
|
| 359 |
+
>>> # of the division by epsilon
|
| 360 |
+
>>> y_true = [1., 0., 2.4, 7.]
|
| 361 |
+
>>> y_pred = [1.2, 0.1, 2.4, 8.]
|
| 362 |
+
>>> mean_absolute_percentage_error(y_true, y_pred)
|
| 363 |
+
112589990684262.48
|
| 364 |
+
"""
|
| 365 |
+
y_type, y_true, y_pred, multioutput = _check_reg_targets(
|
| 366 |
+
y_true, y_pred, multioutput
|
| 367 |
+
)
|
| 368 |
+
check_consistent_length(y_true, y_pred, sample_weight)
|
| 369 |
+
epsilon = np.finfo(np.float64).eps
|
| 370 |
+
mape = np.abs(y_pred - y_true) / np.maximum(np.abs(y_true), epsilon)
|
| 371 |
+
output_errors = np.average(mape, weights=sample_weight, axis=0)
|
| 372 |
+
if isinstance(multioutput, str):
|
| 373 |
+
if multioutput == "raw_values":
|
| 374 |
+
return output_errors
|
| 375 |
+
elif multioutput == "uniform_average":
|
| 376 |
+
# pass None as weights to np.average: uniform mean
|
| 377 |
+
multioutput = None
|
| 378 |
+
|
| 379 |
+
return np.average(output_errors, weights=multioutput)
|
| 380 |
+
|
| 381 |
+
|
| 382 |
+
def mean_squared_error(
|
| 383 |
+
y_true, y_pred, *, sample_weight=None, multioutput="uniform_average", squared=True
|
| 384 |
+
):
|
| 385 |
+
"""Mean squared error regression loss.
|
| 386 |
+
|
| 387 |
+
Read more in the :ref:`User Guide <mean_squared_error>`.
|
| 388 |
+
|
| 389 |
+
Parameters
|
| 390 |
+
----------
|
| 391 |
+
y_true : array-like of shape (n_samples,) or (n_samples, n_outputs)
|
| 392 |
+
Ground truth (correct) target values.
|
| 393 |
+
|
| 394 |
+
y_pred : array-like of shape (n_samples,) or (n_samples, n_outputs)
|
| 395 |
+
Estimated target values.
|
| 396 |
+
|
| 397 |
+
sample_weight : array-like of shape (n_samples,), default=None
|
| 398 |
+
Sample weights.
|
| 399 |
+
|
| 400 |
+
multioutput : {'raw_values', 'uniform_average'} or array-like of shape \
|
| 401 |
+
(n_outputs,), default='uniform_average'
|
| 402 |
+
Defines aggregating of multiple output values.
|
| 403 |
+
Array-like value defines weights used to average errors.
|
| 404 |
+
|
| 405 |
+
'raw_values' :
|
| 406 |
+
Returns a full set of errors in case of multioutput input.
|
| 407 |
+
|
| 408 |
+
'uniform_average' :
|
| 409 |
+
Errors of all outputs are averaged with uniform weight.
|
| 410 |
+
|
| 411 |
+
squared : bool, default=True
|
| 412 |
+
If True returns MSE value, if False returns RMSE value.
|
| 413 |
+
|
| 414 |
+
Returns
|
| 415 |
+
-------
|
| 416 |
+
loss : float or ndarray of floats
|
| 417 |
+
A non-negative floating point value (the best value is 0.0), or an
|
| 418 |
+
array of floating point values, one for each individual target.
|
| 419 |
+
|
| 420 |
+
Examples
|
| 421 |
+
--------
|
| 422 |
+
>>> from sklearn.metrics import mean_squared_error
|
| 423 |
+
>>> y_true = [3, -0.5, 2, 7]
|
| 424 |
+
>>> y_pred = [2.5, 0.0, 2, 8]
|
| 425 |
+
>>> mean_squared_error(y_true, y_pred)
|
| 426 |
+
0.375
|
| 427 |
+
>>> y_true = [3, -0.5, 2, 7]
|
| 428 |
+
>>> y_pred = [2.5, 0.0, 2, 8]
|
| 429 |
+
>>> mean_squared_error(y_true, y_pred, squared=False)
|
| 430 |
+
0.612...
|
| 431 |
+
>>> y_true = [[0.5, 1],[-1, 1],[7, -6]]
|
| 432 |
+
>>> y_pred = [[0, 2],[-1, 2],[8, -5]]
|
| 433 |
+
>>> mean_squared_error(y_true, y_pred)
|
| 434 |
+
0.708...
|
| 435 |
+
>>> mean_squared_error(y_true, y_pred, squared=False)
|
| 436 |
+
0.822...
|
| 437 |
+
>>> mean_squared_error(y_true, y_pred, multioutput='raw_values')
|
| 438 |
+
array([0.41666667, 1. ])
|
| 439 |
+
>>> mean_squared_error(y_true, y_pred, multioutput=[0.3, 0.7])
|
| 440 |
+
0.825...
|
| 441 |
+
"""
|
| 442 |
+
y_type, y_true, y_pred, multioutput = _check_reg_targets(
|
| 443 |
+
y_true, y_pred, multioutput
|
| 444 |
+
)
|
| 445 |
+
check_consistent_length(y_true, y_pred, sample_weight)
|
| 446 |
+
output_errors = np.average((y_true - y_pred) ** 2, axis=0, weights=sample_weight)
|
| 447 |
+
|
| 448 |
+
if not squared:
|
| 449 |
+
output_errors = np.sqrt(output_errors)
|
| 450 |
+
|
| 451 |
+
if isinstance(multioutput, str):
|
| 452 |
+
if multioutput == "raw_values":
|
| 453 |
+
return output_errors
|
| 454 |
+
elif multioutput == "uniform_average":
|
| 455 |
+
# pass None as weights to np.average: uniform mean
|
| 456 |
+
multioutput = None
|
| 457 |
+
|
| 458 |
+
return np.average(output_errors, weights=multioutput)
|
| 459 |
+
|
| 460 |
+
|
| 461 |
+
def mean_squared_log_error(
|
| 462 |
+
y_true, y_pred, *, sample_weight=None, multioutput="uniform_average", squared=True
|
| 463 |
+
):
|
| 464 |
+
"""Mean squared logarithmic error regression loss.
|
| 465 |
+
|
| 466 |
+
Read more in the :ref:`User Guide <mean_squared_log_error>`.
|
| 467 |
+
|
| 468 |
+
Parameters
|
| 469 |
+
----------
|
| 470 |
+
y_true : array-like of shape (n_samples,) or (n_samples, n_outputs)
|
| 471 |
+
Ground truth (correct) target values.
|
| 472 |
+
|
| 473 |
+
y_pred : array-like of shape (n_samples,) or (n_samples, n_outputs)
|
| 474 |
+
Estimated target values.
|
| 475 |
+
|
| 476 |
+
sample_weight : array-like of shape (n_samples,), default=None
|
| 477 |
+
Sample weights.
|
| 478 |
+
|
| 479 |
+
multioutput : {'raw_values', 'uniform_average'} or array-like of shape \
|
| 480 |
+
(n_outputs,), default='uniform_average'
|
| 481 |
+
|
| 482 |
+
Defines aggregating of multiple output values.
|
| 483 |
+
Array-like value defines weights used to average errors.
|
| 484 |
+
|
| 485 |
+
'raw_values' :
|
| 486 |
+
Returns a full set of errors when the input is of multioutput
|
| 487 |
+
format.
|
| 488 |
+
|
| 489 |
+
'uniform_average' :
|
| 490 |
+
Errors of all outputs are averaged with uniform weight.
|
| 491 |
+
squared : bool, default=True
|
| 492 |
+
If True returns MSLE (mean squared log error) value.
|
| 493 |
+
If False returns RMSLE (root mean squared log error) value.
|
| 494 |
+
|
| 495 |
+
Returns
|
| 496 |
+
-------
|
| 497 |
+
loss : float or ndarray of floats
|
| 498 |
+
A non-negative floating point value (the best value is 0.0), or an
|
| 499 |
+
array of floating point values, one for each individual target.
|
| 500 |
+
|
| 501 |
+
Examples
|
| 502 |
+
--------
|
| 503 |
+
>>> from sklearn.metrics import mean_squared_log_error
|
| 504 |
+
>>> y_true = [3, 5, 2.5, 7]
|
| 505 |
+
>>> y_pred = [2.5, 5, 4, 8]
|
| 506 |
+
>>> mean_squared_log_error(y_true, y_pred)
|
| 507 |
+
0.039...
|
| 508 |
+
>>> mean_squared_log_error(y_true, y_pred, squared=False)
|
| 509 |
+
0.199...
|
| 510 |
+
>>> y_true = [[0.5, 1], [1, 2], [7, 6]]
|
| 511 |
+
>>> y_pred = [[0.5, 2], [1, 2.5], [8, 8]]
|
| 512 |
+
>>> mean_squared_log_error(y_true, y_pred)
|
| 513 |
+
0.044...
|
| 514 |
+
>>> mean_squared_log_error(y_true, y_pred, multioutput='raw_values')
|
| 515 |
+
array([0.00462428, 0.08377444])
|
| 516 |
+
>>> mean_squared_log_error(y_true, y_pred, multioutput=[0.3, 0.7])
|
| 517 |
+
0.060...
|
| 518 |
+
"""
|
| 519 |
+
y_type, y_true, y_pred, multioutput = _check_reg_targets(
|
| 520 |
+
y_true, y_pred, multioutput
|
| 521 |
+
)
|
| 522 |
+
check_consistent_length(y_true, y_pred, sample_weight)
|
| 523 |
+
|
| 524 |
+
if (y_true < 0).any() or (y_pred < 0).any():
|
| 525 |
+
raise ValueError(
|
| 526 |
+
"Mean Squared Logarithmic Error cannot be used when "
|
| 527 |
+
"targets contain negative values."
|
| 528 |
+
)
|
| 529 |
+
|
| 530 |
+
return mean_squared_error(
|
| 531 |
+
np.log1p(y_true),
|
| 532 |
+
np.log1p(y_pred),
|
| 533 |
+
sample_weight=sample_weight,
|
| 534 |
+
multioutput=multioutput,
|
| 535 |
+
squared=squared,
|
| 536 |
+
)
|
| 537 |
+
|
| 538 |
+
|
| 539 |
+
def median_absolute_error(
|
| 540 |
+
y_true, y_pred, *, multioutput="uniform_average", sample_weight=None
|
| 541 |
+
):
|
| 542 |
+
"""Median absolute error regression loss.
|
| 543 |
+
|
| 544 |
+
Median absolute error output is non-negative floating point. The best value
|
| 545 |
+
is 0.0. Read more in the :ref:`User Guide <median_absolute_error>`.
|
| 546 |
+
|
| 547 |
+
Parameters
|
| 548 |
+
----------
|
| 549 |
+
y_true : array-like of shape = (n_samples) or (n_samples, n_outputs)
|
| 550 |
+
Ground truth (correct) target values.
|
| 551 |
+
|
| 552 |
+
y_pred : array-like of shape = (n_samples) or (n_samples, n_outputs)
|
| 553 |
+
Estimated target values.
|
| 554 |
+
|
| 555 |
+
multioutput : {'raw_values', 'uniform_average'} or array-like of shape \
|
| 556 |
+
(n_outputs,), default='uniform_average'
|
| 557 |
+
Defines aggregating of multiple output values. Array-like value defines
|
| 558 |
+
weights used to average errors.
|
| 559 |
+
|
| 560 |
+
'raw_values' :
|
| 561 |
+
Returns a full set of errors in case of multioutput input.
|
| 562 |
+
|
| 563 |
+
'uniform_average' :
|
| 564 |
+
Errors of all outputs are averaged with uniform weight.
|
| 565 |
+
|
| 566 |
+
sample_weight : array-like of shape (n_samples,), default=None
|
| 567 |
+
Sample weights.
|
| 568 |
+
|
| 569 |
+
.. versionadded:: 0.24
|
| 570 |
+
|
| 571 |
+
Returns
|
| 572 |
+
-------
|
| 573 |
+
loss : float or ndarray of floats
|
| 574 |
+
If multioutput is 'raw_values', then mean absolute error is returned
|
| 575 |
+
for each output separately.
|
| 576 |
+
If multioutput is 'uniform_average' or an ndarray of weights, then the
|
| 577 |
+
weighted average of all output errors is returned.
|
| 578 |
+
|
| 579 |
+
Examples
|
| 580 |
+
--------
|
| 581 |
+
>>> from sklearn.metrics import median_absolute_error
|
| 582 |
+
>>> y_true = [3, -0.5, 2, 7]
|
| 583 |
+
>>> y_pred = [2.5, 0.0, 2, 8]
|
| 584 |
+
>>> median_absolute_error(y_true, y_pred)
|
| 585 |
+
0.5
|
| 586 |
+
>>> y_true = [[0.5, 1], [-1, 1], [7, -6]]
|
| 587 |
+
>>> y_pred = [[0, 2], [-1, 2], [8, -5]]
|
| 588 |
+
>>> median_absolute_error(y_true, y_pred)
|
| 589 |
+
0.75
|
| 590 |
+
>>> median_absolute_error(y_true, y_pred, multioutput='raw_values')
|
| 591 |
+
array([0.5, 1. ])
|
| 592 |
+
>>> median_absolute_error(y_true, y_pred, multioutput=[0.3, 0.7])
|
| 593 |
+
0.85
|
| 594 |
+
"""
|
| 595 |
+
y_type, y_true, y_pred, multioutput = _check_reg_targets(
|
| 596 |
+
y_true, y_pred, multioutput
|
| 597 |
+
)
|
| 598 |
+
if sample_weight is None:
|
| 599 |
+
output_errors = np.median(np.abs(y_pred - y_true), axis=0)
|
| 600 |
+
else:
|
| 601 |
+
sample_weight = _check_sample_weight(sample_weight, y_pred)
|
| 602 |
+
output_errors = _weighted_percentile(
|
| 603 |
+
np.abs(y_pred - y_true), sample_weight=sample_weight
|
| 604 |
+
)
|
| 605 |
+
if isinstance(multioutput, str):
|
| 606 |
+
if multioutput == "raw_values":
|
| 607 |
+
return output_errors
|
| 608 |
+
elif multioutput == "uniform_average":
|
| 609 |
+
# pass None as weights to np.average: uniform mean
|
| 610 |
+
multioutput = None
|
| 611 |
+
|
| 612 |
+
return np.average(output_errors, weights=multioutput)
|
| 613 |
+
|
| 614 |
+
|
| 615 |
+
def _assemble_r2_explained_variance(
|
| 616 |
+
numerator, denominator, n_outputs, multioutput, force_finite
|
| 617 |
+
):
|
| 618 |
+
"""Common part used by explained variance score and :math:`R^2` score."""
|
| 619 |
+
|
| 620 |
+
nonzero_denominator = denominator != 0
|
| 621 |
+
|
| 622 |
+
if not force_finite:
|
| 623 |
+
# Standard formula, that may lead to NaN or -Inf
|
| 624 |
+
output_scores = 1 - (numerator / denominator)
|
| 625 |
+
else:
|
| 626 |
+
nonzero_numerator = numerator != 0
|
| 627 |
+
# Default = Zero Numerator = perfect predictions. Set to 1.0
|
| 628 |
+
# (note: even if denominator is zero, thus avoiding NaN scores)
|
| 629 |
+
output_scores = np.ones([n_outputs])
|
| 630 |
+
# Non-zero Numerator and Non-zero Denominator: use the formula
|
| 631 |
+
valid_score = nonzero_denominator & nonzero_numerator
|
| 632 |
+
output_scores[valid_score] = 1 - (
|
| 633 |
+
numerator[valid_score] / denominator[valid_score]
|
| 634 |
+
)
|
| 635 |
+
# Non-zero Numerator and Zero Denominator:
|
| 636 |
+
# arbitrary set to 0.0 to avoid -inf scores
|
| 637 |
+
output_scores[nonzero_numerator & ~nonzero_denominator] = 0.0
|
| 638 |
+
|
| 639 |
+
if isinstance(multioutput, str):
|
| 640 |
+
if multioutput == "raw_values":
|
| 641 |
+
# return scores individually
|
| 642 |
+
return output_scores
|
| 643 |
+
elif multioutput == "uniform_average":
|
| 644 |
+
# Passing None as weights to np.average results is uniform mean
|
| 645 |
+
avg_weights = None
|
| 646 |
+
elif multioutput == "variance_weighted":
|
| 647 |
+
avg_weights = denominator
|
| 648 |
+
if not np.any(nonzero_denominator):
|
| 649 |
+
# All weights are zero, np.average would raise a ZeroDiv error.
|
| 650 |
+
# This only happens when all y are constant (or 1-element long)
|
| 651 |
+
# Since weights are all equal, fall back to uniform weights.
|
| 652 |
+
avg_weights = None
|
| 653 |
+
else:
|
| 654 |
+
avg_weights = multioutput
|
| 655 |
+
|
| 656 |
+
return np.average(output_scores, weights=avg_weights)
|
| 657 |
+
|
| 658 |
+
|
| 659 |
+
def explained_variance_score(
|
| 660 |
+
y_true,
|
| 661 |
+
y_pred,
|
| 662 |
+
*,
|
| 663 |
+
sample_weight=None,
|
| 664 |
+
multioutput="uniform_average",
|
| 665 |
+
force_finite=True,
|
| 666 |
+
):
|
| 667 |
+
"""Explained variance regression score function.
|
| 668 |
+
|
| 669 |
+
Best possible score is 1.0, lower values are worse.
|
| 670 |
+
|
| 671 |
+
In the particular case when ``y_true`` is constant, the explained variance
|
| 672 |
+
score is not finite: it is either ``NaN`` (perfect predictions) or
|
| 673 |
+
``-Inf`` (imperfect predictions). To prevent such non-finite numbers to
|
| 674 |
+
pollute higher-level experiments such as a grid search cross-validation,
|
| 675 |
+
by default these cases are replaced with 1.0 (perfect predictions) or 0.0
|
| 676 |
+
(imperfect predictions) respectively. If ``force_finite``
|
| 677 |
+
is set to ``False``, this score falls back on the original :math:`R^2`
|
| 678 |
+
definition.
|
| 679 |
+
|
| 680 |
+
.. note::
|
| 681 |
+
The Explained Variance score is similar to the
|
| 682 |
+
:func:`R^2 score <r2_score>`, with the notable difference that it
|
| 683 |
+
does not account for systematic offsets in the prediction. Most often
|
| 684 |
+
the :func:`R^2 score <r2_score>` should be preferred.
|
| 685 |
+
|
| 686 |
+
Read more in the :ref:`User Guide <explained_variance_score>`.
|
| 687 |
+
|
| 688 |
+
Parameters
|
| 689 |
+
----------
|
| 690 |
+
y_true : array-like of shape (n_samples,) or (n_samples, n_outputs)
|
| 691 |
+
Ground truth (correct) target values.
|
| 692 |
+
|
| 693 |
+
y_pred : array-like of shape (n_samples,) or (n_samples, n_outputs)
|
| 694 |
+
Estimated target values.
|
| 695 |
+
|
| 696 |
+
sample_weight : array-like of shape (n_samples,), default=None
|
| 697 |
+
Sample weights.
|
| 698 |
+
|
| 699 |
+
multioutput : {'raw_values', 'uniform_average', 'variance_weighted'} or \
|
| 700 |
+
array-like of shape (n_outputs,), default='uniform_average'
|
| 701 |
+
Defines aggregating of multiple output scores.
|
| 702 |
+
Array-like value defines weights used to average scores.
|
| 703 |
+
|
| 704 |
+
'raw_values' :
|
| 705 |
+
Returns a full set of scores in case of multioutput input.
|
| 706 |
+
|
| 707 |
+
'uniform_average' :
|
| 708 |
+
Scores of all outputs are averaged with uniform weight.
|
| 709 |
+
|
| 710 |
+
'variance_weighted' :
|
| 711 |
+
Scores of all outputs are averaged, weighted by the variances
|
| 712 |
+
of each individual output.
|
| 713 |
+
|
| 714 |
+
force_finite : bool, default=True
|
| 715 |
+
Flag indicating if ``NaN`` and ``-Inf`` scores resulting from constant
|
| 716 |
+
data should be replaced with real numbers (``1.0`` if prediction is
|
| 717 |
+
perfect, ``0.0`` otherwise). Default is ``True``, a convenient setting
|
| 718 |
+
for hyperparameters' search procedures (e.g. grid search
|
| 719 |
+
cross-validation).
|
| 720 |
+
|
| 721 |
+
.. versionadded:: 1.1
|
| 722 |
+
|
| 723 |
+
Returns
|
| 724 |
+
-------
|
| 725 |
+
score : float or ndarray of floats
|
| 726 |
+
The explained variance or ndarray if 'multioutput' is 'raw_values'.
|
| 727 |
+
|
| 728 |
+
See Also
|
| 729 |
+
--------
|
| 730 |
+
r2_score :
|
| 731 |
+
Similar metric, but accounting for systematic offsets in
|
| 732 |
+
prediction.
|
| 733 |
+
|
| 734 |
+
Notes
|
| 735 |
+
-----
|
| 736 |
+
This is not a symmetric function.
|
| 737 |
+
|
| 738 |
+
Examples
|
| 739 |
+
--------
|
| 740 |
+
>>> from sklearn.metrics import explained_variance_score
|
| 741 |
+
>>> y_true = [3, -0.5, 2, 7]
|
| 742 |
+
>>> y_pred = [2.5, 0.0, 2, 8]
|
| 743 |
+
>>> explained_variance_score(y_true, y_pred)
|
| 744 |
+
0.957...
|
| 745 |
+
>>> y_true = [[0.5, 1], [-1, 1], [7, -6]]
|
| 746 |
+
>>> y_pred = [[0, 2], [-1, 2], [8, -5]]
|
| 747 |
+
>>> explained_variance_score(y_true, y_pred, multioutput='uniform_average')
|
| 748 |
+
0.983...
|
| 749 |
+
>>> y_true = [-2, -2, -2]
|
| 750 |
+
>>> y_pred = [-2, -2, -2]
|
| 751 |
+
>>> explained_variance_score(y_true, y_pred)
|
| 752 |
+
1.0
|
| 753 |
+
>>> explained_variance_score(y_true, y_pred, force_finite=False)
|
| 754 |
+
nan
|
| 755 |
+
>>> y_true = [-2, -2, -2]
|
| 756 |
+
>>> y_pred = [-2, -2, -2 + 1e-8]
|
| 757 |
+
>>> explained_variance_score(y_true, y_pred)
|
| 758 |
+
0.0
|
| 759 |
+
>>> explained_variance_score(y_true, y_pred, force_finite=False)
|
| 760 |
+
-inf
|
| 761 |
+
"""
|
| 762 |
+
y_type, y_true, y_pred, multioutput = _check_reg_targets(
|
| 763 |
+
y_true, y_pred, multioutput
|
| 764 |
+
)
|
| 765 |
+
check_consistent_length(y_true, y_pred, sample_weight)
|
| 766 |
+
|
| 767 |
+
y_diff_avg = np.average(y_true - y_pred, weights=sample_weight, axis=0)
|
| 768 |
+
numerator = np.average(
|
| 769 |
+
(y_true - y_pred - y_diff_avg) ** 2, weights=sample_weight, axis=0
|
| 770 |
+
)
|
| 771 |
+
|
| 772 |
+
y_true_avg = np.average(y_true, weights=sample_weight, axis=0)
|
| 773 |
+
denominator = np.average((y_true - y_true_avg) ** 2, weights=sample_weight, axis=0)
|
| 774 |
+
|
| 775 |
+
return _assemble_r2_explained_variance(
|
| 776 |
+
numerator=numerator,
|
| 777 |
+
denominator=denominator,
|
| 778 |
+
n_outputs=y_true.shape[1],
|
| 779 |
+
multioutput=multioutput,
|
| 780 |
+
force_finite=force_finite,
|
| 781 |
+
)
|
| 782 |
+
|
| 783 |
+
|
| 784 |
+
def r2_score(
|
| 785 |
+
y_true,
|
| 786 |
+
y_pred,
|
| 787 |
+
*,
|
| 788 |
+
sample_weight=None,
|
| 789 |
+
multioutput="uniform_average",
|
| 790 |
+
force_finite=True,
|
| 791 |
+
):
|
| 792 |
+
""":math:`R^2` (coefficient of determination) regression score function.
|
| 793 |
+
|
| 794 |
+
Best possible score is 1.0 and it can be negative (because the
|
| 795 |
+
model can be arbitrarily worse). In the general case when the true y is
|
| 796 |
+
non-constant, a constant model that always predicts the average y
|
| 797 |
+
disregarding the input features would get a :math:`R^2` score of 0.0.
|
| 798 |
+
|
| 799 |
+
In the particular case when ``y_true`` is constant, the :math:`R^2` score
|
| 800 |
+
is not finite: it is either ``NaN`` (perfect predictions) or ``-Inf``
|
| 801 |
+
(imperfect predictions). To prevent such non-finite numbers to pollute
|
| 802 |
+
higher-level experiments such as a grid search cross-validation, by default
|
| 803 |
+
these cases are replaced with 1.0 (perfect predictions) or 0.0 (imperfect
|
| 804 |
+
predictions) respectively. You can set ``force_finite`` to ``False`` to
|
| 805 |
+
prevent this fix from happening.
|
| 806 |
+
|
| 807 |
+
Note: when the prediction residuals have zero mean, the :math:`R^2` score
|
| 808 |
+
is identical to the
|
| 809 |
+
:func:`Explained Variance score <explained_variance_score>`.
|
| 810 |
+
|
| 811 |
+
Read more in the :ref:`User Guide <r2_score>`.
|
| 812 |
+
|
| 813 |
+
Parameters
|
| 814 |
+
----------
|
| 815 |
+
y_true : array-like of shape (n_samples,) or (n_samples, n_outputs)
|
| 816 |
+
Ground truth (correct) target values.
|
| 817 |
+
|
| 818 |
+
y_pred : array-like of shape (n_samples,) or (n_samples, n_outputs)
|
| 819 |
+
Estimated target values.
|
| 820 |
+
|
| 821 |
+
sample_weight : array-like of shape (n_samples,), default=None
|
| 822 |
+
Sample weights.
|
| 823 |
+
|
| 824 |
+
multioutput : {'raw_values', 'uniform_average', 'variance_weighted'}, \
|
| 825 |
+
array-like of shape (n_outputs,) or None, default='uniform_average'
|
| 826 |
+
|
| 827 |
+
Defines aggregating of multiple output scores.
|
| 828 |
+
Array-like value defines weights used to average scores.
|
| 829 |
+
Default is "uniform_average".
|
| 830 |
+
|
| 831 |
+
'raw_values' :
|
| 832 |
+
Returns a full set of scores in case of multioutput input.
|
| 833 |
+
|
| 834 |
+
'uniform_average' :
|
| 835 |
+
Scores of all outputs are averaged with uniform weight.
|
| 836 |
+
|
| 837 |
+
'variance_weighted' :
|
| 838 |
+
Scores of all outputs are averaged, weighted by the variances
|
| 839 |
+
of each individual output.
|
| 840 |
+
|
| 841 |
+
.. versionchanged:: 0.19
|
| 842 |
+
Default value of multioutput is 'uniform_average'.
|
| 843 |
+
|
| 844 |
+
force_finite : bool, default=True
|
| 845 |
+
Flag indicating if ``NaN`` and ``-Inf`` scores resulting from constant
|
| 846 |
+
data should be replaced with real numbers (``1.0`` if prediction is
|
| 847 |
+
perfect, ``0.0`` otherwise). Default is ``True``, a convenient setting
|
| 848 |
+
for hyperparameters' search procedures (e.g. grid search
|
| 849 |
+
cross-validation).
|
| 850 |
+
|
| 851 |
+
.. versionadded:: 1.1
|
| 852 |
+
|
| 853 |
+
Returns
|
| 854 |
+
-------
|
| 855 |
+
z : float or ndarray of floats
|
| 856 |
+
The :math:`R^2` score or ndarray of scores if 'multioutput' is
|
| 857 |
+
'raw_values'.
|
| 858 |
+
|
| 859 |
+
Notes
|
| 860 |
+
-----
|
| 861 |
+
This is not a symmetric function.
|
| 862 |
+
|
| 863 |
+
Unlike most other scores, :math:`R^2` score may be negative (it need not
|
| 864 |
+
actually be the square of a quantity R).
|
| 865 |
+
|
| 866 |
+
This metric is not well-defined for single samples and will return a NaN
|
| 867 |
+
value if n_samples is less than two.
|
| 868 |
+
|
| 869 |
+
References
|
| 870 |
+
----------
|
| 871 |
+
.. [1] `Wikipedia entry on the Coefficient of determination
|
| 872 |
+
<https://en.wikipedia.org/wiki/Coefficient_of_determination>`_
|
| 873 |
+
|
| 874 |
+
Examples
|
| 875 |
+
--------
|
| 876 |
+
>>> from sklearn.metrics import r2_score
|
| 877 |
+
>>> y_true = [3, -0.5, 2, 7]
|
| 878 |
+
>>> y_pred = [2.5, 0.0, 2, 8]
|
| 879 |
+
>>> r2_score(y_true, y_pred)
|
| 880 |
+
0.948...
|
| 881 |
+
>>> y_true = [[0.5, 1], [-1, 1], [7, -6]]
|
| 882 |
+
>>> y_pred = [[0, 2], [-1, 2], [8, -5]]
|
| 883 |
+
>>> r2_score(y_true, y_pred,
|
| 884 |
+
... multioutput='variance_weighted')
|
| 885 |
+
0.938...
|
| 886 |
+
>>> y_true = [1, 2, 3]
|
| 887 |
+
>>> y_pred = [1, 2, 3]
|
| 888 |
+
>>> r2_score(y_true, y_pred)
|
| 889 |
+
1.0
|
| 890 |
+
>>> y_true = [1, 2, 3]
|
| 891 |
+
>>> y_pred = [2, 2, 2]
|
| 892 |
+
>>> r2_score(y_true, y_pred)
|
| 893 |
+
0.0
|
| 894 |
+
>>> y_true = [1, 2, 3]
|
| 895 |
+
>>> y_pred = [3, 2, 1]
|
| 896 |
+
>>> r2_score(y_true, y_pred)
|
| 897 |
+
-3.0
|
| 898 |
+
>>> y_true = [-2, -2, -2]
|
| 899 |
+
>>> y_pred = [-2, -2, -2]
|
| 900 |
+
>>> r2_score(y_true, y_pred)
|
| 901 |
+
1.0
|
| 902 |
+
>>> r2_score(y_true, y_pred, force_finite=False)
|
| 903 |
+
nan
|
| 904 |
+
>>> y_true = [-2, -2, -2]
|
| 905 |
+
>>> y_pred = [-2, -2, -2 + 1e-8]
|
| 906 |
+
>>> r2_score(y_true, y_pred)
|
| 907 |
+
0.0
|
| 908 |
+
>>> r2_score(y_true, y_pred, force_finite=False)
|
| 909 |
+
-inf
|
| 910 |
+
"""
|
| 911 |
+
y_type, y_true, y_pred, multioutput = _check_reg_targets(
|
| 912 |
+
y_true, y_pred, multioutput
|
| 913 |
+
)
|
| 914 |
+
check_consistent_length(y_true, y_pred, sample_weight)
|
| 915 |
+
|
| 916 |
+
if _num_samples(y_pred) < 2:
|
| 917 |
+
msg = "R^2 score is not well-defined with less than two samples."
|
| 918 |
+
warnings.warn(msg, UndefinedMetricWarning)
|
| 919 |
+
return float("nan")
|
| 920 |
+
|
| 921 |
+
if sample_weight is not None:
|
| 922 |
+
sample_weight = column_or_1d(sample_weight)
|
| 923 |
+
weight = sample_weight[:, np.newaxis]
|
| 924 |
+
else:
|
| 925 |
+
weight = 1.0
|
| 926 |
+
|
| 927 |
+
numerator = (weight * (y_true - y_pred) ** 2).sum(axis=0, dtype=np.float64)
|
| 928 |
+
denominator = (
|
| 929 |
+
weight * (y_true - np.average(y_true, axis=0, weights=sample_weight)) ** 2
|
| 930 |
+
).sum(axis=0, dtype=np.float64)
|
| 931 |
+
|
| 932 |
+
return _assemble_r2_explained_variance(
|
| 933 |
+
numerator=numerator,
|
| 934 |
+
denominator=denominator,
|
| 935 |
+
n_outputs=y_true.shape[1],
|
| 936 |
+
multioutput=multioutput,
|
| 937 |
+
force_finite=force_finite,
|
| 938 |
+
)
|
| 939 |
+
|
| 940 |
+
|
| 941 |
+
def max_error(y_true, y_pred):
|
| 942 |
+
"""
|
| 943 |
+
The max_error metric calculates the maximum residual error.
|
| 944 |
+
|
| 945 |
+
Read more in the :ref:`User Guide <max_error>`.
|
| 946 |
+
|
| 947 |
+
Parameters
|
| 948 |
+
----------
|
| 949 |
+
y_true : array-like of shape (n_samples,)
|
| 950 |
+
Ground truth (correct) target values.
|
| 951 |
+
|
| 952 |
+
y_pred : array-like of shape (n_samples,)
|
| 953 |
+
Estimated target values.
|
| 954 |
+
|
| 955 |
+
Returns
|
| 956 |
+
-------
|
| 957 |
+
max_error : float
|
| 958 |
+
A positive floating point value (the best value is 0.0).
|
| 959 |
+
|
| 960 |
+
Examples
|
| 961 |
+
--------
|
| 962 |
+
>>> from sklearn.metrics import max_error
|
| 963 |
+
>>> y_true = [3, 2, 7, 1]
|
| 964 |
+
>>> y_pred = [4, 2, 7, 1]
|
| 965 |
+
>>> max_error(y_true, y_pred)
|
| 966 |
+
1
|
| 967 |
+
"""
|
| 968 |
+
y_type, y_true, y_pred, _ = _check_reg_targets(y_true, y_pred, None)
|
| 969 |
+
if y_type == "continuous-multioutput":
|
| 970 |
+
raise ValueError("Multioutput not supported in max_error")
|
| 971 |
+
return np.max(np.abs(y_true - y_pred))
|
| 972 |
+
|
| 973 |
+
|
| 974 |
+
def _mean_tweedie_deviance(y_true, y_pred, sample_weight, power):
|
| 975 |
+
"""Mean Tweedie deviance regression loss."""
|
| 976 |
+
p = power
|
| 977 |
+
if p < 0:
|
| 978 |
+
# 'Extreme stable', y any real number, y_pred > 0
|
| 979 |
+
dev = 2 * (
|
| 980 |
+
np.power(np.maximum(y_true, 0), 2 - p) / ((1 - p) * (2 - p))
|
| 981 |
+
- y_true * np.power(y_pred, 1 - p) / (1 - p)
|
| 982 |
+
+ np.power(y_pred, 2 - p) / (2 - p)
|
| 983 |
+
)
|
| 984 |
+
elif p == 0:
|
| 985 |
+
# Normal distribution, y and y_pred any real number
|
| 986 |
+
dev = (y_true - y_pred) ** 2
|
| 987 |
+
elif p == 1:
|
| 988 |
+
# Poisson distribution
|
| 989 |
+
dev = 2 * (xlogy(y_true, y_true / y_pred) - y_true + y_pred)
|
| 990 |
+
elif p == 2:
|
| 991 |
+
# Gamma distribution
|
| 992 |
+
dev = 2 * (np.log(y_pred / y_true) + y_true / y_pred - 1)
|
| 993 |
+
else:
|
| 994 |
+
dev = 2 * (
|
| 995 |
+
np.power(y_true, 2 - p) / ((1 - p) * (2 - p))
|
| 996 |
+
- y_true * np.power(y_pred, 1 - p) / (1 - p)
|
| 997 |
+
+ np.power(y_pred, 2 - p) / (2 - p)
|
| 998 |
+
)
|
| 999 |
+
|
| 1000 |
+
return np.average(dev, weights=sample_weight)
|
| 1001 |
+
|
| 1002 |
+
|
| 1003 |
+
def mean_tweedie_deviance(y_true, y_pred, *, sample_weight=None, power=0):
|
| 1004 |
+
"""Mean Tweedie deviance regression loss.
|
| 1005 |
+
|
| 1006 |
+
Read more in the :ref:`User Guide <mean_tweedie_deviance>`.
|
| 1007 |
+
|
| 1008 |
+
Parameters
|
| 1009 |
+
----------
|
| 1010 |
+
y_true : array-like of shape (n_samples,)
|
| 1011 |
+
Ground truth (correct) target values.
|
| 1012 |
+
|
| 1013 |
+
y_pred : array-like of shape (n_samples,)
|
| 1014 |
+
Estimated target values.
|
| 1015 |
+
|
| 1016 |
+
sample_weight : array-like of shape (n_samples,), default=None
|
| 1017 |
+
Sample weights.
|
| 1018 |
+
|
| 1019 |
+
power : float, default=0
|
| 1020 |
+
Tweedie power parameter. Either power <= 0 or power >= 1.
|
| 1021 |
+
|
| 1022 |
+
The higher `p` the less weight is given to extreme
|
| 1023 |
+
deviations between true and predicted targets.
|
| 1024 |
+
|
| 1025 |
+
- power < 0: Extreme stable distribution. Requires: y_pred > 0.
|
| 1026 |
+
- power = 0 : Normal distribution, output corresponds to
|
| 1027 |
+
mean_squared_error. y_true and y_pred can be any real numbers.
|
| 1028 |
+
- power = 1 : Poisson distribution. Requires: y_true >= 0 and
|
| 1029 |
+
y_pred > 0.
|
| 1030 |
+
- 1 < p < 2 : Compound Poisson distribution. Requires: y_true >= 0
|
| 1031 |
+
and y_pred > 0.
|
| 1032 |
+
- power = 2 : Gamma distribution. Requires: y_true > 0 and y_pred > 0.
|
| 1033 |
+
- power = 3 : Inverse Gaussian distribution. Requires: y_true > 0
|
| 1034 |
+
and y_pred > 0.
|
| 1035 |
+
- otherwise : Positive stable distribution. Requires: y_true > 0
|
| 1036 |
+
and y_pred > 0.
|
| 1037 |
+
|
| 1038 |
+
Returns
|
| 1039 |
+
-------
|
| 1040 |
+
loss : float
|
| 1041 |
+
A non-negative floating point value (the best value is 0.0).
|
| 1042 |
+
|
| 1043 |
+
Examples
|
| 1044 |
+
--------
|
| 1045 |
+
>>> from sklearn.metrics import mean_tweedie_deviance
|
| 1046 |
+
>>> y_true = [2, 0, 1, 4]
|
| 1047 |
+
>>> y_pred = [0.5, 0.5, 2., 2.]
|
| 1048 |
+
>>> mean_tweedie_deviance(y_true, y_pred, power=1)
|
| 1049 |
+
1.4260...
|
| 1050 |
+
"""
|
| 1051 |
+
y_type, y_true, y_pred, _ = _check_reg_targets(
|
| 1052 |
+
y_true, y_pred, None, dtype=[np.float64, np.float32]
|
| 1053 |
+
)
|
| 1054 |
+
if y_type == "continuous-multioutput":
|
| 1055 |
+
raise ValueError("Multioutput not supported in mean_tweedie_deviance")
|
| 1056 |
+
check_consistent_length(y_true, y_pred, sample_weight)
|
| 1057 |
+
|
| 1058 |
+
if sample_weight is not None:
|
| 1059 |
+
sample_weight = column_or_1d(sample_weight)
|
| 1060 |
+
sample_weight = sample_weight[:, np.newaxis]
|
| 1061 |
+
|
| 1062 |
+
p = check_scalar(
|
| 1063 |
+
power,
|
| 1064 |
+
name="power",
|
| 1065 |
+
target_type=numbers.Real,
|
| 1066 |
+
)
|
| 1067 |
+
|
| 1068 |
+
message = f"Mean Tweedie deviance error with power={p} can only be used on "
|
| 1069 |
+
if p < 0:
|
| 1070 |
+
# 'Extreme stable', y any real number, y_pred > 0
|
| 1071 |
+
if (y_pred <= 0).any():
|
| 1072 |
+
raise ValueError(message + "strictly positive y_pred.")
|
| 1073 |
+
elif p == 0:
|
| 1074 |
+
# Normal, y and y_pred can be any real number
|
| 1075 |
+
pass
|
| 1076 |
+
elif 0 < p < 1:
|
| 1077 |
+
raise ValueError("Tweedie deviance is only defined for power<=0 and power>=1.")
|
| 1078 |
+
elif 1 <= p < 2:
|
| 1079 |
+
# Poisson and compound Poisson distribution, y >= 0, y_pred > 0
|
| 1080 |
+
if (y_true < 0).any() or (y_pred <= 0).any():
|
| 1081 |
+
raise ValueError(message + "non-negative y and strictly positive y_pred.")
|
| 1082 |
+
elif p >= 2:
|
| 1083 |
+
# Gamma and Extreme stable distribution, y and y_pred > 0
|
| 1084 |
+
if (y_true <= 0).any() or (y_pred <= 0).any():
|
| 1085 |
+
raise ValueError(message + "strictly positive y and y_pred.")
|
| 1086 |
+
else: # pragma: nocover
|
| 1087 |
+
# Unreachable statement
|
| 1088 |
+
raise ValueError
|
| 1089 |
+
|
| 1090 |
+
return _mean_tweedie_deviance(
|
| 1091 |
+
y_true, y_pred, sample_weight=sample_weight, power=power
|
| 1092 |
+
)
|
| 1093 |
+
|
| 1094 |
+
|
| 1095 |
+
def mean_poisson_deviance(y_true, y_pred, *, sample_weight=None):
|
| 1096 |
+
"""Mean Poisson deviance regression loss.
|
| 1097 |
+
|
| 1098 |
+
Poisson deviance is equivalent to the Tweedie deviance with
|
| 1099 |
+
the power parameter `power=1`.
|
| 1100 |
+
|
| 1101 |
+
Read more in the :ref:`User Guide <mean_tweedie_deviance>`.
|
| 1102 |
+
|
| 1103 |
+
Parameters
|
| 1104 |
+
----------
|
| 1105 |
+
y_true : array-like of shape (n_samples,)
|
| 1106 |
+
Ground truth (correct) target values. Requires y_true >= 0.
|
| 1107 |
+
|
| 1108 |
+
y_pred : array-like of shape (n_samples,)
|
| 1109 |
+
Estimated target values. Requires y_pred > 0.
|
| 1110 |
+
|
| 1111 |
+
sample_weight : array-like of shape (n_samples,), default=None
|
| 1112 |
+
Sample weights.
|
| 1113 |
+
|
| 1114 |
+
Returns
|
| 1115 |
+
-------
|
| 1116 |
+
loss : float
|
| 1117 |
+
A non-negative floating point value (the best value is 0.0).
|
| 1118 |
+
|
| 1119 |
+
Examples
|
| 1120 |
+
--------
|
| 1121 |
+
>>> from sklearn.metrics import mean_poisson_deviance
|
| 1122 |
+
>>> y_true = [2, 0, 1, 4]
|
| 1123 |
+
>>> y_pred = [0.5, 0.5, 2., 2.]
|
| 1124 |
+
>>> mean_poisson_deviance(y_true, y_pred)
|
| 1125 |
+
1.4260...
|
| 1126 |
+
"""
|
| 1127 |
+
return mean_tweedie_deviance(y_true, y_pred, sample_weight=sample_weight, power=1)
|
| 1128 |
+
|
| 1129 |
+
|
| 1130 |
+
def mean_gamma_deviance(y_true, y_pred, *, sample_weight=None):
|
| 1131 |
+
"""Mean Gamma deviance regression loss.
|
| 1132 |
+
|
| 1133 |
+
Gamma deviance is equivalent to the Tweedie deviance with
|
| 1134 |
+
the power parameter `power=2`. It is invariant to scaling of
|
| 1135 |
+
the target variable, and measures relative errors.
|
| 1136 |
+
|
| 1137 |
+
Read more in the :ref:`User Guide <mean_tweedie_deviance>`.
|
| 1138 |
+
|
| 1139 |
+
Parameters
|
| 1140 |
+
----------
|
| 1141 |
+
y_true : array-like of shape (n_samples,)
|
| 1142 |
+
Ground truth (correct) target values. Requires y_true > 0.
|
| 1143 |
+
|
| 1144 |
+
y_pred : array-like of shape (n_samples,)
|
| 1145 |
+
Estimated target values. Requires y_pred > 0.
|
| 1146 |
+
|
| 1147 |
+
sample_weight : array-like of shape (n_samples,), default=None
|
| 1148 |
+
Sample weights.
|
| 1149 |
+
|
| 1150 |
+
Returns
|
| 1151 |
+
-------
|
| 1152 |
+
loss : float
|
| 1153 |
+
A non-negative floating point value (the best value is 0.0).
|
| 1154 |
+
|
| 1155 |
+
Examples
|
| 1156 |
+
--------
|
| 1157 |
+
>>> from sklearn.metrics import mean_gamma_deviance
|
| 1158 |
+
>>> y_true = [2, 0.5, 1, 4]
|
| 1159 |
+
>>> y_pred = [0.5, 0.5, 2., 2.]
|
| 1160 |
+
>>> mean_gamma_deviance(y_true, y_pred)
|
| 1161 |
+
1.0568...
|
| 1162 |
+
"""
|
| 1163 |
+
return mean_tweedie_deviance(y_true, y_pred, sample_weight=sample_weight, power=2)
|
| 1164 |
+
|
| 1165 |
+
|
| 1166 |
+
def d2_tweedie_score(y_true, y_pred, *, sample_weight=None, power=0):
|
| 1167 |
+
"""D^2 regression score function, fraction of Tweedie deviance explained.
|
| 1168 |
+
|
| 1169 |
+
Best possible score is 1.0 and it can be negative (because the model can be
|
| 1170 |
+
arbitrarily worse). A model that always uses the empirical mean of `y_true` as
|
| 1171 |
+
constant prediction, disregarding the input features, gets a D^2 score of 0.0.
|
| 1172 |
+
|
| 1173 |
+
Read more in the :ref:`User Guide <d2_score>`.
|
| 1174 |
+
|
| 1175 |
+
.. versionadded:: 1.0
|
| 1176 |
+
|
| 1177 |
+
Parameters
|
| 1178 |
+
----------
|
| 1179 |
+
y_true : array-like of shape (n_samples,)
|
| 1180 |
+
Ground truth (correct) target values.
|
| 1181 |
+
|
| 1182 |
+
y_pred : array-like of shape (n_samples,)
|
| 1183 |
+
Estimated target values.
|
| 1184 |
+
|
| 1185 |
+
sample_weight : array-like of shape (n_samples,), optional
|
| 1186 |
+
Sample weights.
|
| 1187 |
+
|
| 1188 |
+
power : float, default=0
|
| 1189 |
+
Tweedie power parameter. Either power <= 0 or power >= 1.
|
| 1190 |
+
|
| 1191 |
+
The higher `p` the less weight is given to extreme
|
| 1192 |
+
deviations between true and predicted targets.
|
| 1193 |
+
|
| 1194 |
+
- power < 0: Extreme stable distribution. Requires: y_pred > 0.
|
| 1195 |
+
- power = 0 : Normal distribution, output corresponds to r2_score.
|
| 1196 |
+
y_true and y_pred can be any real numbers.
|
| 1197 |
+
- power = 1 : Poisson distribution. Requires: y_true >= 0 and
|
| 1198 |
+
y_pred > 0.
|
| 1199 |
+
- 1 < p < 2 : Compound Poisson distribution. Requires: y_true >= 0
|
| 1200 |
+
and y_pred > 0.
|
| 1201 |
+
- power = 2 : Gamma distribution. Requires: y_true > 0 and y_pred > 0.
|
| 1202 |
+
- power = 3 : Inverse Gaussian distribution. Requires: y_true > 0
|
| 1203 |
+
and y_pred > 0.
|
| 1204 |
+
- otherwise : Positive stable distribution. Requires: y_true > 0
|
| 1205 |
+
and y_pred > 0.
|
| 1206 |
+
|
| 1207 |
+
Returns
|
| 1208 |
+
-------
|
| 1209 |
+
z : float or ndarray of floats
|
| 1210 |
+
The D^2 score.
|
| 1211 |
+
|
| 1212 |
+
Notes
|
| 1213 |
+
-----
|
| 1214 |
+
This is not a symmetric function.
|
| 1215 |
+
|
| 1216 |
+
Like R^2, D^2 score may be negative (it need not actually be the square of
|
| 1217 |
+
a quantity D).
|
| 1218 |
+
|
| 1219 |
+
This metric is not well-defined for single samples and will return a NaN
|
| 1220 |
+
value if n_samples is less than two.
|
| 1221 |
+
|
| 1222 |
+
References
|
| 1223 |
+
----------
|
| 1224 |
+
.. [1] Eq. (3.11) of Hastie, Trevor J., Robert Tibshirani and Martin J.
|
| 1225 |
+
Wainwright. "Statistical Learning with Sparsity: The Lasso and
|
| 1226 |
+
Generalizations." (2015). https://hastie.su.domains/StatLearnSparsity/
|
| 1227 |
+
|
| 1228 |
+
Examples
|
| 1229 |
+
--------
|
| 1230 |
+
>>> from sklearn.metrics import d2_tweedie_score
|
| 1231 |
+
>>> y_true = [0.5, 1, 2.5, 7]
|
| 1232 |
+
>>> y_pred = [1, 1, 5, 3.5]
|
| 1233 |
+
>>> d2_tweedie_score(y_true, y_pred)
|
| 1234 |
+
0.285...
|
| 1235 |
+
>>> d2_tweedie_score(y_true, y_pred, power=1)
|
| 1236 |
+
0.487...
|
| 1237 |
+
>>> d2_tweedie_score(y_true, y_pred, power=2)
|
| 1238 |
+
0.630...
|
| 1239 |
+
>>> d2_tweedie_score(y_true, y_true, power=2)
|
| 1240 |
+
1.0
|
| 1241 |
+
"""
|
| 1242 |
+
y_type, y_true, y_pred, _ = _check_reg_targets(
|
| 1243 |
+
y_true, y_pred, None, dtype=[np.float64, np.float32]
|
| 1244 |
+
)
|
| 1245 |
+
if y_type == "continuous-multioutput":
|
| 1246 |
+
raise ValueError("Multioutput not supported in d2_tweedie_score")
|
| 1247 |
+
|
| 1248 |
+
if _num_samples(y_pred) < 2:
|
| 1249 |
+
msg = "D^2 score is not well-defined with less than two samples."
|
| 1250 |
+
warnings.warn(msg, UndefinedMetricWarning)
|
| 1251 |
+
return float("nan")
|
| 1252 |
+
|
| 1253 |
+
y_true, y_pred = np.squeeze(y_true), np.squeeze(y_pred)
|
| 1254 |
+
numerator = mean_tweedie_deviance(
|
| 1255 |
+
y_true, y_pred, sample_weight=sample_weight, power=power
|
| 1256 |
+
)
|
| 1257 |
+
|
| 1258 |
+
y_avg = np.average(y_true, weights=sample_weight)
|
| 1259 |
+
denominator = _mean_tweedie_deviance(
|
| 1260 |
+
y_true, y_avg, sample_weight=sample_weight, power=power
|
| 1261 |
+
)
|
| 1262 |
+
|
| 1263 |
+
return 1 - numerator / denominator
|
| 1264 |
+
|
| 1265 |
+
|
| 1266 |
+
def d2_pinball_score(
|
| 1267 |
+
y_true, y_pred, *, sample_weight=None, alpha=0.5, multioutput="uniform_average"
|
| 1268 |
+
):
|
| 1269 |
+
"""
|
| 1270 |
+
:math:`D^2` regression score function, fraction of pinball loss explained.
|
| 1271 |
+
|
| 1272 |
+
Best possible score is 1.0 and it can be negative (because the model can be
|
| 1273 |
+
arbitrarily worse). A model that always uses the empirical alpha-quantile of
|
| 1274 |
+
`y_true` as constant prediction, disregarding the input features,
|
| 1275 |
+
gets a :math:`D^2` score of 0.0.
|
| 1276 |
+
|
| 1277 |
+
Read more in the :ref:`User Guide <d2_score>`.
|
| 1278 |
+
|
| 1279 |
+
.. versionadded:: 1.1
|
| 1280 |
+
|
| 1281 |
+
Parameters
|
| 1282 |
+
----------
|
| 1283 |
+
y_true : array-like of shape (n_samples,) or (n_samples, n_outputs)
|
| 1284 |
+
Ground truth (correct) target values.
|
| 1285 |
+
|
| 1286 |
+
y_pred : array-like of shape (n_samples,) or (n_samples, n_outputs)
|
| 1287 |
+
Estimated target values.
|
| 1288 |
+
|
| 1289 |
+
sample_weight : array-like of shape (n_samples,), optional
|
| 1290 |
+
Sample weights.
|
| 1291 |
+
|
| 1292 |
+
alpha : float, default=0.5
|
| 1293 |
+
Slope of the pinball deviance. It determines the quantile level alpha
|
| 1294 |
+
for which the pinball deviance and also D2 are optimal.
|
| 1295 |
+
The default `alpha=0.5` is equivalent to `d2_absolute_error_score`.
|
| 1296 |
+
|
| 1297 |
+
multioutput : {'raw_values', 'uniform_average'} or array-like of shape \
|
| 1298 |
+
(n_outputs,), default='uniform_average'
|
| 1299 |
+
Defines aggregating of multiple output values.
|
| 1300 |
+
Array-like value defines weights used to average scores.
|
| 1301 |
+
|
| 1302 |
+
'raw_values' :
|
| 1303 |
+
Returns a full set of errors in case of multioutput input.
|
| 1304 |
+
|
| 1305 |
+
'uniform_average' :
|
| 1306 |
+
Scores of all outputs are averaged with uniform weight.
|
| 1307 |
+
|
| 1308 |
+
Returns
|
| 1309 |
+
-------
|
| 1310 |
+
score : float or ndarray of floats
|
| 1311 |
+
The :math:`D^2` score with a pinball deviance
|
| 1312 |
+
or ndarray of scores if `multioutput='raw_values'`.
|
| 1313 |
+
|
| 1314 |
+
Notes
|
| 1315 |
+
-----
|
| 1316 |
+
Like :math:`R^2`, :math:`D^2` score may be negative
|
| 1317 |
+
(it need not actually be the square of a quantity D).
|
| 1318 |
+
|
| 1319 |
+
This metric is not well-defined for a single point and will return a NaN
|
| 1320 |
+
value if n_samples is less than two.
|
| 1321 |
+
|
| 1322 |
+
References
|
| 1323 |
+
----------
|
| 1324 |
+
.. [1] Eq. (7) of `Koenker, Roger; Machado, José A. F. (1999).
|
| 1325 |
+
"Goodness of Fit and Related Inference Processes for Quantile Regression"
|
| 1326 |
+
<http://dx.doi.org/10.1080/01621459.1999.10473882>`_
|
| 1327 |
+
.. [2] Eq. (3.11) of Hastie, Trevor J., Robert Tibshirani and Martin J.
|
| 1328 |
+
Wainwright. "Statistical Learning with Sparsity: The Lasso and
|
| 1329 |
+
Generalizations." (2015). https://hastie.su.domains/StatLearnSparsity/
|
| 1330 |
+
|
| 1331 |
+
Examples
|
| 1332 |
+
--------
|
| 1333 |
+
>>> from sklearn.metrics import d2_pinball_score
|
| 1334 |
+
>>> y_true = [1, 2, 3]
|
| 1335 |
+
>>> y_pred = [1, 3, 3]
|
| 1336 |
+
>>> d2_pinball_score(y_true, y_pred)
|
| 1337 |
+
0.5
|
| 1338 |
+
>>> d2_pinball_score(y_true, y_pred, alpha=0.9)
|
| 1339 |
+
0.772...
|
| 1340 |
+
>>> d2_pinball_score(y_true, y_pred, alpha=0.1)
|
| 1341 |
+
-1.045...
|
| 1342 |
+
>>> d2_pinball_score(y_true, y_true, alpha=0.1)
|
| 1343 |
+
1.0
|
| 1344 |
+
"""
|
| 1345 |
+
y_type, y_true, y_pred, multioutput = _check_reg_targets(
|
| 1346 |
+
y_true, y_pred, multioutput
|
| 1347 |
+
)
|
| 1348 |
+
check_consistent_length(y_true, y_pred, sample_weight)
|
| 1349 |
+
|
| 1350 |
+
if _num_samples(y_pred) < 2:
|
| 1351 |
+
msg = "D^2 score is not well-defined with less than two samples."
|
| 1352 |
+
warnings.warn(msg, UndefinedMetricWarning)
|
| 1353 |
+
return float("nan")
|
| 1354 |
+
|
| 1355 |
+
numerator = mean_pinball_loss(
|
| 1356 |
+
y_true,
|
| 1357 |
+
y_pred,
|
| 1358 |
+
sample_weight=sample_weight,
|
| 1359 |
+
alpha=alpha,
|
| 1360 |
+
multioutput="raw_values",
|
| 1361 |
+
)
|
| 1362 |
+
|
| 1363 |
+
if sample_weight is None:
|
| 1364 |
+
y_quantile = np.tile(
|
| 1365 |
+
np.percentile(y_true, q=alpha * 100, axis=0), (len(y_true), 1)
|
| 1366 |
+
)
|
| 1367 |
+
else:
|
| 1368 |
+
sample_weight = _check_sample_weight(sample_weight, y_true)
|
| 1369 |
+
y_quantile = np.tile(
|
| 1370 |
+
_weighted_percentile(
|
| 1371 |
+
y_true, sample_weight=sample_weight, percentile=alpha * 100
|
| 1372 |
+
),
|
| 1373 |
+
(len(y_true), 1),
|
| 1374 |
+
)
|
| 1375 |
+
|
| 1376 |
+
denominator = mean_pinball_loss(
|
| 1377 |
+
y_true,
|
| 1378 |
+
y_quantile,
|
| 1379 |
+
sample_weight=sample_weight,
|
| 1380 |
+
alpha=alpha,
|
| 1381 |
+
multioutput="raw_values",
|
| 1382 |
+
)
|
| 1383 |
+
|
| 1384 |
+
nonzero_numerator = numerator != 0
|
| 1385 |
+
nonzero_denominator = denominator != 0
|
| 1386 |
+
valid_score = nonzero_numerator & nonzero_denominator
|
| 1387 |
+
output_scores = np.ones(y_true.shape[1])
|
| 1388 |
+
|
| 1389 |
+
output_scores[valid_score] = 1 - (numerator[valid_score] / denominator[valid_score])
|
| 1390 |
+
output_scores[nonzero_numerator & ~nonzero_denominator] = 0.0
|
| 1391 |
+
|
| 1392 |
+
if isinstance(multioutput, str):
|
| 1393 |
+
if multioutput == "raw_values":
|
| 1394 |
+
# return scores individually
|
| 1395 |
+
return output_scores
|
| 1396 |
+
elif multioutput == "uniform_average":
|
| 1397 |
+
# passing None as weights to np.average results in uniform mean
|
| 1398 |
+
avg_weights = None
|
| 1399 |
+
else:
|
| 1400 |
+
raise ValueError(
|
| 1401 |
+
"multioutput is expected to be 'raw_values' "
|
| 1402 |
+
"or 'uniform_average' but we got %r"
|
| 1403 |
+
" instead." % multioutput
|
| 1404 |
+
)
|
| 1405 |
+
else:
|
| 1406 |
+
avg_weights = multioutput
|
| 1407 |
+
|
| 1408 |
+
return np.average(output_scores, weights=avg_weights)
|
| 1409 |
+
|
| 1410 |
+
|
| 1411 |
+
def d2_absolute_error_score(
|
| 1412 |
+
y_true, y_pred, *, sample_weight=None, multioutput="uniform_average"
|
| 1413 |
+
):
|
| 1414 |
+
"""
|
| 1415 |
+
:math:`D^2` regression score function, \
|
| 1416 |
+
fraction of absolute error explained.
|
| 1417 |
+
|
| 1418 |
+
Best possible score is 1.0 and it can be negative (because the model can be
|
| 1419 |
+
arbitrarily worse). A model that always uses the empirical median of `y_true`
|
| 1420 |
+
as constant prediction, disregarding the input features,
|
| 1421 |
+
gets a :math:`D^2` score of 0.0.
|
| 1422 |
+
|
| 1423 |
+
Read more in the :ref:`User Guide <d2_score>`.
|
| 1424 |
+
|
| 1425 |
+
.. versionadded:: 1.1
|
| 1426 |
+
|
| 1427 |
+
Parameters
|
| 1428 |
+
----------
|
| 1429 |
+
y_true : array-like of shape (n_samples,) or (n_samples, n_outputs)
|
| 1430 |
+
Ground truth (correct) target values.
|
| 1431 |
+
|
| 1432 |
+
y_pred : array-like of shape (n_samples,) or (n_samples, n_outputs)
|
| 1433 |
+
Estimated target values.
|
| 1434 |
+
|
| 1435 |
+
sample_weight : array-like of shape (n_samples,), optional
|
| 1436 |
+
Sample weights.
|
| 1437 |
+
|
| 1438 |
+
multioutput : {'raw_values', 'uniform_average'} or array-like of shape \
|
| 1439 |
+
(n_outputs,), default='uniform_average'
|
| 1440 |
+
Defines aggregating of multiple output values.
|
| 1441 |
+
Array-like value defines weights used to average scores.
|
| 1442 |
+
|
| 1443 |
+
'raw_values' :
|
| 1444 |
+
Returns a full set of errors in case of multioutput input.
|
| 1445 |
+
|
| 1446 |
+
'uniform_average' :
|
| 1447 |
+
Scores of all outputs are averaged with uniform weight.
|
| 1448 |
+
|
| 1449 |
+
Returns
|
| 1450 |
+
-------
|
| 1451 |
+
score : float or ndarray of floats
|
| 1452 |
+
The :math:`D^2` score with an absolute error deviance
|
| 1453 |
+
or ndarray of scores if 'multioutput' is 'raw_values'.
|
| 1454 |
+
|
| 1455 |
+
Notes
|
| 1456 |
+
-----
|
| 1457 |
+
Like :math:`R^2`, :math:`D^2` score may be negative
|
| 1458 |
+
(it need not actually be the square of a quantity D).
|
| 1459 |
+
|
| 1460 |
+
This metric is not well-defined for single samples and will return a NaN
|
| 1461 |
+
value if n_samples is less than two.
|
| 1462 |
+
|
| 1463 |
+
References
|
| 1464 |
+
----------
|
| 1465 |
+
.. [1] Eq. (3.11) of Hastie, Trevor J., Robert Tibshirani and Martin J.
|
| 1466 |
+
Wainwright. "Statistical Learning with Sparsity: The Lasso and
|
| 1467 |
+
Generalizations." (2015). https://hastie.su.domains/StatLearnSparsity/
|
| 1468 |
+
|
| 1469 |
+
Examples
|
| 1470 |
+
--------
|
| 1471 |
+
>>> from sklearn.metrics import d2_absolute_error_score
|
| 1472 |
+
>>> y_true = [3, -0.5, 2, 7]
|
| 1473 |
+
>>> y_pred = [2.5, 0.0, 2, 8]
|
| 1474 |
+
>>> d2_absolute_error_score(y_true, y_pred)
|
| 1475 |
+
0.764...
|
| 1476 |
+
>>> y_true = [[0.5, 1], [-1, 1], [7, -6]]
|
| 1477 |
+
>>> y_pred = [[0, 2], [-1, 2], [8, -5]]
|
| 1478 |
+
>>> d2_absolute_error_score(y_true, y_pred, multioutput='uniform_average')
|
| 1479 |
+
0.691...
|
| 1480 |
+
>>> d2_absolute_error_score(y_true, y_pred, multioutput='raw_values')
|
| 1481 |
+
array([0.8125 , 0.57142857])
|
| 1482 |
+
>>> y_true = [1, 2, 3]
|
| 1483 |
+
>>> y_pred = [1, 2, 3]
|
| 1484 |
+
>>> d2_absolute_error_score(y_true, y_pred)
|
| 1485 |
+
1.0
|
| 1486 |
+
>>> y_true = [1, 2, 3]
|
| 1487 |
+
>>> y_pred = [2, 2, 2]
|
| 1488 |
+
>>> d2_absolute_error_score(y_true, y_pred)
|
| 1489 |
+
0.0
|
| 1490 |
+
>>> y_true = [1, 2, 3]
|
| 1491 |
+
>>> y_pred = [3, 2, 1]
|
| 1492 |
+
>>> d2_absolute_error_score(y_true, y_pred)
|
| 1493 |
+
-1.0
|
| 1494 |
+
"""
|
| 1495 |
+
return d2_pinball_score(
|
| 1496 |
+
y_true, y_pred, sample_weight=sample_weight, alpha=0.5, multioutput=multioutput
|
| 1497 |
+
)
|
videollama2/lib/python3.10/site-packages/sklearn/metrics/_scorer.py
ADDED
|
@@ -0,0 +1,867 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""
|
| 2 |
+
The :mod:`sklearn.metrics.scorer` submodule implements a flexible
|
| 3 |
+
interface for model selection and evaluation using
|
| 4 |
+
arbitrary score functions.
|
| 5 |
+
|
| 6 |
+
A scorer object is a callable that can be passed to
|
| 7 |
+
:class:`~sklearn.model_selection.GridSearchCV` or
|
| 8 |
+
:func:`sklearn.model_selection.cross_val_score` as the ``scoring``
|
| 9 |
+
parameter, to specify how a model should be evaluated.
|
| 10 |
+
|
| 11 |
+
The signature of the call is ``(estimator, X, y)`` where ``estimator``
|
| 12 |
+
is the model to be evaluated, ``X`` is the test data and ``y`` is the
|
| 13 |
+
ground truth labeling (or ``None`` in the case of unsupervised models).
|
| 14 |
+
"""
|
| 15 |
+
|
| 16 |
+
# Authors: Andreas Mueller <amueller@ais.uni-bonn.de>
|
| 17 |
+
# Lars Buitinck
|
| 18 |
+
# Arnaud Joly <arnaud.v.joly@gmail.com>
|
| 19 |
+
# License: Simplified BSD
|
| 20 |
+
|
| 21 |
+
from collections.abc import Iterable
|
| 22 |
+
from functools import partial
|
| 23 |
+
from collections import Counter
|
| 24 |
+
from traceback import format_exc
|
| 25 |
+
|
| 26 |
+
import numpy as np
|
| 27 |
+
import copy
|
| 28 |
+
import warnings
|
| 29 |
+
|
| 30 |
+
from . import (
|
| 31 |
+
r2_score,
|
| 32 |
+
median_absolute_error,
|
| 33 |
+
max_error,
|
| 34 |
+
mean_absolute_error,
|
| 35 |
+
mean_squared_error,
|
| 36 |
+
mean_squared_log_error,
|
| 37 |
+
mean_poisson_deviance,
|
| 38 |
+
mean_gamma_deviance,
|
| 39 |
+
accuracy_score,
|
| 40 |
+
top_k_accuracy_score,
|
| 41 |
+
f1_score,
|
| 42 |
+
roc_auc_score,
|
| 43 |
+
average_precision_score,
|
| 44 |
+
precision_score,
|
| 45 |
+
recall_score,
|
| 46 |
+
log_loss,
|
| 47 |
+
balanced_accuracy_score,
|
| 48 |
+
explained_variance_score,
|
| 49 |
+
brier_score_loss,
|
| 50 |
+
jaccard_score,
|
| 51 |
+
mean_absolute_percentage_error,
|
| 52 |
+
matthews_corrcoef,
|
| 53 |
+
class_likelihood_ratios,
|
| 54 |
+
)
|
| 55 |
+
|
| 56 |
+
from .cluster import adjusted_rand_score
|
| 57 |
+
from .cluster import rand_score
|
| 58 |
+
from .cluster import homogeneity_score
|
| 59 |
+
from .cluster import completeness_score
|
| 60 |
+
from .cluster import v_measure_score
|
| 61 |
+
from .cluster import mutual_info_score
|
| 62 |
+
from .cluster import adjusted_mutual_info_score
|
| 63 |
+
from .cluster import normalized_mutual_info_score
|
| 64 |
+
from .cluster import fowlkes_mallows_score
|
| 65 |
+
|
| 66 |
+
from ..utils.multiclass import type_of_target
|
| 67 |
+
from ..base import is_regressor
|
| 68 |
+
|
| 69 |
+
|
| 70 |
+
def _cached_call(cache, estimator, method, *args, **kwargs):
|
| 71 |
+
"""Call estimator with method and args and kwargs."""
|
| 72 |
+
if cache is None:
|
| 73 |
+
return getattr(estimator, method)(*args, **kwargs)
|
| 74 |
+
|
| 75 |
+
try:
|
| 76 |
+
return cache[method]
|
| 77 |
+
except KeyError:
|
| 78 |
+
result = getattr(estimator, method)(*args, **kwargs)
|
| 79 |
+
cache[method] = result
|
| 80 |
+
return result
|
| 81 |
+
|
| 82 |
+
|
| 83 |
+
class _MultimetricScorer:
|
| 84 |
+
"""Callable for multimetric scoring used to avoid repeated calls
|
| 85 |
+
to `predict_proba`, `predict`, and `decision_function`.
|
| 86 |
+
|
| 87 |
+
`_MultimetricScorer` will return a dictionary of scores corresponding to
|
| 88 |
+
the scorers in the dictionary. Note that `_MultimetricScorer` can be
|
| 89 |
+
created with a dictionary with one key (i.e. only one actual scorer).
|
| 90 |
+
|
| 91 |
+
Parameters
|
| 92 |
+
----------
|
| 93 |
+
scorers : dict
|
| 94 |
+
Dictionary mapping names to callable scorers.
|
| 95 |
+
|
| 96 |
+
raise_exc : bool, default=True
|
| 97 |
+
Whether to raise the exception in `__call__` or not. If set to `False`
|
| 98 |
+
a formatted string of the exception details is passed as result of
|
| 99 |
+
the failing scorer.
|
| 100 |
+
"""
|
| 101 |
+
|
| 102 |
+
def __init__(self, *, scorers, raise_exc=True):
|
| 103 |
+
self._scorers = scorers
|
| 104 |
+
self._raise_exc = raise_exc
|
| 105 |
+
|
| 106 |
+
def __call__(self, estimator, *args, **kwargs):
|
| 107 |
+
"""Evaluate predicted target values."""
|
| 108 |
+
scores = {}
|
| 109 |
+
cache = {} if self._use_cache(estimator) else None
|
| 110 |
+
cached_call = partial(_cached_call, cache)
|
| 111 |
+
|
| 112 |
+
for name, scorer in self._scorers.items():
|
| 113 |
+
try:
|
| 114 |
+
if isinstance(scorer, _BaseScorer):
|
| 115 |
+
score = scorer._score(cached_call, estimator, *args, **kwargs)
|
| 116 |
+
else:
|
| 117 |
+
score = scorer(estimator, *args, **kwargs)
|
| 118 |
+
scores[name] = score
|
| 119 |
+
except Exception as e:
|
| 120 |
+
if self._raise_exc:
|
| 121 |
+
raise e
|
| 122 |
+
else:
|
| 123 |
+
scores[name] = format_exc()
|
| 124 |
+
|
| 125 |
+
return scores
|
| 126 |
+
|
| 127 |
+
def _use_cache(self, estimator):
|
| 128 |
+
"""Return True if using a cache is beneficial.
|
| 129 |
+
|
| 130 |
+
Caching may be beneficial when one of these conditions holds:
|
| 131 |
+
- `_ProbaScorer` will be called twice.
|
| 132 |
+
- `_PredictScorer` will be called twice.
|
| 133 |
+
- `_ThresholdScorer` will be called twice.
|
| 134 |
+
- `_ThresholdScorer` and `_PredictScorer` are called and
|
| 135 |
+
estimator is a regressor.
|
| 136 |
+
- `_ThresholdScorer` and `_ProbaScorer` are called and
|
| 137 |
+
estimator does not have a `decision_function` attribute.
|
| 138 |
+
|
| 139 |
+
"""
|
| 140 |
+
if len(self._scorers) == 1: # Only one scorer
|
| 141 |
+
return False
|
| 142 |
+
|
| 143 |
+
counter = Counter([type(v) for v in self._scorers.values()])
|
| 144 |
+
|
| 145 |
+
if any(
|
| 146 |
+
counter[known_type] > 1
|
| 147 |
+
for known_type in [_PredictScorer, _ProbaScorer, _ThresholdScorer]
|
| 148 |
+
):
|
| 149 |
+
return True
|
| 150 |
+
|
| 151 |
+
if counter[_ThresholdScorer]:
|
| 152 |
+
if is_regressor(estimator) and counter[_PredictScorer]:
|
| 153 |
+
return True
|
| 154 |
+
elif counter[_ProbaScorer] and not hasattr(estimator, "decision_function"):
|
| 155 |
+
return True
|
| 156 |
+
return False
|
| 157 |
+
|
| 158 |
+
|
| 159 |
+
class _BaseScorer:
|
| 160 |
+
def __init__(self, score_func, sign, kwargs):
|
| 161 |
+
self._kwargs = kwargs
|
| 162 |
+
self._score_func = score_func
|
| 163 |
+
self._sign = sign
|
| 164 |
+
|
| 165 |
+
@staticmethod
|
| 166 |
+
def _check_pos_label(pos_label, classes):
|
| 167 |
+
if pos_label not in list(classes):
|
| 168 |
+
raise ValueError(f"pos_label={pos_label} is not a valid label: {classes}")
|
| 169 |
+
|
| 170 |
+
def _select_proba_binary(self, y_pred, classes):
|
| 171 |
+
"""Select the column of the positive label in `y_pred` when
|
| 172 |
+
probabilities are provided.
|
| 173 |
+
|
| 174 |
+
Parameters
|
| 175 |
+
----------
|
| 176 |
+
y_pred : ndarray of shape (n_samples, n_classes)
|
| 177 |
+
The prediction given by `predict_proba`.
|
| 178 |
+
|
| 179 |
+
classes : ndarray of shape (n_classes,)
|
| 180 |
+
The class labels for the estimator.
|
| 181 |
+
|
| 182 |
+
Returns
|
| 183 |
+
-------
|
| 184 |
+
y_pred : ndarray of shape (n_samples,)
|
| 185 |
+
Probability predictions of the positive class.
|
| 186 |
+
"""
|
| 187 |
+
if y_pred.shape[1] == 2:
|
| 188 |
+
pos_label = self._kwargs.get("pos_label", classes[1])
|
| 189 |
+
self._check_pos_label(pos_label, classes)
|
| 190 |
+
col_idx = np.flatnonzero(classes == pos_label)[0]
|
| 191 |
+
return y_pred[:, col_idx]
|
| 192 |
+
|
| 193 |
+
err_msg = (
|
| 194 |
+
f"Got predict_proba of shape {y_pred.shape}, but need "
|
| 195 |
+
f"classifier with two classes for {self._score_func.__name__} "
|
| 196 |
+
"scoring"
|
| 197 |
+
)
|
| 198 |
+
raise ValueError(err_msg)
|
| 199 |
+
|
| 200 |
+
def __repr__(self):
|
| 201 |
+
kwargs_string = "".join(
|
| 202 |
+
[", %s=%s" % (str(k), str(v)) for k, v in self._kwargs.items()]
|
| 203 |
+
)
|
| 204 |
+
return "make_scorer(%s%s%s%s)" % (
|
| 205 |
+
self._score_func.__name__,
|
| 206 |
+
"" if self._sign > 0 else ", greater_is_better=False",
|
| 207 |
+
self._factory_args(),
|
| 208 |
+
kwargs_string,
|
| 209 |
+
)
|
| 210 |
+
|
| 211 |
+
def __call__(self, estimator, X, y_true, sample_weight=None):
|
| 212 |
+
"""Evaluate predicted target values for X relative to y_true.
|
| 213 |
+
|
| 214 |
+
Parameters
|
| 215 |
+
----------
|
| 216 |
+
estimator : object
|
| 217 |
+
Trained estimator to use for scoring. Must have a predict_proba
|
| 218 |
+
method; the output of that is used to compute the score.
|
| 219 |
+
|
| 220 |
+
X : {array-like, sparse matrix}
|
| 221 |
+
Test data that will be fed to estimator.predict.
|
| 222 |
+
|
| 223 |
+
y_true : array-like
|
| 224 |
+
Gold standard target values for X.
|
| 225 |
+
|
| 226 |
+
sample_weight : array-like of shape (n_samples,), default=None
|
| 227 |
+
Sample weights.
|
| 228 |
+
|
| 229 |
+
Returns
|
| 230 |
+
-------
|
| 231 |
+
score : float
|
| 232 |
+
Score function applied to prediction of estimator on X.
|
| 233 |
+
"""
|
| 234 |
+
return self._score(
|
| 235 |
+
partial(_cached_call, None),
|
| 236 |
+
estimator,
|
| 237 |
+
X,
|
| 238 |
+
y_true,
|
| 239 |
+
sample_weight=sample_weight,
|
| 240 |
+
)
|
| 241 |
+
|
| 242 |
+
def _factory_args(self):
|
| 243 |
+
"""Return non-default make_scorer arguments for repr."""
|
| 244 |
+
return ""
|
| 245 |
+
|
| 246 |
+
|
| 247 |
+
class _PredictScorer(_BaseScorer):
|
| 248 |
+
def _score(self, method_caller, estimator, X, y_true, sample_weight=None):
|
| 249 |
+
"""Evaluate predicted target values for X relative to y_true.
|
| 250 |
+
|
| 251 |
+
Parameters
|
| 252 |
+
----------
|
| 253 |
+
method_caller : callable
|
| 254 |
+
Returns predictions given an estimator, method name, and other
|
| 255 |
+
arguments, potentially caching results.
|
| 256 |
+
|
| 257 |
+
estimator : object
|
| 258 |
+
Trained estimator to use for scoring. Must have a `predict`
|
| 259 |
+
method; the output of that is used to compute the score.
|
| 260 |
+
|
| 261 |
+
X : {array-like, sparse matrix}
|
| 262 |
+
Test data that will be fed to estimator.predict.
|
| 263 |
+
|
| 264 |
+
y_true : array-like
|
| 265 |
+
Gold standard target values for X.
|
| 266 |
+
|
| 267 |
+
sample_weight : array-like of shape (n_samples,), default=None
|
| 268 |
+
Sample weights.
|
| 269 |
+
|
| 270 |
+
Returns
|
| 271 |
+
-------
|
| 272 |
+
score : float
|
| 273 |
+
Score function applied to prediction of estimator on X.
|
| 274 |
+
"""
|
| 275 |
+
|
| 276 |
+
y_pred = method_caller(estimator, "predict", X)
|
| 277 |
+
if sample_weight is not None:
|
| 278 |
+
return self._sign * self._score_func(
|
| 279 |
+
y_true, y_pred, sample_weight=sample_weight, **self._kwargs
|
| 280 |
+
)
|
| 281 |
+
else:
|
| 282 |
+
return self._sign * self._score_func(y_true, y_pred, **self._kwargs)
|
| 283 |
+
|
| 284 |
+
|
| 285 |
+
class _ProbaScorer(_BaseScorer):
|
| 286 |
+
def _score(self, method_caller, clf, X, y, sample_weight=None):
|
| 287 |
+
"""Evaluate predicted probabilities for X relative to y_true.
|
| 288 |
+
|
| 289 |
+
Parameters
|
| 290 |
+
----------
|
| 291 |
+
method_caller : callable
|
| 292 |
+
Returns predictions given an estimator, method name, and other
|
| 293 |
+
arguments, potentially caching results.
|
| 294 |
+
|
| 295 |
+
clf : object
|
| 296 |
+
Trained classifier to use for scoring. Must have a `predict_proba`
|
| 297 |
+
method; the output of that is used to compute the score.
|
| 298 |
+
|
| 299 |
+
X : {array-like, sparse matrix}
|
| 300 |
+
Test data that will be fed to clf.predict_proba.
|
| 301 |
+
|
| 302 |
+
y : array-like
|
| 303 |
+
Gold standard target values for X. These must be class labels,
|
| 304 |
+
not probabilities.
|
| 305 |
+
|
| 306 |
+
sample_weight : array-like, default=None
|
| 307 |
+
Sample weights.
|
| 308 |
+
|
| 309 |
+
Returns
|
| 310 |
+
-------
|
| 311 |
+
score : float
|
| 312 |
+
Score function applied to prediction of estimator on X.
|
| 313 |
+
"""
|
| 314 |
+
|
| 315 |
+
y_type = type_of_target(y)
|
| 316 |
+
y_pred = method_caller(clf, "predict_proba", X)
|
| 317 |
+
if y_type == "binary" and y_pred.shape[1] <= 2:
|
| 318 |
+
# `y_type` could be equal to "binary" even in a multi-class
|
| 319 |
+
# problem: (when only 2 class are given to `y_true` during scoring)
|
| 320 |
+
# Thus, we need to check for the shape of `y_pred`.
|
| 321 |
+
y_pred = self._select_proba_binary(y_pred, clf.classes_)
|
| 322 |
+
if sample_weight is not None:
|
| 323 |
+
return self._sign * self._score_func(
|
| 324 |
+
y, y_pred, sample_weight=sample_weight, **self._kwargs
|
| 325 |
+
)
|
| 326 |
+
else:
|
| 327 |
+
return self._sign * self._score_func(y, y_pred, **self._kwargs)
|
| 328 |
+
|
| 329 |
+
def _factory_args(self):
|
| 330 |
+
return ", needs_proba=True"
|
| 331 |
+
|
| 332 |
+
|
| 333 |
+
class _ThresholdScorer(_BaseScorer):
|
| 334 |
+
def _score(self, method_caller, clf, X, y, sample_weight=None):
|
| 335 |
+
"""Evaluate decision function output for X relative to y_true.
|
| 336 |
+
|
| 337 |
+
Parameters
|
| 338 |
+
----------
|
| 339 |
+
method_caller : callable
|
| 340 |
+
Returns predictions given an estimator, method name, and other
|
| 341 |
+
arguments, potentially caching results.
|
| 342 |
+
|
| 343 |
+
clf : object
|
| 344 |
+
Trained classifier to use for scoring. Must have either a
|
| 345 |
+
decision_function method or a predict_proba method; the output of
|
| 346 |
+
that is used to compute the score.
|
| 347 |
+
|
| 348 |
+
X : {array-like, sparse matrix}
|
| 349 |
+
Test data that will be fed to clf.decision_function or
|
| 350 |
+
clf.predict_proba.
|
| 351 |
+
|
| 352 |
+
y : array-like
|
| 353 |
+
Gold standard target values for X. These must be class labels,
|
| 354 |
+
not decision function values.
|
| 355 |
+
|
| 356 |
+
sample_weight : array-like, default=None
|
| 357 |
+
Sample weights.
|
| 358 |
+
|
| 359 |
+
Returns
|
| 360 |
+
-------
|
| 361 |
+
score : float
|
| 362 |
+
Score function applied to prediction of estimator on X.
|
| 363 |
+
"""
|
| 364 |
+
|
| 365 |
+
y_type = type_of_target(y)
|
| 366 |
+
if y_type not in ("binary", "multilabel-indicator"):
|
| 367 |
+
raise ValueError("{0} format is not supported".format(y_type))
|
| 368 |
+
|
| 369 |
+
if is_regressor(clf):
|
| 370 |
+
y_pred = method_caller(clf, "predict", X)
|
| 371 |
+
else:
|
| 372 |
+
try:
|
| 373 |
+
y_pred = method_caller(clf, "decision_function", X)
|
| 374 |
+
|
| 375 |
+
if isinstance(y_pred, list):
|
| 376 |
+
# For multi-output multi-class estimator
|
| 377 |
+
y_pred = np.vstack([p for p in y_pred]).T
|
| 378 |
+
elif y_type == "binary" and "pos_label" in self._kwargs:
|
| 379 |
+
self._check_pos_label(self._kwargs["pos_label"], clf.classes_)
|
| 380 |
+
if self._kwargs["pos_label"] == clf.classes_[0]:
|
| 381 |
+
# The implicit positive class of the binary classifier
|
| 382 |
+
# does not match `pos_label`: we need to invert the
|
| 383 |
+
# predictions
|
| 384 |
+
y_pred *= -1
|
| 385 |
+
|
| 386 |
+
except (NotImplementedError, AttributeError):
|
| 387 |
+
y_pred = method_caller(clf, "predict_proba", X)
|
| 388 |
+
|
| 389 |
+
if y_type == "binary":
|
| 390 |
+
y_pred = self._select_proba_binary(y_pred, clf.classes_)
|
| 391 |
+
elif isinstance(y_pred, list):
|
| 392 |
+
y_pred = np.vstack([p[:, -1] for p in y_pred]).T
|
| 393 |
+
|
| 394 |
+
if sample_weight is not None:
|
| 395 |
+
return self._sign * self._score_func(
|
| 396 |
+
y, y_pred, sample_weight=sample_weight, **self._kwargs
|
| 397 |
+
)
|
| 398 |
+
else:
|
| 399 |
+
return self._sign * self._score_func(y, y_pred, **self._kwargs)
|
| 400 |
+
|
| 401 |
+
def _factory_args(self):
|
| 402 |
+
return ", needs_threshold=True"
|
| 403 |
+
|
| 404 |
+
|
| 405 |
+
def get_scorer(scoring):
|
| 406 |
+
"""Get a scorer from string.
|
| 407 |
+
|
| 408 |
+
Read more in the :ref:`User Guide <scoring_parameter>`.
|
| 409 |
+
:func:`~sklearn.metrics.get_scorer_names` can be used to retrieve the names
|
| 410 |
+
of all available scorers.
|
| 411 |
+
|
| 412 |
+
Parameters
|
| 413 |
+
----------
|
| 414 |
+
scoring : str or callable
|
| 415 |
+
Scoring method as string. If callable it is returned as is.
|
| 416 |
+
|
| 417 |
+
Returns
|
| 418 |
+
-------
|
| 419 |
+
scorer : callable
|
| 420 |
+
The scorer.
|
| 421 |
+
|
| 422 |
+
Notes
|
| 423 |
+
-----
|
| 424 |
+
When passed a string, this function always returns a copy of the scorer
|
| 425 |
+
object. Calling `get_scorer` twice for the same scorer results in two
|
| 426 |
+
separate scorer objects.
|
| 427 |
+
"""
|
| 428 |
+
if isinstance(scoring, str):
|
| 429 |
+
try:
|
| 430 |
+
scorer = copy.deepcopy(_SCORERS[scoring])
|
| 431 |
+
except KeyError:
|
| 432 |
+
raise ValueError(
|
| 433 |
+
"%r is not a valid scoring value. "
|
| 434 |
+
"Use sklearn.metrics.get_scorer_names() "
|
| 435 |
+
"to get valid options." % scoring
|
| 436 |
+
)
|
| 437 |
+
else:
|
| 438 |
+
scorer = scoring
|
| 439 |
+
return scorer
|
| 440 |
+
|
| 441 |
+
|
| 442 |
+
def _passthrough_scorer(estimator, *args, **kwargs):
|
| 443 |
+
"""Function that wraps estimator.score"""
|
| 444 |
+
return estimator.score(*args, **kwargs)
|
| 445 |
+
|
| 446 |
+
|
| 447 |
+
def check_scoring(estimator, scoring=None, *, allow_none=False):
|
| 448 |
+
"""Determine scorer from user options.
|
| 449 |
+
|
| 450 |
+
A TypeError will be thrown if the estimator cannot be scored.
|
| 451 |
+
|
| 452 |
+
Parameters
|
| 453 |
+
----------
|
| 454 |
+
estimator : estimator object implementing 'fit'
|
| 455 |
+
The object to use to fit the data.
|
| 456 |
+
|
| 457 |
+
scoring : str or callable, default=None
|
| 458 |
+
A string (see model evaluation documentation) or
|
| 459 |
+
a scorer callable object / function with signature
|
| 460 |
+
``scorer(estimator, X, y)``.
|
| 461 |
+
If None, the provided estimator object's `score` method is used.
|
| 462 |
+
|
| 463 |
+
allow_none : bool, default=False
|
| 464 |
+
If no scoring is specified and the estimator has no score function, we
|
| 465 |
+
can either return None or raise an exception.
|
| 466 |
+
|
| 467 |
+
Returns
|
| 468 |
+
-------
|
| 469 |
+
scoring : callable
|
| 470 |
+
A scorer callable object / function with signature
|
| 471 |
+
``scorer(estimator, X, y)``.
|
| 472 |
+
"""
|
| 473 |
+
if not hasattr(estimator, "fit"):
|
| 474 |
+
raise TypeError(
|
| 475 |
+
"estimator should be an estimator implementing 'fit' method, %r was passed"
|
| 476 |
+
% estimator
|
| 477 |
+
)
|
| 478 |
+
if isinstance(scoring, str):
|
| 479 |
+
return get_scorer(scoring)
|
| 480 |
+
elif callable(scoring):
|
| 481 |
+
# Heuristic to ensure user has not passed a metric
|
| 482 |
+
module = getattr(scoring, "__module__", None)
|
| 483 |
+
if (
|
| 484 |
+
hasattr(module, "startswith")
|
| 485 |
+
and module.startswith("sklearn.metrics.")
|
| 486 |
+
and not module.startswith("sklearn.metrics._scorer")
|
| 487 |
+
and not module.startswith("sklearn.metrics.tests.")
|
| 488 |
+
):
|
| 489 |
+
raise ValueError(
|
| 490 |
+
"scoring value %r looks like it is a metric "
|
| 491 |
+
"function rather than a scorer. A scorer should "
|
| 492 |
+
"require an estimator as its first parameter. "
|
| 493 |
+
"Please use `make_scorer` to convert a metric "
|
| 494 |
+
"to a scorer." % scoring
|
| 495 |
+
)
|
| 496 |
+
return get_scorer(scoring)
|
| 497 |
+
elif scoring is None:
|
| 498 |
+
if hasattr(estimator, "score"):
|
| 499 |
+
return _passthrough_scorer
|
| 500 |
+
elif allow_none:
|
| 501 |
+
return None
|
| 502 |
+
else:
|
| 503 |
+
raise TypeError(
|
| 504 |
+
"If no scoring is specified, the estimator passed should "
|
| 505 |
+
"have a 'score' method. The estimator %r does not." % estimator
|
| 506 |
+
)
|
| 507 |
+
elif isinstance(scoring, Iterable):
|
| 508 |
+
raise ValueError(
|
| 509 |
+
"For evaluating multiple scores, use "
|
| 510 |
+
"sklearn.model_selection.cross_validate instead. "
|
| 511 |
+
"{0} was passed.".format(scoring)
|
| 512 |
+
)
|
| 513 |
+
else:
|
| 514 |
+
raise ValueError(
|
| 515 |
+
"scoring value should either be a callable, string or None. %r was passed"
|
| 516 |
+
% scoring
|
| 517 |
+
)
|
| 518 |
+
|
| 519 |
+
|
| 520 |
+
def _check_multimetric_scoring(estimator, scoring):
|
| 521 |
+
"""Check the scoring parameter in cases when multiple metrics are allowed.
|
| 522 |
+
|
| 523 |
+
Parameters
|
| 524 |
+
----------
|
| 525 |
+
estimator : sklearn estimator instance
|
| 526 |
+
The estimator for which the scoring will be applied.
|
| 527 |
+
|
| 528 |
+
scoring : list, tuple or dict
|
| 529 |
+
Strategy to evaluate the performance of the cross-validated model on
|
| 530 |
+
the test set.
|
| 531 |
+
|
| 532 |
+
The possibilities are:
|
| 533 |
+
|
| 534 |
+
- a list or tuple of unique strings;
|
| 535 |
+
- a callable returning a dictionary where they keys are the metric
|
| 536 |
+
names and the values are the metric scores;
|
| 537 |
+
- a dictionary with metric names as keys and callables a values.
|
| 538 |
+
|
| 539 |
+
See :ref:`multimetric_grid_search` for an example.
|
| 540 |
+
|
| 541 |
+
Returns
|
| 542 |
+
-------
|
| 543 |
+
scorers_dict : dict
|
| 544 |
+
A dict mapping each scorer name to its validated scorer.
|
| 545 |
+
"""
|
| 546 |
+
err_msg_generic = (
|
| 547 |
+
f"scoring is invalid (got {scoring!r}). Refer to the "
|
| 548 |
+
"scoring glossary for details: "
|
| 549 |
+
"https://scikit-learn.org/stable/glossary.html#term-scoring"
|
| 550 |
+
)
|
| 551 |
+
|
| 552 |
+
if isinstance(scoring, (list, tuple, set)):
|
| 553 |
+
err_msg = (
|
| 554 |
+
"The list/tuple elements must be unique strings of predefined scorers. "
|
| 555 |
+
)
|
| 556 |
+
try:
|
| 557 |
+
keys = set(scoring)
|
| 558 |
+
except TypeError as e:
|
| 559 |
+
raise ValueError(err_msg) from e
|
| 560 |
+
|
| 561 |
+
if len(keys) != len(scoring):
|
| 562 |
+
raise ValueError(
|
| 563 |
+
f"{err_msg} Duplicate elements were found in"
|
| 564 |
+
f" the given list. {scoring!r}"
|
| 565 |
+
)
|
| 566 |
+
elif len(keys) > 0:
|
| 567 |
+
if not all(isinstance(k, str) for k in keys):
|
| 568 |
+
if any(callable(k) for k in keys):
|
| 569 |
+
raise ValueError(
|
| 570 |
+
f"{err_msg} One or more of the elements "
|
| 571 |
+
"were callables. Use a dict of score "
|
| 572 |
+
"name mapped to the scorer callable. "
|
| 573 |
+
f"Got {scoring!r}"
|
| 574 |
+
)
|
| 575 |
+
else:
|
| 576 |
+
raise ValueError(
|
| 577 |
+
f"{err_msg} Non-string types were found "
|
| 578 |
+
f"in the given list. Got {scoring!r}"
|
| 579 |
+
)
|
| 580 |
+
scorers = {
|
| 581 |
+
scorer: check_scoring(estimator, scoring=scorer) for scorer in scoring
|
| 582 |
+
}
|
| 583 |
+
else:
|
| 584 |
+
raise ValueError(f"{err_msg} Empty list was given. {scoring!r}")
|
| 585 |
+
|
| 586 |
+
elif isinstance(scoring, dict):
|
| 587 |
+
keys = set(scoring)
|
| 588 |
+
if not all(isinstance(k, str) for k in keys):
|
| 589 |
+
raise ValueError(
|
| 590 |
+
"Non-string types were found in the keys of "
|
| 591 |
+
f"the given dict. scoring={scoring!r}"
|
| 592 |
+
)
|
| 593 |
+
if len(keys) == 0:
|
| 594 |
+
raise ValueError(f"An empty dict was passed. {scoring!r}")
|
| 595 |
+
scorers = {
|
| 596 |
+
key: check_scoring(estimator, scoring=scorer)
|
| 597 |
+
for key, scorer in scoring.items()
|
| 598 |
+
}
|
| 599 |
+
else:
|
| 600 |
+
raise ValueError(err_msg_generic)
|
| 601 |
+
return scorers
|
| 602 |
+
|
| 603 |
+
|
| 604 |
+
def make_scorer(
|
| 605 |
+
score_func,
|
| 606 |
+
*,
|
| 607 |
+
greater_is_better=True,
|
| 608 |
+
needs_proba=False,
|
| 609 |
+
needs_threshold=False,
|
| 610 |
+
**kwargs,
|
| 611 |
+
):
|
| 612 |
+
"""Make a scorer from a performance metric or loss function.
|
| 613 |
+
|
| 614 |
+
This factory function wraps scoring functions for use in
|
| 615 |
+
:class:`~sklearn.model_selection.GridSearchCV` and
|
| 616 |
+
:func:`~sklearn.model_selection.cross_val_score`.
|
| 617 |
+
It takes a score function, such as :func:`~sklearn.metrics.accuracy_score`,
|
| 618 |
+
:func:`~sklearn.metrics.mean_squared_error`,
|
| 619 |
+
:func:`~sklearn.metrics.adjusted_rand_score` or
|
| 620 |
+
:func:`~sklearn.metrics.average_precision_score`
|
| 621 |
+
and returns a callable that scores an estimator's output.
|
| 622 |
+
The signature of the call is `(estimator, X, y)` where `estimator`
|
| 623 |
+
is the model to be evaluated, `X` is the data and `y` is the
|
| 624 |
+
ground truth labeling (or `None` in the case of unsupervised models).
|
| 625 |
+
|
| 626 |
+
Read more in the :ref:`User Guide <scoring>`.
|
| 627 |
+
|
| 628 |
+
Parameters
|
| 629 |
+
----------
|
| 630 |
+
score_func : callable
|
| 631 |
+
Score function (or loss function) with signature
|
| 632 |
+
`score_func(y, y_pred, **kwargs)`.
|
| 633 |
+
|
| 634 |
+
greater_is_better : bool, default=True
|
| 635 |
+
Whether `score_func` is a score function (default), meaning high is
|
| 636 |
+
good, or a loss function, meaning low is good. In the latter case, the
|
| 637 |
+
scorer object will sign-flip the outcome of the `score_func`.
|
| 638 |
+
|
| 639 |
+
needs_proba : bool, default=False
|
| 640 |
+
Whether `score_func` requires `predict_proba` to get probability
|
| 641 |
+
estimates out of a classifier.
|
| 642 |
+
|
| 643 |
+
If True, for binary `y_true`, the score function is supposed to accept
|
| 644 |
+
a 1D `y_pred` (i.e., probability of the positive class, shape
|
| 645 |
+
`(n_samples,)`).
|
| 646 |
+
|
| 647 |
+
needs_threshold : bool, default=False
|
| 648 |
+
Whether `score_func` takes a continuous decision certainty.
|
| 649 |
+
This only works for binary classification using estimators that
|
| 650 |
+
have either a `decision_function` or `predict_proba` method.
|
| 651 |
+
|
| 652 |
+
If True, for binary `y_true`, the score function is supposed to accept
|
| 653 |
+
a 1D `y_pred` (i.e., probability of the positive class or the decision
|
| 654 |
+
function, shape `(n_samples,)`).
|
| 655 |
+
|
| 656 |
+
For example `average_precision` or the area under the roc curve
|
| 657 |
+
can not be computed using discrete predictions alone.
|
| 658 |
+
|
| 659 |
+
**kwargs : additional arguments
|
| 660 |
+
Additional parameters to be passed to `score_func`.
|
| 661 |
+
|
| 662 |
+
Returns
|
| 663 |
+
-------
|
| 664 |
+
scorer : callable
|
| 665 |
+
Callable object that returns a scalar score; greater is better.
|
| 666 |
+
|
| 667 |
+
Notes
|
| 668 |
+
-----
|
| 669 |
+
If `needs_proba=False` and `needs_threshold=False`, the score
|
| 670 |
+
function is supposed to accept the output of :term:`predict`. If
|
| 671 |
+
`needs_proba=True`, the score function is supposed to accept the
|
| 672 |
+
output of :term:`predict_proba` (For binary `y_true`, the score function is
|
| 673 |
+
supposed to accept probability of the positive class). If
|
| 674 |
+
`needs_threshold=True`, the score function is supposed to accept the
|
| 675 |
+
output of :term:`decision_function` or :term:`predict_proba` when
|
| 676 |
+
:term:`decision_function` is not present.
|
| 677 |
+
|
| 678 |
+
Examples
|
| 679 |
+
--------
|
| 680 |
+
>>> from sklearn.metrics import fbeta_score, make_scorer
|
| 681 |
+
>>> ftwo_scorer = make_scorer(fbeta_score, beta=2)
|
| 682 |
+
>>> ftwo_scorer
|
| 683 |
+
make_scorer(fbeta_score, beta=2)
|
| 684 |
+
>>> from sklearn.model_selection import GridSearchCV
|
| 685 |
+
>>> from sklearn.svm import LinearSVC
|
| 686 |
+
>>> grid = GridSearchCV(LinearSVC(), param_grid={'C': [1, 10]},
|
| 687 |
+
... scoring=ftwo_scorer)
|
| 688 |
+
"""
|
| 689 |
+
sign = 1 if greater_is_better else -1
|
| 690 |
+
if needs_proba and needs_threshold:
|
| 691 |
+
raise ValueError(
|
| 692 |
+
"Set either needs_proba or needs_threshold to True, but not both."
|
| 693 |
+
)
|
| 694 |
+
if needs_proba:
|
| 695 |
+
cls = _ProbaScorer
|
| 696 |
+
elif needs_threshold:
|
| 697 |
+
cls = _ThresholdScorer
|
| 698 |
+
else:
|
| 699 |
+
cls = _PredictScorer
|
| 700 |
+
return cls(score_func, sign, kwargs)
|
| 701 |
+
|
| 702 |
+
|
| 703 |
+
# Standard regression scores
|
| 704 |
+
explained_variance_scorer = make_scorer(explained_variance_score)
|
| 705 |
+
r2_scorer = make_scorer(r2_score)
|
| 706 |
+
max_error_scorer = make_scorer(max_error, greater_is_better=False)
|
| 707 |
+
neg_mean_squared_error_scorer = make_scorer(mean_squared_error, greater_is_better=False)
|
| 708 |
+
neg_mean_squared_log_error_scorer = make_scorer(
|
| 709 |
+
mean_squared_log_error, greater_is_better=False
|
| 710 |
+
)
|
| 711 |
+
neg_mean_absolute_error_scorer = make_scorer(
|
| 712 |
+
mean_absolute_error, greater_is_better=False
|
| 713 |
+
)
|
| 714 |
+
neg_mean_absolute_percentage_error_scorer = make_scorer(
|
| 715 |
+
mean_absolute_percentage_error, greater_is_better=False
|
| 716 |
+
)
|
| 717 |
+
neg_median_absolute_error_scorer = make_scorer(
|
| 718 |
+
median_absolute_error, greater_is_better=False
|
| 719 |
+
)
|
| 720 |
+
neg_root_mean_squared_error_scorer = make_scorer(
|
| 721 |
+
mean_squared_error, greater_is_better=False, squared=False
|
| 722 |
+
)
|
| 723 |
+
neg_mean_poisson_deviance_scorer = make_scorer(
|
| 724 |
+
mean_poisson_deviance, greater_is_better=False
|
| 725 |
+
)
|
| 726 |
+
|
| 727 |
+
neg_mean_gamma_deviance_scorer = make_scorer(
|
| 728 |
+
mean_gamma_deviance, greater_is_better=False
|
| 729 |
+
)
|
| 730 |
+
|
| 731 |
+
# Standard Classification Scores
|
| 732 |
+
accuracy_scorer = make_scorer(accuracy_score)
|
| 733 |
+
balanced_accuracy_scorer = make_scorer(balanced_accuracy_score)
|
| 734 |
+
matthews_corrcoef_scorer = make_scorer(matthews_corrcoef)
|
| 735 |
+
|
| 736 |
+
|
| 737 |
+
def positive_likelihood_ratio(y_true, y_pred):
|
| 738 |
+
return class_likelihood_ratios(y_true, y_pred)[0]
|
| 739 |
+
|
| 740 |
+
|
| 741 |
+
def negative_likelihood_ratio(y_true, y_pred):
|
| 742 |
+
return class_likelihood_ratios(y_true, y_pred)[1]
|
| 743 |
+
|
| 744 |
+
|
| 745 |
+
positive_likelihood_ratio_scorer = make_scorer(positive_likelihood_ratio)
|
| 746 |
+
neg_negative_likelihood_ratio_scorer = make_scorer(
|
| 747 |
+
negative_likelihood_ratio, greater_is_better=False
|
| 748 |
+
)
|
| 749 |
+
|
| 750 |
+
# Score functions that need decision values
|
| 751 |
+
top_k_accuracy_scorer = make_scorer(
|
| 752 |
+
top_k_accuracy_score, greater_is_better=True, needs_threshold=True
|
| 753 |
+
)
|
| 754 |
+
roc_auc_scorer = make_scorer(
|
| 755 |
+
roc_auc_score, greater_is_better=True, needs_threshold=True
|
| 756 |
+
)
|
| 757 |
+
average_precision_scorer = make_scorer(average_precision_score, needs_threshold=True)
|
| 758 |
+
roc_auc_ovo_scorer = make_scorer(roc_auc_score, needs_proba=True, multi_class="ovo")
|
| 759 |
+
roc_auc_ovo_weighted_scorer = make_scorer(
|
| 760 |
+
roc_auc_score, needs_proba=True, multi_class="ovo", average="weighted"
|
| 761 |
+
)
|
| 762 |
+
roc_auc_ovr_scorer = make_scorer(roc_auc_score, needs_proba=True, multi_class="ovr")
|
| 763 |
+
roc_auc_ovr_weighted_scorer = make_scorer(
|
| 764 |
+
roc_auc_score, needs_proba=True, multi_class="ovr", average="weighted"
|
| 765 |
+
)
|
| 766 |
+
|
| 767 |
+
# Score function for probabilistic classification
|
| 768 |
+
neg_log_loss_scorer = make_scorer(log_loss, greater_is_better=False, needs_proba=True)
|
| 769 |
+
neg_brier_score_scorer = make_scorer(
|
| 770 |
+
brier_score_loss, greater_is_better=False, needs_proba=True
|
| 771 |
+
)
|
| 772 |
+
brier_score_loss_scorer = make_scorer(
|
| 773 |
+
brier_score_loss, greater_is_better=False, needs_proba=True
|
| 774 |
+
)
|
| 775 |
+
|
| 776 |
+
|
| 777 |
+
# Clustering scores
|
| 778 |
+
adjusted_rand_scorer = make_scorer(adjusted_rand_score)
|
| 779 |
+
rand_scorer = make_scorer(rand_score)
|
| 780 |
+
homogeneity_scorer = make_scorer(homogeneity_score)
|
| 781 |
+
completeness_scorer = make_scorer(completeness_score)
|
| 782 |
+
v_measure_scorer = make_scorer(v_measure_score)
|
| 783 |
+
mutual_info_scorer = make_scorer(mutual_info_score)
|
| 784 |
+
adjusted_mutual_info_scorer = make_scorer(adjusted_mutual_info_score)
|
| 785 |
+
normalized_mutual_info_scorer = make_scorer(normalized_mutual_info_score)
|
| 786 |
+
fowlkes_mallows_scorer = make_scorer(fowlkes_mallows_score)
|
| 787 |
+
|
| 788 |
+
|
| 789 |
+
# TODO(1.3) Remove
|
| 790 |
+
class _DeprecatedScorers(dict):
|
| 791 |
+
"""A temporary class to deprecate SCORERS."""
|
| 792 |
+
|
| 793 |
+
def __getitem__(self, item):
|
| 794 |
+
warnings.warn(
|
| 795 |
+
"sklearn.metrics.SCORERS is deprecated and will be removed in v1.3. "
|
| 796 |
+
"Please use sklearn.metrics.get_scorer_names to get a list of available "
|
| 797 |
+
"scorers and sklearn.metrics.get_metric to get scorer.",
|
| 798 |
+
FutureWarning,
|
| 799 |
+
)
|
| 800 |
+
return super().__getitem__(item)
|
| 801 |
+
|
| 802 |
+
|
| 803 |
+
_SCORERS = dict(
|
| 804 |
+
explained_variance=explained_variance_scorer,
|
| 805 |
+
r2=r2_scorer,
|
| 806 |
+
max_error=max_error_scorer,
|
| 807 |
+
matthews_corrcoef=matthews_corrcoef_scorer,
|
| 808 |
+
neg_median_absolute_error=neg_median_absolute_error_scorer,
|
| 809 |
+
neg_mean_absolute_error=neg_mean_absolute_error_scorer,
|
| 810 |
+
neg_mean_absolute_percentage_error=neg_mean_absolute_percentage_error_scorer, # noqa
|
| 811 |
+
neg_mean_squared_error=neg_mean_squared_error_scorer,
|
| 812 |
+
neg_mean_squared_log_error=neg_mean_squared_log_error_scorer,
|
| 813 |
+
neg_root_mean_squared_error=neg_root_mean_squared_error_scorer,
|
| 814 |
+
neg_mean_poisson_deviance=neg_mean_poisson_deviance_scorer,
|
| 815 |
+
neg_mean_gamma_deviance=neg_mean_gamma_deviance_scorer,
|
| 816 |
+
accuracy=accuracy_scorer,
|
| 817 |
+
top_k_accuracy=top_k_accuracy_scorer,
|
| 818 |
+
roc_auc=roc_auc_scorer,
|
| 819 |
+
roc_auc_ovr=roc_auc_ovr_scorer,
|
| 820 |
+
roc_auc_ovo=roc_auc_ovo_scorer,
|
| 821 |
+
roc_auc_ovr_weighted=roc_auc_ovr_weighted_scorer,
|
| 822 |
+
roc_auc_ovo_weighted=roc_auc_ovo_weighted_scorer,
|
| 823 |
+
balanced_accuracy=balanced_accuracy_scorer,
|
| 824 |
+
average_precision=average_precision_scorer,
|
| 825 |
+
neg_log_loss=neg_log_loss_scorer,
|
| 826 |
+
neg_brier_score=neg_brier_score_scorer,
|
| 827 |
+
positive_likelihood_ratio=positive_likelihood_ratio_scorer,
|
| 828 |
+
neg_negative_likelihood_ratio=neg_negative_likelihood_ratio_scorer,
|
| 829 |
+
# Cluster metrics that use supervised evaluation
|
| 830 |
+
adjusted_rand_score=adjusted_rand_scorer,
|
| 831 |
+
rand_score=rand_scorer,
|
| 832 |
+
homogeneity_score=homogeneity_scorer,
|
| 833 |
+
completeness_score=completeness_scorer,
|
| 834 |
+
v_measure_score=v_measure_scorer,
|
| 835 |
+
mutual_info_score=mutual_info_scorer,
|
| 836 |
+
adjusted_mutual_info_score=adjusted_mutual_info_scorer,
|
| 837 |
+
normalized_mutual_info_score=normalized_mutual_info_scorer,
|
| 838 |
+
fowlkes_mallows_score=fowlkes_mallows_scorer,
|
| 839 |
+
)
|
| 840 |
+
|
| 841 |
+
|
| 842 |
+
def get_scorer_names():
|
| 843 |
+
"""Get the names of all available scorers.
|
| 844 |
+
|
| 845 |
+
These names can be passed to :func:`~sklearn.metrics.get_scorer` to
|
| 846 |
+
retrieve the scorer object.
|
| 847 |
+
|
| 848 |
+
Returns
|
| 849 |
+
-------
|
| 850 |
+
list of str
|
| 851 |
+
Names of all available scorers.
|
| 852 |
+
"""
|
| 853 |
+
return sorted(_SCORERS.keys())
|
| 854 |
+
|
| 855 |
+
|
| 856 |
+
for name, metric in [
|
| 857 |
+
("precision", precision_score),
|
| 858 |
+
("recall", recall_score),
|
| 859 |
+
("f1", f1_score),
|
| 860 |
+
("jaccard", jaccard_score),
|
| 861 |
+
]:
|
| 862 |
+
_SCORERS[name] = make_scorer(metric, average="binary")
|
| 863 |
+
for average in ["macro", "micro", "samples", "weighted"]:
|
| 864 |
+
qualified_name = "{0}_{1}".format(name, average)
|
| 865 |
+
_SCORERS[qualified_name] = make_scorer(metric, pos_label=None, average=average)
|
| 866 |
+
|
| 867 |
+
SCORERS = _DeprecatedScorers(_SCORERS)
|
videollama2/lib/python3.10/site-packages/sklearn/metrics/pairwise.py
ADDED
|
@@ -0,0 +1,2205 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Authors: Alexandre Gramfort <alexandre.gramfort@inria.fr>
|
| 2 |
+
# Mathieu Blondel <mathieu@mblondel.org>
|
| 3 |
+
# Robert Layton <robertlayton@gmail.com>
|
| 4 |
+
# Andreas Mueller <amueller@ais.uni-bonn.de>
|
| 5 |
+
# Philippe Gervais <philippe.gervais@inria.fr>
|
| 6 |
+
# Lars Buitinck
|
| 7 |
+
# Joel Nothman <joel.nothman@gmail.com>
|
| 8 |
+
# License: BSD 3 clause
|
| 9 |
+
|
| 10 |
+
import itertools
|
| 11 |
+
from functools import partial
|
| 12 |
+
import warnings
|
| 13 |
+
|
| 14 |
+
import numpy as np
|
| 15 |
+
from scipy.spatial import distance
|
| 16 |
+
from scipy.sparse import csr_matrix
|
| 17 |
+
from scipy.sparse import issparse
|
| 18 |
+
from joblib import effective_n_jobs
|
| 19 |
+
|
| 20 |
+
from .. import config_context
|
| 21 |
+
from ..utils.validation import _num_samples
|
| 22 |
+
from ..utils.validation import check_non_negative
|
| 23 |
+
from ..utils import check_array
|
| 24 |
+
from ..utils import gen_even_slices
|
| 25 |
+
from ..utils import gen_batches, get_chunk_n_rows
|
| 26 |
+
from ..utils import is_scalar_nan
|
| 27 |
+
from ..utils.extmath import row_norms, safe_sparse_dot
|
| 28 |
+
from ..preprocessing import normalize
|
| 29 |
+
from ..utils._mask import _get_mask
|
| 30 |
+
from ..utils.parallel import delayed, Parallel
|
| 31 |
+
from ..utils.fixes import sp_version, parse_version
|
| 32 |
+
|
| 33 |
+
from ._pairwise_distances_reduction import ArgKmin
|
| 34 |
+
from ._pairwise_fast import _chi2_kernel_fast, _sparse_manhattan
|
| 35 |
+
from ..exceptions import DataConversionWarning
|
| 36 |
+
|
| 37 |
+
|
| 38 |
+
# Utility Functions
|
| 39 |
+
def _return_float_dtype(X, Y):
|
| 40 |
+
"""
|
| 41 |
+
1. If dtype of X and Y is float32, then dtype float32 is returned.
|
| 42 |
+
2. Else dtype float is returned.
|
| 43 |
+
"""
|
| 44 |
+
if not issparse(X) and not isinstance(X, np.ndarray):
|
| 45 |
+
X = np.asarray(X)
|
| 46 |
+
|
| 47 |
+
if Y is None:
|
| 48 |
+
Y_dtype = X.dtype
|
| 49 |
+
elif not issparse(Y) and not isinstance(Y, np.ndarray):
|
| 50 |
+
Y = np.asarray(Y)
|
| 51 |
+
Y_dtype = Y.dtype
|
| 52 |
+
else:
|
| 53 |
+
Y_dtype = Y.dtype
|
| 54 |
+
|
| 55 |
+
if X.dtype == Y_dtype == np.float32:
|
| 56 |
+
dtype = np.float32
|
| 57 |
+
else:
|
| 58 |
+
dtype = float
|
| 59 |
+
|
| 60 |
+
return X, Y, dtype
|
| 61 |
+
|
| 62 |
+
|
| 63 |
+
def check_pairwise_arrays(
|
| 64 |
+
X,
|
| 65 |
+
Y,
|
| 66 |
+
*,
|
| 67 |
+
precomputed=False,
|
| 68 |
+
dtype=None,
|
| 69 |
+
accept_sparse="csr",
|
| 70 |
+
force_all_finite=True,
|
| 71 |
+
copy=False,
|
| 72 |
+
):
|
| 73 |
+
"""Set X and Y appropriately and checks inputs.
|
| 74 |
+
|
| 75 |
+
If Y is None, it is set as a pointer to X (i.e. not a copy).
|
| 76 |
+
If Y is given, this does not happen.
|
| 77 |
+
All distance metrics should use this function first to assert that the
|
| 78 |
+
given parameters are correct and safe to use.
|
| 79 |
+
|
| 80 |
+
Specifically, this function first ensures that both X and Y are arrays,
|
| 81 |
+
then checks that they are at least two dimensional while ensuring that
|
| 82 |
+
their elements are floats (or dtype if provided). Finally, the function
|
| 83 |
+
checks that the size of the second dimension of the two arrays is equal, or
|
| 84 |
+
the equivalent check for a precomputed distance matrix.
|
| 85 |
+
|
| 86 |
+
Parameters
|
| 87 |
+
----------
|
| 88 |
+
X : {array-like, sparse matrix} of shape (n_samples_X, n_features)
|
| 89 |
+
|
| 90 |
+
Y : {array-like, sparse matrix} of shape (n_samples_Y, n_features)
|
| 91 |
+
|
| 92 |
+
precomputed : bool, default=False
|
| 93 |
+
True if X is to be treated as precomputed distances to the samples in
|
| 94 |
+
Y.
|
| 95 |
+
|
| 96 |
+
dtype : str, type, list of type, default=None
|
| 97 |
+
Data type required for X and Y. If None, the dtype will be an
|
| 98 |
+
appropriate float type selected by _return_float_dtype.
|
| 99 |
+
|
| 100 |
+
.. versionadded:: 0.18
|
| 101 |
+
|
| 102 |
+
accept_sparse : str, bool or list/tuple of str, default='csr'
|
| 103 |
+
String[s] representing allowed sparse matrix formats, such as 'csc',
|
| 104 |
+
'csr', etc. If the input is sparse but not in the allowed format,
|
| 105 |
+
it will be converted to the first listed format. True allows the input
|
| 106 |
+
to be any format. False means that a sparse matrix input will
|
| 107 |
+
raise an error.
|
| 108 |
+
|
| 109 |
+
force_all_finite : bool or 'allow-nan', default=True
|
| 110 |
+
Whether to raise an error on np.inf, np.nan, pd.NA in array. The
|
| 111 |
+
possibilities are:
|
| 112 |
+
|
| 113 |
+
- True: Force all values of array to be finite.
|
| 114 |
+
- False: accepts np.inf, np.nan, pd.NA in array.
|
| 115 |
+
- 'allow-nan': accepts only np.nan and pd.NA values in array. Values
|
| 116 |
+
cannot be infinite.
|
| 117 |
+
|
| 118 |
+
.. versionadded:: 0.22
|
| 119 |
+
``force_all_finite`` accepts the string ``'allow-nan'``.
|
| 120 |
+
|
| 121 |
+
.. versionchanged:: 0.23
|
| 122 |
+
Accepts `pd.NA` and converts it into `np.nan`.
|
| 123 |
+
|
| 124 |
+
copy : bool, default=False
|
| 125 |
+
Whether a forced copy will be triggered. If copy=False, a copy might
|
| 126 |
+
be triggered by a conversion.
|
| 127 |
+
|
| 128 |
+
.. versionadded:: 0.22
|
| 129 |
+
|
| 130 |
+
Returns
|
| 131 |
+
-------
|
| 132 |
+
safe_X : {array-like, sparse matrix} of shape (n_samples_X, n_features)
|
| 133 |
+
An array equal to X, guaranteed to be a numpy array.
|
| 134 |
+
|
| 135 |
+
safe_Y : {array-like, sparse matrix} of shape (n_samples_Y, n_features)
|
| 136 |
+
An array equal to Y if Y was not None, guaranteed to be a numpy array.
|
| 137 |
+
If Y was None, safe_Y will be a pointer to X.
|
| 138 |
+
"""
|
| 139 |
+
X, Y, dtype_float = _return_float_dtype(X, Y)
|
| 140 |
+
|
| 141 |
+
estimator = "check_pairwise_arrays"
|
| 142 |
+
if dtype is None:
|
| 143 |
+
dtype = dtype_float
|
| 144 |
+
|
| 145 |
+
if Y is X or Y is None:
|
| 146 |
+
X = Y = check_array(
|
| 147 |
+
X,
|
| 148 |
+
accept_sparse=accept_sparse,
|
| 149 |
+
dtype=dtype,
|
| 150 |
+
copy=copy,
|
| 151 |
+
force_all_finite=force_all_finite,
|
| 152 |
+
estimator=estimator,
|
| 153 |
+
)
|
| 154 |
+
else:
|
| 155 |
+
X = check_array(
|
| 156 |
+
X,
|
| 157 |
+
accept_sparse=accept_sparse,
|
| 158 |
+
dtype=dtype,
|
| 159 |
+
copy=copy,
|
| 160 |
+
force_all_finite=force_all_finite,
|
| 161 |
+
estimator=estimator,
|
| 162 |
+
)
|
| 163 |
+
Y = check_array(
|
| 164 |
+
Y,
|
| 165 |
+
accept_sparse=accept_sparse,
|
| 166 |
+
dtype=dtype,
|
| 167 |
+
copy=copy,
|
| 168 |
+
force_all_finite=force_all_finite,
|
| 169 |
+
estimator=estimator,
|
| 170 |
+
)
|
| 171 |
+
|
| 172 |
+
if precomputed:
|
| 173 |
+
if X.shape[1] != Y.shape[0]:
|
| 174 |
+
raise ValueError(
|
| 175 |
+
"Precomputed metric requires shape "
|
| 176 |
+
"(n_queries, n_indexed). Got (%d, %d) "
|
| 177 |
+
"for %d indexed." % (X.shape[0], X.shape[1], Y.shape[0])
|
| 178 |
+
)
|
| 179 |
+
elif X.shape[1] != Y.shape[1]:
|
| 180 |
+
raise ValueError(
|
| 181 |
+
"Incompatible dimension for X and Y matrices: "
|
| 182 |
+
"X.shape[1] == %d while Y.shape[1] == %d" % (X.shape[1], Y.shape[1])
|
| 183 |
+
)
|
| 184 |
+
|
| 185 |
+
return X, Y
|
| 186 |
+
|
| 187 |
+
|
| 188 |
+
def check_paired_arrays(X, Y):
|
| 189 |
+
"""Set X and Y appropriately and checks inputs for paired distances.
|
| 190 |
+
|
| 191 |
+
All paired distance metrics should use this function first to assert that
|
| 192 |
+
the given parameters are correct and safe to use.
|
| 193 |
+
|
| 194 |
+
Specifically, this function first ensures that both X and Y are arrays,
|
| 195 |
+
then checks that they are at least two dimensional while ensuring that
|
| 196 |
+
their elements are floats. Finally, the function checks that the size
|
| 197 |
+
of the dimensions of the two arrays are equal.
|
| 198 |
+
|
| 199 |
+
Parameters
|
| 200 |
+
----------
|
| 201 |
+
X : {array-like, sparse matrix} of shape (n_samples_X, n_features)
|
| 202 |
+
|
| 203 |
+
Y : {array-like, sparse matrix} of shape (n_samples_Y, n_features)
|
| 204 |
+
|
| 205 |
+
Returns
|
| 206 |
+
-------
|
| 207 |
+
safe_X : {array-like, sparse matrix} of shape (n_samples_X, n_features)
|
| 208 |
+
An array equal to X, guaranteed to be a numpy array.
|
| 209 |
+
|
| 210 |
+
safe_Y : {array-like, sparse matrix} of shape (n_samples_Y, n_features)
|
| 211 |
+
An array equal to Y if Y was not None, guaranteed to be a numpy array.
|
| 212 |
+
If Y was None, safe_Y will be a pointer to X.
|
| 213 |
+
"""
|
| 214 |
+
X, Y = check_pairwise_arrays(X, Y)
|
| 215 |
+
if X.shape != Y.shape:
|
| 216 |
+
raise ValueError(
|
| 217 |
+
"X and Y should be of same shape. They were respectively %r and %r long."
|
| 218 |
+
% (X.shape, Y.shape)
|
| 219 |
+
)
|
| 220 |
+
return X, Y
|
| 221 |
+
|
| 222 |
+
|
| 223 |
+
# Pairwise distances
|
| 224 |
+
def euclidean_distances(
|
| 225 |
+
X, Y=None, *, Y_norm_squared=None, squared=False, X_norm_squared=None
|
| 226 |
+
):
|
| 227 |
+
"""
|
| 228 |
+
Compute the distance matrix between each pair from a vector array X and Y.
|
| 229 |
+
|
| 230 |
+
For efficiency reasons, the euclidean distance between a pair of row
|
| 231 |
+
vector x and y is computed as::
|
| 232 |
+
|
| 233 |
+
dist(x, y) = sqrt(dot(x, x) - 2 * dot(x, y) + dot(y, y))
|
| 234 |
+
|
| 235 |
+
This formulation has two advantages over other ways of computing distances.
|
| 236 |
+
First, it is computationally efficient when dealing with sparse data.
|
| 237 |
+
Second, if one argument varies but the other remains unchanged, then
|
| 238 |
+
`dot(x, x)` and/or `dot(y, y)` can be pre-computed.
|
| 239 |
+
|
| 240 |
+
However, this is not the most precise way of doing this computation,
|
| 241 |
+
because this equation potentially suffers from "catastrophic cancellation".
|
| 242 |
+
Also, the distance matrix returned by this function may not be exactly
|
| 243 |
+
symmetric as required by, e.g., ``scipy.spatial.distance`` functions.
|
| 244 |
+
|
| 245 |
+
Read more in the :ref:`User Guide <metrics>`.
|
| 246 |
+
|
| 247 |
+
Parameters
|
| 248 |
+
----------
|
| 249 |
+
X : {array-like, sparse matrix} of shape (n_samples_X, n_features)
|
| 250 |
+
An array where each row is a sample and each column is a feature.
|
| 251 |
+
|
| 252 |
+
Y : {array-like, sparse matrix} of shape (n_samples_Y, n_features), \
|
| 253 |
+
default=None
|
| 254 |
+
An array where each row is a sample and each column is a feature.
|
| 255 |
+
If `None`, method uses `Y=X`.
|
| 256 |
+
|
| 257 |
+
Y_norm_squared : array-like of shape (n_samples_Y,) or (n_samples_Y, 1) \
|
| 258 |
+
or (1, n_samples_Y), default=None
|
| 259 |
+
Pre-computed dot-products of vectors in Y (e.g.,
|
| 260 |
+
``(Y**2).sum(axis=1)``)
|
| 261 |
+
May be ignored in some cases, see the note below.
|
| 262 |
+
|
| 263 |
+
squared : bool, default=False
|
| 264 |
+
Return squared Euclidean distances.
|
| 265 |
+
|
| 266 |
+
X_norm_squared : array-like of shape (n_samples_X,) or (n_samples_X, 1) \
|
| 267 |
+
or (1, n_samples_X), default=None
|
| 268 |
+
Pre-computed dot-products of vectors in X (e.g.,
|
| 269 |
+
``(X**2).sum(axis=1)``)
|
| 270 |
+
May be ignored in some cases, see the note below.
|
| 271 |
+
|
| 272 |
+
Returns
|
| 273 |
+
-------
|
| 274 |
+
distances : ndarray of shape (n_samples_X, n_samples_Y)
|
| 275 |
+
Returns the distances between the row vectors of `X`
|
| 276 |
+
and the row vectors of `Y`.
|
| 277 |
+
|
| 278 |
+
See Also
|
| 279 |
+
--------
|
| 280 |
+
paired_distances : Distances between pairs of elements of X and Y.
|
| 281 |
+
|
| 282 |
+
Notes
|
| 283 |
+
-----
|
| 284 |
+
To achieve a better accuracy, `X_norm_squared` and `Y_norm_squared` may be
|
| 285 |
+
unused if they are passed as `np.float32`.
|
| 286 |
+
|
| 287 |
+
Examples
|
| 288 |
+
--------
|
| 289 |
+
>>> from sklearn.metrics.pairwise import euclidean_distances
|
| 290 |
+
>>> X = [[0, 1], [1, 1]]
|
| 291 |
+
>>> # distance between rows of X
|
| 292 |
+
>>> euclidean_distances(X, X)
|
| 293 |
+
array([[0., 1.],
|
| 294 |
+
[1., 0.]])
|
| 295 |
+
>>> # get distance to origin
|
| 296 |
+
>>> euclidean_distances(X, [[0, 0]])
|
| 297 |
+
array([[1. ],
|
| 298 |
+
[1.41421356]])
|
| 299 |
+
"""
|
| 300 |
+
X, Y = check_pairwise_arrays(X, Y)
|
| 301 |
+
|
| 302 |
+
if X_norm_squared is not None:
|
| 303 |
+
X_norm_squared = check_array(X_norm_squared, ensure_2d=False)
|
| 304 |
+
original_shape = X_norm_squared.shape
|
| 305 |
+
if X_norm_squared.shape == (X.shape[0],):
|
| 306 |
+
X_norm_squared = X_norm_squared.reshape(-1, 1)
|
| 307 |
+
if X_norm_squared.shape == (1, X.shape[0]):
|
| 308 |
+
X_norm_squared = X_norm_squared.T
|
| 309 |
+
if X_norm_squared.shape != (X.shape[0], 1):
|
| 310 |
+
raise ValueError(
|
| 311 |
+
f"Incompatible dimensions for X of shape {X.shape} and "
|
| 312 |
+
f"X_norm_squared of shape {original_shape}."
|
| 313 |
+
)
|
| 314 |
+
|
| 315 |
+
if Y_norm_squared is not None:
|
| 316 |
+
Y_norm_squared = check_array(Y_norm_squared, ensure_2d=False)
|
| 317 |
+
original_shape = Y_norm_squared.shape
|
| 318 |
+
if Y_norm_squared.shape == (Y.shape[0],):
|
| 319 |
+
Y_norm_squared = Y_norm_squared.reshape(1, -1)
|
| 320 |
+
if Y_norm_squared.shape == (Y.shape[0], 1):
|
| 321 |
+
Y_norm_squared = Y_norm_squared.T
|
| 322 |
+
if Y_norm_squared.shape != (1, Y.shape[0]):
|
| 323 |
+
raise ValueError(
|
| 324 |
+
f"Incompatible dimensions for Y of shape {Y.shape} and "
|
| 325 |
+
f"Y_norm_squared of shape {original_shape}."
|
| 326 |
+
)
|
| 327 |
+
|
| 328 |
+
return _euclidean_distances(X, Y, X_norm_squared, Y_norm_squared, squared)
|
| 329 |
+
|
| 330 |
+
|
| 331 |
+
def _euclidean_distances(X, Y, X_norm_squared=None, Y_norm_squared=None, squared=False):
|
| 332 |
+
"""Computational part of euclidean_distances
|
| 333 |
+
|
| 334 |
+
Assumes inputs are already checked.
|
| 335 |
+
|
| 336 |
+
If norms are passed as float32, they are unused. If arrays are passed as
|
| 337 |
+
float32, norms needs to be recomputed on upcast chunks.
|
| 338 |
+
TODO: use a float64 accumulator in row_norms to avoid the latter.
|
| 339 |
+
"""
|
| 340 |
+
if X_norm_squared is not None:
|
| 341 |
+
if X_norm_squared.dtype == np.float32:
|
| 342 |
+
XX = None
|
| 343 |
+
else:
|
| 344 |
+
XX = X_norm_squared.reshape(-1, 1)
|
| 345 |
+
elif X.dtype == np.float32:
|
| 346 |
+
XX = None
|
| 347 |
+
else:
|
| 348 |
+
XX = row_norms(X, squared=True)[:, np.newaxis]
|
| 349 |
+
|
| 350 |
+
if Y is X:
|
| 351 |
+
YY = None if XX is None else XX.T
|
| 352 |
+
else:
|
| 353 |
+
if Y_norm_squared is not None:
|
| 354 |
+
if Y_norm_squared.dtype == np.float32:
|
| 355 |
+
YY = None
|
| 356 |
+
else:
|
| 357 |
+
YY = Y_norm_squared.reshape(1, -1)
|
| 358 |
+
elif Y.dtype == np.float32:
|
| 359 |
+
YY = None
|
| 360 |
+
else:
|
| 361 |
+
YY = row_norms(Y, squared=True)[np.newaxis, :]
|
| 362 |
+
|
| 363 |
+
if X.dtype == np.float32:
|
| 364 |
+
# To minimize precision issues with float32, we compute the distance
|
| 365 |
+
# matrix on chunks of X and Y upcast to float64
|
| 366 |
+
distances = _euclidean_distances_upcast(X, XX, Y, YY)
|
| 367 |
+
else:
|
| 368 |
+
# if dtype is already float64, no need to chunk and upcast
|
| 369 |
+
distances = -2 * safe_sparse_dot(X, Y.T, dense_output=True)
|
| 370 |
+
distances += XX
|
| 371 |
+
distances += YY
|
| 372 |
+
np.maximum(distances, 0, out=distances)
|
| 373 |
+
|
| 374 |
+
# Ensure that distances between vectors and themselves are set to 0.0.
|
| 375 |
+
# This may not be the case due to floating point rounding errors.
|
| 376 |
+
if X is Y:
|
| 377 |
+
np.fill_diagonal(distances, 0)
|
| 378 |
+
|
| 379 |
+
return distances if squared else np.sqrt(distances, out=distances)
|
| 380 |
+
|
| 381 |
+
|
| 382 |
+
def nan_euclidean_distances(
|
| 383 |
+
X, Y=None, *, squared=False, missing_values=np.nan, copy=True
|
| 384 |
+
):
|
| 385 |
+
"""Calculate the euclidean distances in the presence of missing values.
|
| 386 |
+
|
| 387 |
+
Compute the euclidean distance between each pair of samples in X and Y,
|
| 388 |
+
where Y=X is assumed if Y=None. When calculating the distance between a
|
| 389 |
+
pair of samples, this formulation ignores feature coordinates with a
|
| 390 |
+
missing value in either sample and scales up the weight of the remaining
|
| 391 |
+
coordinates:
|
| 392 |
+
|
| 393 |
+
dist(x,y) = sqrt(weight * sq. distance from present coordinates)
|
| 394 |
+
where,
|
| 395 |
+
weight = Total # of coordinates / # of present coordinates
|
| 396 |
+
|
| 397 |
+
For example, the distance between ``[3, na, na, 6]`` and ``[1, na, 4, 5]``
|
| 398 |
+
is:
|
| 399 |
+
|
| 400 |
+
.. math::
|
| 401 |
+
\\sqrt{\\frac{4}{2}((3-1)^2 + (6-5)^2)}
|
| 402 |
+
|
| 403 |
+
If all the coordinates are missing or if there are no common present
|
| 404 |
+
coordinates then NaN is returned for that pair.
|
| 405 |
+
|
| 406 |
+
Read more in the :ref:`User Guide <metrics>`.
|
| 407 |
+
|
| 408 |
+
.. versionadded:: 0.22
|
| 409 |
+
|
| 410 |
+
Parameters
|
| 411 |
+
----------
|
| 412 |
+
X : array-like of shape (n_samples_X, n_features)
|
| 413 |
+
An array where each row is a sample and each column is a feature.
|
| 414 |
+
|
| 415 |
+
Y : array-like of shape (n_samples_Y, n_features), default=None
|
| 416 |
+
An array where each row is a sample and each column is a feature.
|
| 417 |
+
If `None`, method uses `Y=X`.
|
| 418 |
+
|
| 419 |
+
squared : bool, default=False
|
| 420 |
+
Return squared Euclidean distances.
|
| 421 |
+
|
| 422 |
+
missing_values : np.nan or int, default=np.nan
|
| 423 |
+
Representation of missing value.
|
| 424 |
+
|
| 425 |
+
copy : bool, default=True
|
| 426 |
+
Make and use a deep copy of X and Y (if Y exists).
|
| 427 |
+
|
| 428 |
+
Returns
|
| 429 |
+
-------
|
| 430 |
+
distances : ndarray of shape (n_samples_X, n_samples_Y)
|
| 431 |
+
Returns the distances between the row vectors of `X`
|
| 432 |
+
and the row vectors of `Y`.
|
| 433 |
+
|
| 434 |
+
See Also
|
| 435 |
+
--------
|
| 436 |
+
paired_distances : Distances between pairs of elements of X and Y.
|
| 437 |
+
|
| 438 |
+
References
|
| 439 |
+
----------
|
| 440 |
+
* John K. Dixon, "Pattern Recognition with Partly Missing Data",
|
| 441 |
+
IEEE Transactions on Systems, Man, and Cybernetics, Volume: 9, Issue:
|
| 442 |
+
10, pp. 617 - 621, Oct. 1979.
|
| 443 |
+
http://ieeexplore.ieee.org/abstract/document/4310090/
|
| 444 |
+
|
| 445 |
+
Examples
|
| 446 |
+
--------
|
| 447 |
+
>>> from sklearn.metrics.pairwise import nan_euclidean_distances
|
| 448 |
+
>>> nan = float("NaN")
|
| 449 |
+
>>> X = [[0, 1], [1, nan]]
|
| 450 |
+
>>> nan_euclidean_distances(X, X) # distance between rows of X
|
| 451 |
+
array([[0. , 1.41421356],
|
| 452 |
+
[1.41421356, 0. ]])
|
| 453 |
+
|
| 454 |
+
>>> # get distance to origin
|
| 455 |
+
>>> nan_euclidean_distances(X, [[0, 0]])
|
| 456 |
+
array([[1. ],
|
| 457 |
+
[1.41421356]])
|
| 458 |
+
"""
|
| 459 |
+
|
| 460 |
+
force_all_finite = "allow-nan" if is_scalar_nan(missing_values) else True
|
| 461 |
+
X, Y = check_pairwise_arrays(
|
| 462 |
+
X, Y, accept_sparse=False, force_all_finite=force_all_finite, copy=copy
|
| 463 |
+
)
|
| 464 |
+
# Get missing mask for X
|
| 465 |
+
missing_X = _get_mask(X, missing_values)
|
| 466 |
+
|
| 467 |
+
# Get missing mask for Y
|
| 468 |
+
missing_Y = missing_X if Y is X else _get_mask(Y, missing_values)
|
| 469 |
+
|
| 470 |
+
# set missing values to zero
|
| 471 |
+
X[missing_X] = 0
|
| 472 |
+
Y[missing_Y] = 0
|
| 473 |
+
|
| 474 |
+
distances = euclidean_distances(X, Y, squared=True)
|
| 475 |
+
|
| 476 |
+
# Adjust distances for missing values
|
| 477 |
+
XX = X * X
|
| 478 |
+
YY = Y * Y
|
| 479 |
+
distances -= np.dot(XX, missing_Y.T)
|
| 480 |
+
distances -= np.dot(missing_X, YY.T)
|
| 481 |
+
|
| 482 |
+
np.clip(distances, 0, None, out=distances)
|
| 483 |
+
|
| 484 |
+
if X is Y:
|
| 485 |
+
# Ensure that distances between vectors and themselves are set to 0.0.
|
| 486 |
+
# This may not be the case due to floating point rounding errors.
|
| 487 |
+
np.fill_diagonal(distances, 0.0)
|
| 488 |
+
|
| 489 |
+
present_X = 1 - missing_X
|
| 490 |
+
present_Y = present_X if Y is X else ~missing_Y
|
| 491 |
+
present_count = np.dot(present_X, present_Y.T)
|
| 492 |
+
distances[present_count == 0] = np.nan
|
| 493 |
+
# avoid divide by zero
|
| 494 |
+
np.maximum(1, present_count, out=present_count)
|
| 495 |
+
distances /= present_count
|
| 496 |
+
distances *= X.shape[1]
|
| 497 |
+
|
| 498 |
+
if not squared:
|
| 499 |
+
np.sqrt(distances, out=distances)
|
| 500 |
+
|
| 501 |
+
return distances
|
| 502 |
+
|
| 503 |
+
|
| 504 |
+
def _euclidean_distances_upcast(X, XX=None, Y=None, YY=None, batch_size=None):
|
| 505 |
+
"""Euclidean distances between X and Y.
|
| 506 |
+
|
| 507 |
+
Assumes X and Y have float32 dtype.
|
| 508 |
+
Assumes XX and YY have float64 dtype or are None.
|
| 509 |
+
|
| 510 |
+
X and Y are upcast to float64 by chunks, which size is chosen to limit
|
| 511 |
+
memory increase by approximately 10% (at least 10MiB).
|
| 512 |
+
"""
|
| 513 |
+
n_samples_X = X.shape[0]
|
| 514 |
+
n_samples_Y = Y.shape[0]
|
| 515 |
+
n_features = X.shape[1]
|
| 516 |
+
|
| 517 |
+
distances = np.empty((n_samples_X, n_samples_Y), dtype=np.float32)
|
| 518 |
+
|
| 519 |
+
if batch_size is None:
|
| 520 |
+
x_density = X.nnz / np.prod(X.shape) if issparse(X) else 1
|
| 521 |
+
y_density = Y.nnz / np.prod(Y.shape) if issparse(Y) else 1
|
| 522 |
+
|
| 523 |
+
# Allow 10% more memory than X, Y and the distance matrix take (at
|
| 524 |
+
# least 10MiB)
|
| 525 |
+
maxmem = max(
|
| 526 |
+
(
|
| 527 |
+
(x_density * n_samples_X + y_density * n_samples_Y) * n_features
|
| 528 |
+
+ (x_density * n_samples_X * y_density * n_samples_Y)
|
| 529 |
+
)
|
| 530 |
+
/ 10,
|
| 531 |
+
10 * 2**17,
|
| 532 |
+
)
|
| 533 |
+
|
| 534 |
+
# The increase amount of memory in 8-byte blocks is:
|
| 535 |
+
# - x_density * batch_size * n_features (copy of chunk of X)
|
| 536 |
+
# - y_density * batch_size * n_features (copy of chunk of Y)
|
| 537 |
+
# - batch_size * batch_size (chunk of distance matrix)
|
| 538 |
+
# Hence x² + (xd+yd)kx = M, where x=batch_size, k=n_features, M=maxmem
|
| 539 |
+
# xd=x_density and yd=y_density
|
| 540 |
+
tmp = (x_density + y_density) * n_features
|
| 541 |
+
batch_size = (-tmp + np.sqrt(tmp**2 + 4 * maxmem)) / 2
|
| 542 |
+
batch_size = max(int(batch_size), 1)
|
| 543 |
+
|
| 544 |
+
x_batches = gen_batches(n_samples_X, batch_size)
|
| 545 |
+
|
| 546 |
+
for i, x_slice in enumerate(x_batches):
|
| 547 |
+
X_chunk = X[x_slice].astype(np.float64)
|
| 548 |
+
if XX is None:
|
| 549 |
+
XX_chunk = row_norms(X_chunk, squared=True)[:, np.newaxis]
|
| 550 |
+
else:
|
| 551 |
+
XX_chunk = XX[x_slice]
|
| 552 |
+
|
| 553 |
+
y_batches = gen_batches(n_samples_Y, batch_size)
|
| 554 |
+
|
| 555 |
+
for j, y_slice in enumerate(y_batches):
|
| 556 |
+
if X is Y and j < i:
|
| 557 |
+
# when X is Y the distance matrix is symmetric so we only need
|
| 558 |
+
# to compute half of it.
|
| 559 |
+
d = distances[y_slice, x_slice].T
|
| 560 |
+
|
| 561 |
+
else:
|
| 562 |
+
Y_chunk = Y[y_slice].astype(np.float64)
|
| 563 |
+
if YY is None:
|
| 564 |
+
YY_chunk = row_norms(Y_chunk, squared=True)[np.newaxis, :]
|
| 565 |
+
else:
|
| 566 |
+
YY_chunk = YY[:, y_slice]
|
| 567 |
+
|
| 568 |
+
d = -2 * safe_sparse_dot(X_chunk, Y_chunk.T, dense_output=True)
|
| 569 |
+
d += XX_chunk
|
| 570 |
+
d += YY_chunk
|
| 571 |
+
|
| 572 |
+
distances[x_slice, y_slice] = d.astype(np.float32, copy=False)
|
| 573 |
+
|
| 574 |
+
return distances
|
| 575 |
+
|
| 576 |
+
|
| 577 |
+
def _argmin_min_reduce(dist, start):
|
| 578 |
+
# `start` is specified in the signature but not used. This is because the higher
|
| 579 |
+
# order `pairwise_distances_chunked` function needs reduction functions that are
|
| 580 |
+
# passed as argument to have a two arguments signature.
|
| 581 |
+
indices = dist.argmin(axis=1)
|
| 582 |
+
values = dist[np.arange(dist.shape[0]), indices]
|
| 583 |
+
return indices, values
|
| 584 |
+
|
| 585 |
+
|
| 586 |
+
def _argmin_reduce(dist, start):
|
| 587 |
+
# `start` is specified in the signature but not used. This is because the higher
|
| 588 |
+
# order `pairwise_distances_chunked` function needs reduction functions that are
|
| 589 |
+
# passed as argument to have a two arguments signature.
|
| 590 |
+
return dist.argmin(axis=1)
|
| 591 |
+
|
| 592 |
+
|
| 593 |
+
def pairwise_distances_argmin_min(
|
| 594 |
+
X, Y, *, axis=1, metric="euclidean", metric_kwargs=None
|
| 595 |
+
):
|
| 596 |
+
"""Compute minimum distances between one point and a set of points.
|
| 597 |
+
|
| 598 |
+
This function computes for each row in X, the index of the row of Y which
|
| 599 |
+
is closest (according to the specified distance). The minimal distances are
|
| 600 |
+
also returned.
|
| 601 |
+
|
| 602 |
+
This is mostly equivalent to calling:
|
| 603 |
+
|
| 604 |
+
(pairwise_distances(X, Y=Y, metric=metric).argmin(axis=axis),
|
| 605 |
+
pairwise_distances(X, Y=Y, metric=metric).min(axis=axis))
|
| 606 |
+
|
| 607 |
+
but uses much less memory, and is faster for large arrays.
|
| 608 |
+
|
| 609 |
+
Parameters
|
| 610 |
+
----------
|
| 611 |
+
X : {array-like, sparse matrix} of shape (n_samples_X, n_features)
|
| 612 |
+
Array containing points.
|
| 613 |
+
|
| 614 |
+
Y : {array-like, sparse matrix} of shape (n_samples_Y, n_features)
|
| 615 |
+
Array containing points.
|
| 616 |
+
|
| 617 |
+
axis : int, default=1
|
| 618 |
+
Axis along which the argmin and distances are to be computed.
|
| 619 |
+
|
| 620 |
+
metric : str or callable, default='euclidean'
|
| 621 |
+
Metric to use for distance computation. Any metric from scikit-learn
|
| 622 |
+
or scipy.spatial.distance can be used.
|
| 623 |
+
|
| 624 |
+
If metric is a callable function, it is called on each
|
| 625 |
+
pair of instances (rows) and the resulting value recorded. The callable
|
| 626 |
+
should take two arrays as input and return one value indicating the
|
| 627 |
+
distance between them. This works for Scipy's metrics, but is less
|
| 628 |
+
efficient than passing the metric name as a string.
|
| 629 |
+
|
| 630 |
+
Distance matrices are not supported.
|
| 631 |
+
|
| 632 |
+
Valid values for metric are:
|
| 633 |
+
|
| 634 |
+
- from scikit-learn: ['cityblock', 'cosine', 'euclidean', 'l1', 'l2',
|
| 635 |
+
'manhattan']
|
| 636 |
+
|
| 637 |
+
- from scipy.spatial.distance: ['braycurtis', 'canberra', 'chebyshev',
|
| 638 |
+
'correlation', 'dice', 'hamming', 'jaccard', 'kulsinski',
|
| 639 |
+
'mahalanobis', 'minkowski', 'rogerstanimoto', 'russellrao',
|
| 640 |
+
'seuclidean', 'sokalmichener', 'sokalsneath', 'sqeuclidean',
|
| 641 |
+
'yule']
|
| 642 |
+
|
| 643 |
+
See the documentation for scipy.spatial.distance for details on these
|
| 644 |
+
metrics.
|
| 645 |
+
|
| 646 |
+
metric_kwargs : dict, default=None
|
| 647 |
+
Keyword arguments to pass to specified metric function.
|
| 648 |
+
|
| 649 |
+
Returns
|
| 650 |
+
-------
|
| 651 |
+
argmin : ndarray
|
| 652 |
+
Y[argmin[i], :] is the row in Y that is closest to X[i, :].
|
| 653 |
+
|
| 654 |
+
distances : ndarray
|
| 655 |
+
The array of minimum distances. `distances[i]` is the distance between
|
| 656 |
+
the i-th row in X and the argmin[i]-th row in Y.
|
| 657 |
+
|
| 658 |
+
See Also
|
| 659 |
+
--------
|
| 660 |
+
pairwise_distances : Distances between every pair of samples of X and Y.
|
| 661 |
+
pairwise_distances_argmin : Same as `pairwise_distances_argmin_min` but only
|
| 662 |
+
returns the argmins.
|
| 663 |
+
"""
|
| 664 |
+
X, Y = check_pairwise_arrays(X, Y)
|
| 665 |
+
|
| 666 |
+
if axis == 0:
|
| 667 |
+
X, Y = Y, X
|
| 668 |
+
|
| 669 |
+
if metric_kwargs is None:
|
| 670 |
+
metric_kwargs = {}
|
| 671 |
+
|
| 672 |
+
if ArgKmin.is_usable_for(X, Y, metric):
|
| 673 |
+
# This is an adaptor for one "sqeuclidean" specification.
|
| 674 |
+
# For this backend, we can directly use "sqeuclidean".
|
| 675 |
+
if metric_kwargs.get("squared", False) and metric == "euclidean":
|
| 676 |
+
metric = "sqeuclidean"
|
| 677 |
+
metric_kwargs = {}
|
| 678 |
+
|
| 679 |
+
values, indices = ArgKmin.compute(
|
| 680 |
+
X=X,
|
| 681 |
+
Y=Y,
|
| 682 |
+
k=1,
|
| 683 |
+
metric=metric,
|
| 684 |
+
metric_kwargs=metric_kwargs,
|
| 685 |
+
strategy="auto",
|
| 686 |
+
return_distance=True,
|
| 687 |
+
)
|
| 688 |
+
values = values.flatten()
|
| 689 |
+
indices = indices.flatten()
|
| 690 |
+
else:
|
| 691 |
+
# Joblib-based backend, which is used when user-defined callable
|
| 692 |
+
# are passed for metric.
|
| 693 |
+
|
| 694 |
+
# This won't be used in the future once PairwiseDistancesReductions support:
|
| 695 |
+
# - DistanceMetrics which work on supposedly binary data
|
| 696 |
+
# - CSR-dense and dense-CSR case if 'euclidean' in metric.
|
| 697 |
+
|
| 698 |
+
# Turn off check for finiteness because this is costly and because arrays
|
| 699 |
+
# have already been validated.
|
| 700 |
+
with config_context(assume_finite=True):
|
| 701 |
+
indices, values = zip(
|
| 702 |
+
*pairwise_distances_chunked(
|
| 703 |
+
X, Y, reduce_func=_argmin_min_reduce, metric=metric, **metric_kwargs
|
| 704 |
+
)
|
| 705 |
+
)
|
| 706 |
+
indices = np.concatenate(indices)
|
| 707 |
+
values = np.concatenate(values)
|
| 708 |
+
|
| 709 |
+
return indices, values
|
| 710 |
+
|
| 711 |
+
|
| 712 |
+
def pairwise_distances_argmin(X, Y, *, axis=1, metric="euclidean", metric_kwargs=None):
|
| 713 |
+
"""Compute minimum distances between one point and a set of points.
|
| 714 |
+
|
| 715 |
+
This function computes for each row in X, the index of the row of Y which
|
| 716 |
+
is closest (according to the specified distance).
|
| 717 |
+
|
| 718 |
+
This is mostly equivalent to calling:
|
| 719 |
+
|
| 720 |
+
pairwise_distances(X, Y=Y, metric=metric).argmin(axis=axis)
|
| 721 |
+
|
| 722 |
+
but uses much less memory, and is faster for large arrays.
|
| 723 |
+
|
| 724 |
+
This function works with dense 2D arrays only.
|
| 725 |
+
|
| 726 |
+
Parameters
|
| 727 |
+
----------
|
| 728 |
+
X : {array-like, sparse matrix} of shape (n_samples_X, n_features)
|
| 729 |
+
Array containing points.
|
| 730 |
+
|
| 731 |
+
Y : {array-like, sparse matrix} of shape (n_samples_Y, n_features)
|
| 732 |
+
Arrays containing points.
|
| 733 |
+
|
| 734 |
+
axis : int, default=1
|
| 735 |
+
Axis along which the argmin and distances are to be computed.
|
| 736 |
+
|
| 737 |
+
metric : str or callable, default="euclidean"
|
| 738 |
+
Metric to use for distance computation. Any metric from scikit-learn
|
| 739 |
+
or scipy.spatial.distance can be used.
|
| 740 |
+
|
| 741 |
+
If metric is a callable function, it is called on each
|
| 742 |
+
pair of instances (rows) and the resulting value recorded. The callable
|
| 743 |
+
should take two arrays as input and return one value indicating the
|
| 744 |
+
distance between them. This works for Scipy's metrics, but is less
|
| 745 |
+
efficient than passing the metric name as a string.
|
| 746 |
+
|
| 747 |
+
Distance matrices are not supported.
|
| 748 |
+
|
| 749 |
+
Valid values for metric are:
|
| 750 |
+
|
| 751 |
+
- from scikit-learn: ['cityblock', 'cosine', 'euclidean', 'l1', 'l2',
|
| 752 |
+
'manhattan']
|
| 753 |
+
|
| 754 |
+
- from scipy.spatial.distance: ['braycurtis', 'canberra', 'chebyshev',
|
| 755 |
+
'correlation', 'dice', 'hamming', 'jaccard', 'kulsinski',
|
| 756 |
+
'mahalanobis', 'minkowski', 'rogerstanimoto', 'russellrao',
|
| 757 |
+
'seuclidean', 'sokalmichener', 'sokalsneath', 'sqeuclidean',
|
| 758 |
+
'yule']
|
| 759 |
+
|
| 760 |
+
See the documentation for scipy.spatial.distance for details on these
|
| 761 |
+
metrics.
|
| 762 |
+
|
| 763 |
+
metric_kwargs : dict, default=None
|
| 764 |
+
Keyword arguments to pass to specified metric function.
|
| 765 |
+
|
| 766 |
+
Returns
|
| 767 |
+
-------
|
| 768 |
+
argmin : numpy.ndarray
|
| 769 |
+
Y[argmin[i], :] is the row in Y that is closest to X[i, :].
|
| 770 |
+
|
| 771 |
+
See Also
|
| 772 |
+
--------
|
| 773 |
+
pairwise_distances : Distances between every pair of samples of X and Y.
|
| 774 |
+
pairwise_distances_argmin_min : Same as `pairwise_distances_argmin` but also
|
| 775 |
+
returns the distances.
|
| 776 |
+
"""
|
| 777 |
+
if metric_kwargs is None:
|
| 778 |
+
metric_kwargs = {}
|
| 779 |
+
|
| 780 |
+
X, Y = check_pairwise_arrays(X, Y)
|
| 781 |
+
|
| 782 |
+
if axis == 0:
|
| 783 |
+
X, Y = Y, X
|
| 784 |
+
|
| 785 |
+
if metric_kwargs is None:
|
| 786 |
+
metric_kwargs = {}
|
| 787 |
+
|
| 788 |
+
if ArgKmin.is_usable_for(X, Y, metric):
|
| 789 |
+
# This is an adaptor for one "sqeuclidean" specification.
|
| 790 |
+
# For this backend, we can directly use "sqeuclidean".
|
| 791 |
+
if metric_kwargs.get("squared", False) and metric == "euclidean":
|
| 792 |
+
metric = "sqeuclidean"
|
| 793 |
+
metric_kwargs = {}
|
| 794 |
+
|
| 795 |
+
indices = ArgKmin.compute(
|
| 796 |
+
X=X,
|
| 797 |
+
Y=Y,
|
| 798 |
+
k=1,
|
| 799 |
+
metric=metric,
|
| 800 |
+
metric_kwargs=metric_kwargs,
|
| 801 |
+
strategy="auto",
|
| 802 |
+
return_distance=False,
|
| 803 |
+
)
|
| 804 |
+
indices = indices.flatten()
|
| 805 |
+
else:
|
| 806 |
+
# Joblib-based backend, which is used when user-defined callable
|
| 807 |
+
# are passed for metric.
|
| 808 |
+
|
| 809 |
+
# This won't be used in the future once PairwiseDistancesReductions support:
|
| 810 |
+
# - DistanceMetrics which work on supposedly binary data
|
| 811 |
+
# - CSR-dense and dense-CSR case if 'euclidean' in metric.
|
| 812 |
+
|
| 813 |
+
# Turn off check for finiteness because this is costly and because arrays
|
| 814 |
+
# have already been validated.
|
| 815 |
+
with config_context(assume_finite=True):
|
| 816 |
+
indices = np.concatenate(
|
| 817 |
+
list(
|
| 818 |
+
# This returns a np.ndarray generator whose arrays we need
|
| 819 |
+
# to flatten into one.
|
| 820 |
+
pairwise_distances_chunked(
|
| 821 |
+
X, Y, reduce_func=_argmin_reduce, metric=metric, **metric_kwargs
|
| 822 |
+
)
|
| 823 |
+
)
|
| 824 |
+
)
|
| 825 |
+
|
| 826 |
+
return indices
|
| 827 |
+
|
| 828 |
+
|
| 829 |
+
def haversine_distances(X, Y=None):
|
| 830 |
+
"""Compute the Haversine distance between samples in X and Y.
|
| 831 |
+
|
| 832 |
+
The Haversine (or great circle) distance is the angular distance between
|
| 833 |
+
two points on the surface of a sphere. The first coordinate of each point
|
| 834 |
+
is assumed to be the latitude, the second is the longitude, given
|
| 835 |
+
in radians. The dimension of the data must be 2.
|
| 836 |
+
|
| 837 |
+
.. math::
|
| 838 |
+
D(x, y) = 2\\arcsin[\\sqrt{\\sin^2((x1 - y1) / 2)
|
| 839 |
+
+ \\cos(x1)\\cos(y1)\\sin^2((x2 - y2) / 2)}]
|
| 840 |
+
|
| 841 |
+
Parameters
|
| 842 |
+
----------
|
| 843 |
+
X : array-like of shape (n_samples_X, 2)
|
| 844 |
+
A feature array.
|
| 845 |
+
|
| 846 |
+
Y : array-like of shape (n_samples_Y, 2), default=None
|
| 847 |
+
An optional second feature array. If `None`, uses `Y=X`.
|
| 848 |
+
|
| 849 |
+
Returns
|
| 850 |
+
-------
|
| 851 |
+
distance : ndarray of shape (n_samples_X, n_samples_Y)
|
| 852 |
+
The distance matrix.
|
| 853 |
+
|
| 854 |
+
Notes
|
| 855 |
+
-----
|
| 856 |
+
As the Earth is nearly spherical, the haversine formula provides a good
|
| 857 |
+
approximation of the distance between two points of the Earth surface, with
|
| 858 |
+
a less than 1% error on average.
|
| 859 |
+
|
| 860 |
+
Examples
|
| 861 |
+
--------
|
| 862 |
+
We want to calculate the distance between the Ezeiza Airport
|
| 863 |
+
(Buenos Aires, Argentina) and the Charles de Gaulle Airport (Paris,
|
| 864 |
+
France).
|
| 865 |
+
|
| 866 |
+
>>> from sklearn.metrics.pairwise import haversine_distances
|
| 867 |
+
>>> from math import radians
|
| 868 |
+
>>> bsas = [-34.83333, -58.5166646]
|
| 869 |
+
>>> paris = [49.0083899664, 2.53844117956]
|
| 870 |
+
>>> bsas_in_radians = [radians(_) for _ in bsas]
|
| 871 |
+
>>> paris_in_radians = [radians(_) for _ in paris]
|
| 872 |
+
>>> result = haversine_distances([bsas_in_radians, paris_in_radians])
|
| 873 |
+
>>> result * 6371000/1000 # multiply by Earth radius to get kilometers
|
| 874 |
+
array([[ 0. , 11099.54035582],
|
| 875 |
+
[11099.54035582, 0. ]])
|
| 876 |
+
"""
|
| 877 |
+
from ..metrics import DistanceMetric
|
| 878 |
+
|
| 879 |
+
return DistanceMetric.get_metric("haversine").pairwise(X, Y)
|
| 880 |
+
|
| 881 |
+
|
| 882 |
+
def manhattan_distances(X, Y=None, *, sum_over_features="deprecated"):
|
| 883 |
+
"""Compute the L1 distances between the vectors in X and Y.
|
| 884 |
+
|
| 885 |
+
With sum_over_features equal to False it returns the componentwise
|
| 886 |
+
distances.
|
| 887 |
+
|
| 888 |
+
Read more in the :ref:`User Guide <metrics>`.
|
| 889 |
+
|
| 890 |
+
Parameters
|
| 891 |
+
----------
|
| 892 |
+
X : array-like of shape (n_samples_X, n_features)
|
| 893 |
+
An array where each row is a sample and each column is a feature.
|
| 894 |
+
|
| 895 |
+
Y : array-like of shape (n_samples_Y, n_features), default=None
|
| 896 |
+
An array where each row is a sample and each column is a feature.
|
| 897 |
+
If `None`, method uses `Y=X`.
|
| 898 |
+
|
| 899 |
+
sum_over_features : bool, default=True
|
| 900 |
+
If True the function returns the pairwise distance matrix
|
| 901 |
+
else it returns the componentwise L1 pairwise-distances.
|
| 902 |
+
Not supported for sparse matrix inputs.
|
| 903 |
+
|
| 904 |
+
.. deprecated:: 1.2
|
| 905 |
+
``sum_over_features`` was deprecated in version 1.2 and will be removed in
|
| 906 |
+
1.4.
|
| 907 |
+
|
| 908 |
+
Returns
|
| 909 |
+
-------
|
| 910 |
+
D : ndarray of shape (n_samples_X * n_samples_Y, n_features) or \
|
| 911 |
+
(n_samples_X, n_samples_Y)
|
| 912 |
+
If sum_over_features is False shape is
|
| 913 |
+
(n_samples_X * n_samples_Y, n_features) and D contains the
|
| 914 |
+
componentwise L1 pairwise-distances (ie. absolute difference),
|
| 915 |
+
else shape is (n_samples_X, n_samples_Y) and D contains
|
| 916 |
+
the pairwise L1 distances.
|
| 917 |
+
|
| 918 |
+
Notes
|
| 919 |
+
-----
|
| 920 |
+
When X and/or Y are CSR sparse matrices and they are not already
|
| 921 |
+
in canonical format, this function modifies them in-place to
|
| 922 |
+
make them canonical.
|
| 923 |
+
|
| 924 |
+
Examples
|
| 925 |
+
--------
|
| 926 |
+
>>> from sklearn.metrics.pairwise import manhattan_distances
|
| 927 |
+
>>> manhattan_distances([[3]], [[3]])
|
| 928 |
+
array([[0.]])
|
| 929 |
+
>>> manhattan_distances([[3]], [[2]])
|
| 930 |
+
array([[1.]])
|
| 931 |
+
>>> manhattan_distances([[2]], [[3]])
|
| 932 |
+
array([[1.]])
|
| 933 |
+
>>> manhattan_distances([[1, 2], [3, 4]],\
|
| 934 |
+
[[1, 2], [0, 3]])
|
| 935 |
+
array([[0., 2.],
|
| 936 |
+
[4., 4.]])
|
| 937 |
+
"""
|
| 938 |
+
# TODO(1.4): remove sum_over_features
|
| 939 |
+
if sum_over_features != "deprecated":
|
| 940 |
+
warnings.warn(
|
| 941 |
+
"`sum_over_features` is deprecated in version 1.2 and will be"
|
| 942 |
+
" removed in version 1.4.",
|
| 943 |
+
FutureWarning,
|
| 944 |
+
)
|
| 945 |
+
else:
|
| 946 |
+
sum_over_features = True
|
| 947 |
+
|
| 948 |
+
X, Y = check_pairwise_arrays(X, Y)
|
| 949 |
+
|
| 950 |
+
if issparse(X) or issparse(Y):
|
| 951 |
+
if not sum_over_features:
|
| 952 |
+
raise TypeError(
|
| 953 |
+
"sum_over_features=%r not supported for sparse matrices"
|
| 954 |
+
% sum_over_features
|
| 955 |
+
)
|
| 956 |
+
|
| 957 |
+
X = csr_matrix(X, copy=False)
|
| 958 |
+
Y = csr_matrix(Y, copy=False)
|
| 959 |
+
X.sum_duplicates() # this also sorts indices in-place
|
| 960 |
+
Y.sum_duplicates()
|
| 961 |
+
D = np.zeros((X.shape[0], Y.shape[0]))
|
| 962 |
+
_sparse_manhattan(X.data, X.indices, X.indptr, Y.data, Y.indices, Y.indptr, D)
|
| 963 |
+
return D
|
| 964 |
+
|
| 965 |
+
if sum_over_features:
|
| 966 |
+
return distance.cdist(X, Y, "cityblock")
|
| 967 |
+
|
| 968 |
+
D = X[:, np.newaxis, :] - Y[np.newaxis, :, :]
|
| 969 |
+
D = np.abs(D, D)
|
| 970 |
+
return D.reshape((-1, X.shape[1]))
|
| 971 |
+
|
| 972 |
+
|
| 973 |
+
def cosine_distances(X, Y=None):
|
| 974 |
+
"""Compute cosine distance between samples in X and Y.
|
| 975 |
+
|
| 976 |
+
Cosine distance is defined as 1.0 minus the cosine similarity.
|
| 977 |
+
|
| 978 |
+
Read more in the :ref:`User Guide <metrics>`.
|
| 979 |
+
|
| 980 |
+
Parameters
|
| 981 |
+
----------
|
| 982 |
+
X : {array-like, sparse matrix} of shape (n_samples_X, n_features)
|
| 983 |
+
Matrix `X`.
|
| 984 |
+
|
| 985 |
+
Y : {array-like, sparse matrix} of shape (n_samples_Y, n_features), \
|
| 986 |
+
default=None
|
| 987 |
+
Matrix `Y`.
|
| 988 |
+
|
| 989 |
+
Returns
|
| 990 |
+
-------
|
| 991 |
+
distance matrix : ndarray of shape (n_samples_X, n_samples_Y)
|
| 992 |
+
Returns the cosine distance between samples in X and Y.
|
| 993 |
+
|
| 994 |
+
See Also
|
| 995 |
+
--------
|
| 996 |
+
cosine_similarity : Compute cosine similarity between samples in X and Y.
|
| 997 |
+
scipy.spatial.distance.cosine : Dense matrices only.
|
| 998 |
+
"""
|
| 999 |
+
# 1.0 - cosine_similarity(X, Y) without copy
|
| 1000 |
+
S = cosine_similarity(X, Y)
|
| 1001 |
+
S *= -1
|
| 1002 |
+
S += 1
|
| 1003 |
+
np.clip(S, 0, 2, out=S)
|
| 1004 |
+
if X is Y or Y is None:
|
| 1005 |
+
# Ensure that distances between vectors and themselves are set to 0.0.
|
| 1006 |
+
# This may not be the case due to floating point rounding errors.
|
| 1007 |
+
S[np.diag_indices_from(S)] = 0.0
|
| 1008 |
+
return S
|
| 1009 |
+
|
| 1010 |
+
|
| 1011 |
+
# Paired distances
|
| 1012 |
+
def paired_euclidean_distances(X, Y):
|
| 1013 |
+
"""Compute the paired euclidean distances between X and Y.
|
| 1014 |
+
|
| 1015 |
+
Read more in the :ref:`User Guide <metrics>`.
|
| 1016 |
+
|
| 1017 |
+
Parameters
|
| 1018 |
+
----------
|
| 1019 |
+
X : array-like of shape (n_samples, n_features)
|
| 1020 |
+
Input array/matrix X.
|
| 1021 |
+
|
| 1022 |
+
Y : array-like of shape (n_samples, n_features)
|
| 1023 |
+
Input array/matrix Y.
|
| 1024 |
+
|
| 1025 |
+
Returns
|
| 1026 |
+
-------
|
| 1027 |
+
distances : ndarray of shape (n_samples,)
|
| 1028 |
+
Output array/matrix containing the calculated paired euclidean
|
| 1029 |
+
distances.
|
| 1030 |
+
"""
|
| 1031 |
+
X, Y = check_paired_arrays(X, Y)
|
| 1032 |
+
return row_norms(X - Y)
|
| 1033 |
+
|
| 1034 |
+
|
| 1035 |
+
def paired_manhattan_distances(X, Y):
|
| 1036 |
+
"""Compute the paired L1 distances between X and Y.
|
| 1037 |
+
|
| 1038 |
+
Distances are calculated between (X[0], Y[0]), (X[1], Y[1]), ...,
|
| 1039 |
+
(X[n_samples], Y[n_samples]).
|
| 1040 |
+
|
| 1041 |
+
Read more in the :ref:`User Guide <metrics>`.
|
| 1042 |
+
|
| 1043 |
+
Parameters
|
| 1044 |
+
----------
|
| 1045 |
+
X : array-like of shape (n_samples, n_features)
|
| 1046 |
+
An array-like where each row is a sample and each column is a feature.
|
| 1047 |
+
|
| 1048 |
+
Y : array-like of shape (n_samples, n_features)
|
| 1049 |
+
An array-like where each row is a sample and each column is a feature.
|
| 1050 |
+
|
| 1051 |
+
Returns
|
| 1052 |
+
-------
|
| 1053 |
+
distances : ndarray of shape (n_samples,)
|
| 1054 |
+
L1 paired distances between the row vectors of `X`
|
| 1055 |
+
and the row vectors of `Y`.
|
| 1056 |
+
|
| 1057 |
+
Examples
|
| 1058 |
+
--------
|
| 1059 |
+
>>> from sklearn.metrics.pairwise import paired_manhattan_distances
|
| 1060 |
+
>>> import numpy as np
|
| 1061 |
+
>>> X = np.array([[1, 1, 0], [0, 1, 0], [0, 0, 1]])
|
| 1062 |
+
>>> Y = np.array([[0, 1, 0], [0, 0, 1], [0, 0, 0]])
|
| 1063 |
+
>>> paired_manhattan_distances(X, Y)
|
| 1064 |
+
array([1., 2., 1.])
|
| 1065 |
+
"""
|
| 1066 |
+
X, Y = check_paired_arrays(X, Y)
|
| 1067 |
+
diff = X - Y
|
| 1068 |
+
if issparse(diff):
|
| 1069 |
+
diff.data = np.abs(diff.data)
|
| 1070 |
+
return np.squeeze(np.array(diff.sum(axis=1)))
|
| 1071 |
+
else:
|
| 1072 |
+
return np.abs(diff).sum(axis=-1)
|
| 1073 |
+
|
| 1074 |
+
|
| 1075 |
+
def paired_cosine_distances(X, Y):
|
| 1076 |
+
"""
|
| 1077 |
+
Compute the paired cosine distances between X and Y.
|
| 1078 |
+
|
| 1079 |
+
Read more in the :ref:`User Guide <metrics>`.
|
| 1080 |
+
|
| 1081 |
+
Parameters
|
| 1082 |
+
----------
|
| 1083 |
+
X : array-like of shape (n_samples, n_features)
|
| 1084 |
+
An array where each row is a sample and each column is a feature.
|
| 1085 |
+
|
| 1086 |
+
Y : array-like of shape (n_samples, n_features)
|
| 1087 |
+
An array where each row is a sample and each column is a feature.
|
| 1088 |
+
|
| 1089 |
+
Returns
|
| 1090 |
+
-------
|
| 1091 |
+
distances : ndarray of shape (n_samples,)
|
| 1092 |
+
Returns the distances between the row vectors of `X`
|
| 1093 |
+
and the row vectors of `Y`, where `distances[i]` is the
|
| 1094 |
+
distance between `X[i]` and `Y[i]`.
|
| 1095 |
+
|
| 1096 |
+
Notes
|
| 1097 |
+
-----
|
| 1098 |
+
The cosine distance is equivalent to the half the squared
|
| 1099 |
+
euclidean distance if each sample is normalized to unit norm.
|
| 1100 |
+
"""
|
| 1101 |
+
X, Y = check_paired_arrays(X, Y)
|
| 1102 |
+
return 0.5 * row_norms(normalize(X) - normalize(Y), squared=True)
|
| 1103 |
+
|
| 1104 |
+
|
| 1105 |
+
PAIRED_DISTANCES = {
|
| 1106 |
+
"cosine": paired_cosine_distances,
|
| 1107 |
+
"euclidean": paired_euclidean_distances,
|
| 1108 |
+
"l2": paired_euclidean_distances,
|
| 1109 |
+
"l1": paired_manhattan_distances,
|
| 1110 |
+
"manhattan": paired_manhattan_distances,
|
| 1111 |
+
"cityblock": paired_manhattan_distances,
|
| 1112 |
+
}
|
| 1113 |
+
|
| 1114 |
+
|
| 1115 |
+
def paired_distances(X, Y, *, metric="euclidean", **kwds):
|
| 1116 |
+
"""
|
| 1117 |
+
Compute the paired distances between X and Y.
|
| 1118 |
+
|
| 1119 |
+
Compute the distances between (X[0], Y[0]), (X[1], Y[1]), etc...
|
| 1120 |
+
|
| 1121 |
+
Read more in the :ref:`User Guide <metrics>`.
|
| 1122 |
+
|
| 1123 |
+
Parameters
|
| 1124 |
+
----------
|
| 1125 |
+
X : ndarray of shape (n_samples, n_features)
|
| 1126 |
+
Array 1 for distance computation.
|
| 1127 |
+
|
| 1128 |
+
Y : ndarray of shape (n_samples, n_features)
|
| 1129 |
+
Array 2 for distance computation.
|
| 1130 |
+
|
| 1131 |
+
metric : str or callable, default="euclidean"
|
| 1132 |
+
The metric to use when calculating distance between instances in a
|
| 1133 |
+
feature array. If metric is a string, it must be one of the options
|
| 1134 |
+
specified in PAIRED_DISTANCES, including "euclidean",
|
| 1135 |
+
"manhattan", or "cosine".
|
| 1136 |
+
Alternatively, if metric is a callable function, it is called on each
|
| 1137 |
+
pair of instances (rows) and the resulting value recorded. The callable
|
| 1138 |
+
should take two arrays from `X` as input and return a value indicating
|
| 1139 |
+
the distance between them.
|
| 1140 |
+
|
| 1141 |
+
**kwds : dict
|
| 1142 |
+
Unused parameters.
|
| 1143 |
+
|
| 1144 |
+
Returns
|
| 1145 |
+
-------
|
| 1146 |
+
distances : ndarray of shape (n_samples,)
|
| 1147 |
+
Returns the distances between the row vectors of `X`
|
| 1148 |
+
and the row vectors of `Y`.
|
| 1149 |
+
|
| 1150 |
+
See Also
|
| 1151 |
+
--------
|
| 1152 |
+
pairwise_distances : Computes the distance between every pair of samples.
|
| 1153 |
+
|
| 1154 |
+
Examples
|
| 1155 |
+
--------
|
| 1156 |
+
>>> from sklearn.metrics.pairwise import paired_distances
|
| 1157 |
+
>>> X = [[0, 1], [1, 1]]
|
| 1158 |
+
>>> Y = [[0, 1], [2, 1]]
|
| 1159 |
+
>>> paired_distances(X, Y)
|
| 1160 |
+
array([0., 1.])
|
| 1161 |
+
"""
|
| 1162 |
+
|
| 1163 |
+
if metric in PAIRED_DISTANCES:
|
| 1164 |
+
func = PAIRED_DISTANCES[metric]
|
| 1165 |
+
return func(X, Y)
|
| 1166 |
+
elif callable(metric):
|
| 1167 |
+
# Check the matrix first (it is usually done by the metric)
|
| 1168 |
+
X, Y = check_paired_arrays(X, Y)
|
| 1169 |
+
distances = np.zeros(len(X))
|
| 1170 |
+
for i in range(len(X)):
|
| 1171 |
+
distances[i] = metric(X[i], Y[i])
|
| 1172 |
+
return distances
|
| 1173 |
+
else:
|
| 1174 |
+
raise ValueError("Unknown distance %s" % metric)
|
| 1175 |
+
|
| 1176 |
+
|
| 1177 |
+
# Kernels
|
| 1178 |
+
def linear_kernel(X, Y=None, dense_output=True):
|
| 1179 |
+
"""
|
| 1180 |
+
Compute the linear kernel between X and Y.
|
| 1181 |
+
|
| 1182 |
+
Read more in the :ref:`User Guide <linear_kernel>`.
|
| 1183 |
+
|
| 1184 |
+
Parameters
|
| 1185 |
+
----------
|
| 1186 |
+
X : ndarray of shape (n_samples_X, n_features)
|
| 1187 |
+
A feature array.
|
| 1188 |
+
|
| 1189 |
+
Y : ndarray of shape (n_samples_Y, n_features), default=None
|
| 1190 |
+
An optional second feature array. If `None`, uses `Y=X`.
|
| 1191 |
+
|
| 1192 |
+
dense_output : bool, default=True
|
| 1193 |
+
Whether to return dense output even when the input is sparse. If
|
| 1194 |
+
``False``, the output is sparse if both input arrays are sparse.
|
| 1195 |
+
|
| 1196 |
+
.. versionadded:: 0.20
|
| 1197 |
+
|
| 1198 |
+
Returns
|
| 1199 |
+
-------
|
| 1200 |
+
Gram matrix : ndarray of shape (n_samples_X, n_samples_Y)
|
| 1201 |
+
The Gram matrix of the linear kernel, i.e. `X @ Y.T`.
|
| 1202 |
+
"""
|
| 1203 |
+
X, Y = check_pairwise_arrays(X, Y)
|
| 1204 |
+
return safe_sparse_dot(X, Y.T, dense_output=dense_output)
|
| 1205 |
+
|
| 1206 |
+
|
| 1207 |
+
def polynomial_kernel(X, Y=None, degree=3, gamma=None, coef0=1):
|
| 1208 |
+
"""
|
| 1209 |
+
Compute the polynomial kernel between X and Y.
|
| 1210 |
+
|
| 1211 |
+
:math:`K(X, Y) = (gamma <X, Y> + coef0)^{degree}`
|
| 1212 |
+
|
| 1213 |
+
Read more in the :ref:`User Guide <polynomial_kernel>`.
|
| 1214 |
+
|
| 1215 |
+
Parameters
|
| 1216 |
+
----------
|
| 1217 |
+
X : ndarray of shape (n_samples_X, n_features)
|
| 1218 |
+
A feature array.
|
| 1219 |
+
|
| 1220 |
+
Y : ndarray of shape (n_samples_Y, n_features), default=None
|
| 1221 |
+
An optional second feature array. If `None`, uses `Y=X`.
|
| 1222 |
+
|
| 1223 |
+
degree : int, default=3
|
| 1224 |
+
Kernel degree.
|
| 1225 |
+
|
| 1226 |
+
gamma : float, default=None
|
| 1227 |
+
Coefficient of the vector inner product. If None, defaults to 1.0 / n_features.
|
| 1228 |
+
|
| 1229 |
+
coef0 : float, default=1
|
| 1230 |
+
Constant offset added to scaled inner product.
|
| 1231 |
+
|
| 1232 |
+
Returns
|
| 1233 |
+
-------
|
| 1234 |
+
Gram matrix : ndarray of shape (n_samples_X, n_samples_Y)
|
| 1235 |
+
The polynomial kernel.
|
| 1236 |
+
"""
|
| 1237 |
+
X, Y = check_pairwise_arrays(X, Y)
|
| 1238 |
+
if gamma is None:
|
| 1239 |
+
gamma = 1.0 / X.shape[1]
|
| 1240 |
+
|
| 1241 |
+
K = safe_sparse_dot(X, Y.T, dense_output=True)
|
| 1242 |
+
K *= gamma
|
| 1243 |
+
K += coef0
|
| 1244 |
+
K **= degree
|
| 1245 |
+
return K
|
| 1246 |
+
|
| 1247 |
+
|
| 1248 |
+
def sigmoid_kernel(X, Y=None, gamma=None, coef0=1):
|
| 1249 |
+
"""Compute the sigmoid kernel between X and Y.
|
| 1250 |
+
|
| 1251 |
+
K(X, Y) = tanh(gamma <X, Y> + coef0)
|
| 1252 |
+
|
| 1253 |
+
Read more in the :ref:`User Guide <sigmoid_kernel>`.
|
| 1254 |
+
|
| 1255 |
+
Parameters
|
| 1256 |
+
----------
|
| 1257 |
+
X : ndarray of shape (n_samples_X, n_features)
|
| 1258 |
+
A feature array.
|
| 1259 |
+
|
| 1260 |
+
Y : ndarray of shape (n_samples_Y, n_features), default=None
|
| 1261 |
+
An optional second feature array. If `None`, uses `Y=X`.
|
| 1262 |
+
|
| 1263 |
+
gamma : float, default=None
|
| 1264 |
+
Coefficient of the vector inner product. If None, defaults to 1.0 / n_features.
|
| 1265 |
+
|
| 1266 |
+
coef0 : float, default=1
|
| 1267 |
+
Constant offset added to scaled inner product.
|
| 1268 |
+
|
| 1269 |
+
Returns
|
| 1270 |
+
-------
|
| 1271 |
+
Gram matrix : ndarray of shape (n_samples_X, n_samples_Y)
|
| 1272 |
+
Sigmoid kernel between two arrays.
|
| 1273 |
+
"""
|
| 1274 |
+
X, Y = check_pairwise_arrays(X, Y)
|
| 1275 |
+
if gamma is None:
|
| 1276 |
+
gamma = 1.0 / X.shape[1]
|
| 1277 |
+
|
| 1278 |
+
K = safe_sparse_dot(X, Y.T, dense_output=True)
|
| 1279 |
+
K *= gamma
|
| 1280 |
+
K += coef0
|
| 1281 |
+
np.tanh(K, K) # compute tanh in-place
|
| 1282 |
+
return K
|
| 1283 |
+
|
| 1284 |
+
|
| 1285 |
+
def rbf_kernel(X, Y=None, gamma=None):
|
| 1286 |
+
"""Compute the rbf (gaussian) kernel between X and Y.
|
| 1287 |
+
|
| 1288 |
+
K(x, y) = exp(-gamma ||x-y||^2)
|
| 1289 |
+
|
| 1290 |
+
for each pair of rows x in X and y in Y.
|
| 1291 |
+
|
| 1292 |
+
Read more in the :ref:`User Guide <rbf_kernel>`.
|
| 1293 |
+
|
| 1294 |
+
Parameters
|
| 1295 |
+
----------
|
| 1296 |
+
X : ndarray of shape (n_samples_X, n_features)
|
| 1297 |
+
A feature array.
|
| 1298 |
+
|
| 1299 |
+
Y : ndarray of shape (n_samples_Y, n_features), default=None
|
| 1300 |
+
An optional second feature array. If `None`, uses `Y=X`.
|
| 1301 |
+
|
| 1302 |
+
gamma : float, default=None
|
| 1303 |
+
If None, defaults to 1.0 / n_features.
|
| 1304 |
+
|
| 1305 |
+
Returns
|
| 1306 |
+
-------
|
| 1307 |
+
kernel_matrix : ndarray of shape (n_samples_X, n_samples_Y)
|
| 1308 |
+
The RBF kernel.
|
| 1309 |
+
"""
|
| 1310 |
+
X, Y = check_pairwise_arrays(X, Y)
|
| 1311 |
+
if gamma is None:
|
| 1312 |
+
gamma = 1.0 / X.shape[1]
|
| 1313 |
+
|
| 1314 |
+
K = euclidean_distances(X, Y, squared=True)
|
| 1315 |
+
K *= -gamma
|
| 1316 |
+
np.exp(K, K) # exponentiate K in-place
|
| 1317 |
+
return K
|
| 1318 |
+
|
| 1319 |
+
|
| 1320 |
+
def laplacian_kernel(X, Y=None, gamma=None):
|
| 1321 |
+
"""Compute the laplacian kernel between X and Y.
|
| 1322 |
+
|
| 1323 |
+
The laplacian kernel is defined as::
|
| 1324 |
+
|
| 1325 |
+
K(x, y) = exp(-gamma ||x-y||_1)
|
| 1326 |
+
|
| 1327 |
+
for each pair of rows x in X and y in Y.
|
| 1328 |
+
Read more in the :ref:`User Guide <laplacian_kernel>`.
|
| 1329 |
+
|
| 1330 |
+
.. versionadded:: 0.17
|
| 1331 |
+
|
| 1332 |
+
Parameters
|
| 1333 |
+
----------
|
| 1334 |
+
X : ndarray of shape (n_samples_X, n_features)
|
| 1335 |
+
A feature array.
|
| 1336 |
+
|
| 1337 |
+
Y : ndarray of shape (n_samples_Y, n_features), default=None
|
| 1338 |
+
An optional second feature array. If `None`, uses `Y=X`.
|
| 1339 |
+
|
| 1340 |
+
gamma : float, default=None
|
| 1341 |
+
If None, defaults to 1.0 / n_features.
|
| 1342 |
+
|
| 1343 |
+
Returns
|
| 1344 |
+
-------
|
| 1345 |
+
kernel_matrix : ndarray of shape (n_samples_X, n_samples_Y)
|
| 1346 |
+
The kernel matrix.
|
| 1347 |
+
"""
|
| 1348 |
+
X, Y = check_pairwise_arrays(X, Y)
|
| 1349 |
+
if gamma is None:
|
| 1350 |
+
gamma = 1.0 / X.shape[1]
|
| 1351 |
+
|
| 1352 |
+
K = -gamma * manhattan_distances(X, Y)
|
| 1353 |
+
np.exp(K, K) # exponentiate K in-place
|
| 1354 |
+
return K
|
| 1355 |
+
|
| 1356 |
+
|
| 1357 |
+
def cosine_similarity(X, Y=None, dense_output=True):
|
| 1358 |
+
"""Compute cosine similarity between samples in X and Y.
|
| 1359 |
+
|
| 1360 |
+
Cosine similarity, or the cosine kernel, computes similarity as the
|
| 1361 |
+
normalized dot product of X and Y:
|
| 1362 |
+
|
| 1363 |
+
K(X, Y) = <X, Y> / (||X||*||Y||)
|
| 1364 |
+
|
| 1365 |
+
On L2-normalized data, this function is equivalent to linear_kernel.
|
| 1366 |
+
|
| 1367 |
+
Read more in the :ref:`User Guide <cosine_similarity>`.
|
| 1368 |
+
|
| 1369 |
+
Parameters
|
| 1370 |
+
----------
|
| 1371 |
+
X : {ndarray, sparse matrix} of shape (n_samples_X, n_features)
|
| 1372 |
+
Input data.
|
| 1373 |
+
|
| 1374 |
+
Y : {ndarray, sparse matrix} of shape (n_samples_Y, n_features), \
|
| 1375 |
+
default=None
|
| 1376 |
+
Input data. If ``None``, the output will be the pairwise
|
| 1377 |
+
similarities between all samples in ``X``.
|
| 1378 |
+
|
| 1379 |
+
dense_output : bool, default=True
|
| 1380 |
+
Whether to return dense output even when the input is sparse. If
|
| 1381 |
+
``False``, the output is sparse if both input arrays are sparse.
|
| 1382 |
+
|
| 1383 |
+
.. versionadded:: 0.17
|
| 1384 |
+
parameter ``dense_output`` for dense output.
|
| 1385 |
+
|
| 1386 |
+
Returns
|
| 1387 |
+
-------
|
| 1388 |
+
kernel matrix : ndarray of shape (n_samples_X, n_samples_Y)
|
| 1389 |
+
Returns the cosine similarity between samples in X and Y.
|
| 1390 |
+
"""
|
| 1391 |
+
# to avoid recursive import
|
| 1392 |
+
|
| 1393 |
+
X, Y = check_pairwise_arrays(X, Y)
|
| 1394 |
+
|
| 1395 |
+
X_normalized = normalize(X, copy=True)
|
| 1396 |
+
if X is Y:
|
| 1397 |
+
Y_normalized = X_normalized
|
| 1398 |
+
else:
|
| 1399 |
+
Y_normalized = normalize(Y, copy=True)
|
| 1400 |
+
|
| 1401 |
+
K = safe_sparse_dot(X_normalized, Y_normalized.T, dense_output=dense_output)
|
| 1402 |
+
|
| 1403 |
+
return K
|
| 1404 |
+
|
| 1405 |
+
|
| 1406 |
+
def additive_chi2_kernel(X, Y=None):
|
| 1407 |
+
"""Compute the additive chi-squared kernel between observations in X and Y.
|
| 1408 |
+
|
| 1409 |
+
The chi-squared kernel is computed between each pair of rows in X and Y. X
|
| 1410 |
+
and Y have to be non-negative. This kernel is most commonly applied to
|
| 1411 |
+
histograms.
|
| 1412 |
+
|
| 1413 |
+
The chi-squared kernel is given by::
|
| 1414 |
+
|
| 1415 |
+
k(x, y) = -Sum [(x - y)^2 / (x + y)]
|
| 1416 |
+
|
| 1417 |
+
It can be interpreted as a weighted difference per entry.
|
| 1418 |
+
|
| 1419 |
+
Read more in the :ref:`User Guide <chi2_kernel>`.
|
| 1420 |
+
|
| 1421 |
+
Parameters
|
| 1422 |
+
----------
|
| 1423 |
+
X : array-like of shape (n_samples_X, n_features)
|
| 1424 |
+
A feature array.
|
| 1425 |
+
|
| 1426 |
+
Y : ndarray of shape (n_samples_Y, n_features), default=None
|
| 1427 |
+
An optional second feature array. If `None`, uses `Y=X`.
|
| 1428 |
+
|
| 1429 |
+
Returns
|
| 1430 |
+
-------
|
| 1431 |
+
kernel_matrix : ndarray of shape (n_samples_X, n_samples_Y)
|
| 1432 |
+
The kernel matrix.
|
| 1433 |
+
|
| 1434 |
+
See Also
|
| 1435 |
+
--------
|
| 1436 |
+
chi2_kernel : The exponentiated version of the kernel, which is usually
|
| 1437 |
+
preferable.
|
| 1438 |
+
sklearn.kernel_approximation.AdditiveChi2Sampler : A Fourier approximation
|
| 1439 |
+
to this kernel.
|
| 1440 |
+
|
| 1441 |
+
Notes
|
| 1442 |
+
-----
|
| 1443 |
+
As the negative of a distance, this kernel is only conditionally positive
|
| 1444 |
+
definite.
|
| 1445 |
+
|
| 1446 |
+
References
|
| 1447 |
+
----------
|
| 1448 |
+
* Zhang, J. and Marszalek, M. and Lazebnik, S. and Schmid, C.
|
| 1449 |
+
Local features and kernels for classification of texture and object
|
| 1450 |
+
categories: A comprehensive study
|
| 1451 |
+
International Journal of Computer Vision 2007
|
| 1452 |
+
https://hal.archives-ouvertes.fr/hal-00171412/document
|
| 1453 |
+
"""
|
| 1454 |
+
if issparse(X) or issparse(Y):
|
| 1455 |
+
raise ValueError("additive_chi2 does not support sparse matrices.")
|
| 1456 |
+
X, Y = check_pairwise_arrays(X, Y)
|
| 1457 |
+
if (X < 0).any():
|
| 1458 |
+
raise ValueError("X contains negative values.")
|
| 1459 |
+
if Y is not X and (Y < 0).any():
|
| 1460 |
+
raise ValueError("Y contains negative values.")
|
| 1461 |
+
|
| 1462 |
+
result = np.zeros((X.shape[0], Y.shape[0]), dtype=X.dtype)
|
| 1463 |
+
_chi2_kernel_fast(X, Y, result)
|
| 1464 |
+
return result
|
| 1465 |
+
|
| 1466 |
+
|
| 1467 |
+
def chi2_kernel(X, Y=None, gamma=1.0):
|
| 1468 |
+
"""Compute the exponential chi-squared kernel between X and Y.
|
| 1469 |
+
|
| 1470 |
+
The chi-squared kernel is computed between each pair of rows in X and Y. X
|
| 1471 |
+
and Y have to be non-negative. This kernel is most commonly applied to
|
| 1472 |
+
histograms.
|
| 1473 |
+
|
| 1474 |
+
The chi-squared kernel is given by::
|
| 1475 |
+
|
| 1476 |
+
k(x, y) = exp(-gamma Sum [(x - y)^2 / (x + y)])
|
| 1477 |
+
|
| 1478 |
+
It can be interpreted as a weighted difference per entry.
|
| 1479 |
+
|
| 1480 |
+
Read more in the :ref:`User Guide <chi2_kernel>`.
|
| 1481 |
+
|
| 1482 |
+
Parameters
|
| 1483 |
+
----------
|
| 1484 |
+
X : array-like of shape (n_samples_X, n_features)
|
| 1485 |
+
A feature array.
|
| 1486 |
+
|
| 1487 |
+
Y : ndarray of shape (n_samples_Y, n_features), default=None
|
| 1488 |
+
An optional second feature array. If `None`, uses `Y=X`.
|
| 1489 |
+
|
| 1490 |
+
gamma : float, default=1
|
| 1491 |
+
Scaling parameter of the chi2 kernel.
|
| 1492 |
+
|
| 1493 |
+
Returns
|
| 1494 |
+
-------
|
| 1495 |
+
kernel_matrix : ndarray of shape (n_samples_X, n_samples_Y)
|
| 1496 |
+
The kernel matrix.
|
| 1497 |
+
|
| 1498 |
+
See Also
|
| 1499 |
+
--------
|
| 1500 |
+
additive_chi2_kernel : The additive version of this kernel.
|
| 1501 |
+
sklearn.kernel_approximation.AdditiveChi2Sampler : A Fourier approximation
|
| 1502 |
+
to the additive version of this kernel.
|
| 1503 |
+
|
| 1504 |
+
References
|
| 1505 |
+
----------
|
| 1506 |
+
* Zhang, J. and Marszalek, M. and Lazebnik, S. and Schmid, C.
|
| 1507 |
+
Local features and kernels for classification of texture and object
|
| 1508 |
+
categories: A comprehensive study
|
| 1509 |
+
International Journal of Computer Vision 2007
|
| 1510 |
+
https://hal.archives-ouvertes.fr/hal-00171412/document
|
| 1511 |
+
"""
|
| 1512 |
+
K = additive_chi2_kernel(X, Y)
|
| 1513 |
+
K *= gamma
|
| 1514 |
+
return np.exp(K, K)
|
| 1515 |
+
|
| 1516 |
+
|
| 1517 |
+
# Helper functions - distance
|
| 1518 |
+
PAIRWISE_DISTANCE_FUNCTIONS = {
|
| 1519 |
+
# If updating this dictionary, update the doc in both distance_metrics()
|
| 1520 |
+
# and also in pairwise_distances()!
|
| 1521 |
+
"cityblock": manhattan_distances,
|
| 1522 |
+
"cosine": cosine_distances,
|
| 1523 |
+
"euclidean": euclidean_distances,
|
| 1524 |
+
"haversine": haversine_distances,
|
| 1525 |
+
"l2": euclidean_distances,
|
| 1526 |
+
"l1": manhattan_distances,
|
| 1527 |
+
"manhattan": manhattan_distances,
|
| 1528 |
+
"precomputed": None, # HACK: precomputed is always allowed, never called
|
| 1529 |
+
"nan_euclidean": nan_euclidean_distances,
|
| 1530 |
+
}
|
| 1531 |
+
|
| 1532 |
+
|
| 1533 |
+
def distance_metrics():
|
| 1534 |
+
"""Valid metrics for pairwise_distances.
|
| 1535 |
+
|
| 1536 |
+
This function simply returns the valid pairwise distance metrics.
|
| 1537 |
+
It exists to allow for a description of the mapping for
|
| 1538 |
+
each of the valid strings.
|
| 1539 |
+
|
| 1540 |
+
The valid distance metrics, and the function they map to, are:
|
| 1541 |
+
|
| 1542 |
+
=============== ========================================
|
| 1543 |
+
metric Function
|
| 1544 |
+
=============== ========================================
|
| 1545 |
+
'cityblock' metrics.pairwise.manhattan_distances
|
| 1546 |
+
'cosine' metrics.pairwise.cosine_distances
|
| 1547 |
+
'euclidean' metrics.pairwise.euclidean_distances
|
| 1548 |
+
'haversine' metrics.pairwise.haversine_distances
|
| 1549 |
+
'l1' metrics.pairwise.manhattan_distances
|
| 1550 |
+
'l2' metrics.pairwise.euclidean_distances
|
| 1551 |
+
'manhattan' metrics.pairwise.manhattan_distances
|
| 1552 |
+
'nan_euclidean' metrics.pairwise.nan_euclidean_distances
|
| 1553 |
+
=============== ========================================
|
| 1554 |
+
|
| 1555 |
+
Read more in the :ref:`User Guide <metrics>`.
|
| 1556 |
+
|
| 1557 |
+
Returns
|
| 1558 |
+
-------
|
| 1559 |
+
distance_metrics : dict
|
| 1560 |
+
Returns valid metrics for pairwise_distances.
|
| 1561 |
+
"""
|
| 1562 |
+
return PAIRWISE_DISTANCE_FUNCTIONS
|
| 1563 |
+
|
| 1564 |
+
|
| 1565 |
+
def _dist_wrapper(dist_func, dist_matrix, slice_, *args, **kwargs):
|
| 1566 |
+
"""Write in-place to a slice of a distance matrix."""
|
| 1567 |
+
dist_matrix[:, slice_] = dist_func(*args, **kwargs)
|
| 1568 |
+
|
| 1569 |
+
|
| 1570 |
+
def _parallel_pairwise(X, Y, func, n_jobs, **kwds):
|
| 1571 |
+
"""Break the pairwise matrix in n_jobs even slices
|
| 1572 |
+
and compute them in parallel."""
|
| 1573 |
+
|
| 1574 |
+
if Y is None:
|
| 1575 |
+
Y = X
|
| 1576 |
+
X, Y, dtype = _return_float_dtype(X, Y)
|
| 1577 |
+
|
| 1578 |
+
if effective_n_jobs(n_jobs) == 1:
|
| 1579 |
+
return func(X, Y, **kwds)
|
| 1580 |
+
|
| 1581 |
+
# enforce a threading backend to prevent data communication overhead
|
| 1582 |
+
fd = delayed(_dist_wrapper)
|
| 1583 |
+
ret = np.empty((X.shape[0], Y.shape[0]), dtype=dtype, order="F")
|
| 1584 |
+
Parallel(backend="threading", n_jobs=n_jobs)(
|
| 1585 |
+
fd(func, ret, s, X, Y[s], **kwds)
|
| 1586 |
+
for s in gen_even_slices(_num_samples(Y), effective_n_jobs(n_jobs))
|
| 1587 |
+
)
|
| 1588 |
+
|
| 1589 |
+
if (X is Y or Y is None) and func is euclidean_distances:
|
| 1590 |
+
# zeroing diagonal for euclidean norm.
|
| 1591 |
+
# TODO: do it also for other norms.
|
| 1592 |
+
np.fill_diagonal(ret, 0)
|
| 1593 |
+
|
| 1594 |
+
return ret
|
| 1595 |
+
|
| 1596 |
+
|
| 1597 |
+
def _pairwise_callable(X, Y, metric, force_all_finite=True, **kwds):
|
| 1598 |
+
"""Handle the callable case for pairwise_{distances,kernels}."""
|
| 1599 |
+
X, Y = check_pairwise_arrays(X, Y, force_all_finite=force_all_finite)
|
| 1600 |
+
|
| 1601 |
+
if X is Y:
|
| 1602 |
+
# Only calculate metric for upper triangle
|
| 1603 |
+
out = np.zeros((X.shape[0], Y.shape[0]), dtype="float")
|
| 1604 |
+
iterator = itertools.combinations(range(X.shape[0]), 2)
|
| 1605 |
+
for i, j in iterator:
|
| 1606 |
+
out[i, j] = metric(X[i], Y[j], **kwds)
|
| 1607 |
+
|
| 1608 |
+
# Make symmetric
|
| 1609 |
+
# NB: out += out.T will produce incorrect results
|
| 1610 |
+
out = out + out.T
|
| 1611 |
+
|
| 1612 |
+
# Calculate diagonal
|
| 1613 |
+
# NB: nonzero diagonals are allowed for both metrics and kernels
|
| 1614 |
+
for i in range(X.shape[0]):
|
| 1615 |
+
x = X[i]
|
| 1616 |
+
out[i, i] = metric(x, x, **kwds)
|
| 1617 |
+
|
| 1618 |
+
else:
|
| 1619 |
+
# Calculate all cells
|
| 1620 |
+
out = np.empty((X.shape[0], Y.shape[0]), dtype="float")
|
| 1621 |
+
iterator = itertools.product(range(X.shape[0]), range(Y.shape[0]))
|
| 1622 |
+
for i, j in iterator:
|
| 1623 |
+
out[i, j] = metric(X[i], Y[j], **kwds)
|
| 1624 |
+
|
| 1625 |
+
return out
|
| 1626 |
+
|
| 1627 |
+
|
| 1628 |
+
_VALID_METRICS = [
|
| 1629 |
+
"euclidean",
|
| 1630 |
+
"l2",
|
| 1631 |
+
"l1",
|
| 1632 |
+
"manhattan",
|
| 1633 |
+
"cityblock",
|
| 1634 |
+
"braycurtis",
|
| 1635 |
+
"canberra",
|
| 1636 |
+
"chebyshev",
|
| 1637 |
+
"correlation",
|
| 1638 |
+
"cosine",
|
| 1639 |
+
"dice",
|
| 1640 |
+
"hamming",
|
| 1641 |
+
"jaccard",
|
| 1642 |
+
"kulsinski",
|
| 1643 |
+
"mahalanobis",
|
| 1644 |
+
"matching",
|
| 1645 |
+
"minkowski",
|
| 1646 |
+
"rogerstanimoto",
|
| 1647 |
+
"russellrao",
|
| 1648 |
+
"seuclidean",
|
| 1649 |
+
"sokalmichener",
|
| 1650 |
+
"sokalsneath",
|
| 1651 |
+
"sqeuclidean",
|
| 1652 |
+
"yule",
|
| 1653 |
+
"wminkowski",
|
| 1654 |
+
"nan_euclidean",
|
| 1655 |
+
"haversine",
|
| 1656 |
+
]
|
| 1657 |
+
|
| 1658 |
+
_NAN_METRICS = ["nan_euclidean"]
|
| 1659 |
+
|
| 1660 |
+
|
| 1661 |
+
def _check_chunk_size(reduced, chunk_size):
|
| 1662 |
+
"""Checks chunk is a sequence of expected size or a tuple of same."""
|
| 1663 |
+
if reduced is None:
|
| 1664 |
+
return
|
| 1665 |
+
is_tuple = isinstance(reduced, tuple)
|
| 1666 |
+
if not is_tuple:
|
| 1667 |
+
reduced = (reduced,)
|
| 1668 |
+
if any(isinstance(r, tuple) or not hasattr(r, "__iter__") for r in reduced):
|
| 1669 |
+
raise TypeError(
|
| 1670 |
+
"reduce_func returned %r. Expected sequence(s) of length %d."
|
| 1671 |
+
% (reduced if is_tuple else reduced[0], chunk_size)
|
| 1672 |
+
)
|
| 1673 |
+
if any(_num_samples(r) != chunk_size for r in reduced):
|
| 1674 |
+
actual_size = tuple(_num_samples(r) for r in reduced)
|
| 1675 |
+
raise ValueError(
|
| 1676 |
+
"reduce_func returned object of length %s. "
|
| 1677 |
+
"Expected same length as input: %d."
|
| 1678 |
+
% (actual_size if is_tuple else actual_size[0], chunk_size)
|
| 1679 |
+
)
|
| 1680 |
+
|
| 1681 |
+
|
| 1682 |
+
def _precompute_metric_params(X, Y, metric=None, **kwds):
|
| 1683 |
+
"""Precompute data-derived metric parameters if not provided."""
|
| 1684 |
+
if metric == "seuclidean" and "V" not in kwds:
|
| 1685 |
+
# There is a bug in scipy < 1.5 that will cause a crash if
|
| 1686 |
+
# X.dtype != np.double (float64). See PR #15730
|
| 1687 |
+
dtype = np.float64 if sp_version < parse_version("1.5") else None
|
| 1688 |
+
if X is Y:
|
| 1689 |
+
V = np.var(X, axis=0, ddof=1, dtype=dtype)
|
| 1690 |
+
else:
|
| 1691 |
+
raise ValueError(
|
| 1692 |
+
"The 'V' parameter is required for the seuclidean metric "
|
| 1693 |
+
"when Y is passed."
|
| 1694 |
+
)
|
| 1695 |
+
return {"V": V}
|
| 1696 |
+
if metric == "mahalanobis" and "VI" not in kwds:
|
| 1697 |
+
if X is Y:
|
| 1698 |
+
VI = np.linalg.inv(np.cov(X.T)).T
|
| 1699 |
+
else:
|
| 1700 |
+
raise ValueError(
|
| 1701 |
+
"The 'VI' parameter is required for the mahalanobis metric "
|
| 1702 |
+
"when Y is passed."
|
| 1703 |
+
)
|
| 1704 |
+
return {"VI": VI}
|
| 1705 |
+
return {}
|
| 1706 |
+
|
| 1707 |
+
|
| 1708 |
+
def pairwise_distances_chunked(
|
| 1709 |
+
X,
|
| 1710 |
+
Y=None,
|
| 1711 |
+
*,
|
| 1712 |
+
reduce_func=None,
|
| 1713 |
+
metric="euclidean",
|
| 1714 |
+
n_jobs=None,
|
| 1715 |
+
working_memory=None,
|
| 1716 |
+
**kwds,
|
| 1717 |
+
):
|
| 1718 |
+
"""Generate a distance matrix chunk by chunk with optional reduction.
|
| 1719 |
+
|
| 1720 |
+
In cases where not all of a pairwise distance matrix needs to be
|
| 1721 |
+
stored at once, this is used to calculate pairwise distances in
|
| 1722 |
+
``working_memory``-sized chunks. If ``reduce_func`` is given, it is
|
| 1723 |
+
run on each chunk and its return values are concatenated into lists,
|
| 1724 |
+
arrays or sparse matrices.
|
| 1725 |
+
|
| 1726 |
+
Parameters
|
| 1727 |
+
----------
|
| 1728 |
+
X : ndarray of shape (n_samples_X, n_samples_X) or \
|
| 1729 |
+
(n_samples_X, n_features)
|
| 1730 |
+
Array of pairwise distances between samples, or a feature array.
|
| 1731 |
+
The shape the array should be (n_samples_X, n_samples_X) if
|
| 1732 |
+
metric='precomputed' and (n_samples_X, n_features) otherwise.
|
| 1733 |
+
|
| 1734 |
+
Y : ndarray of shape (n_samples_Y, n_features), default=None
|
| 1735 |
+
An optional second feature array. Only allowed if
|
| 1736 |
+
metric != "precomputed".
|
| 1737 |
+
|
| 1738 |
+
reduce_func : callable, default=None
|
| 1739 |
+
The function which is applied on each chunk of the distance matrix,
|
| 1740 |
+
reducing it to needed values. ``reduce_func(D_chunk, start)``
|
| 1741 |
+
is called repeatedly, where ``D_chunk`` is a contiguous vertical
|
| 1742 |
+
slice of the pairwise distance matrix, starting at row ``start``.
|
| 1743 |
+
It should return one of: None; an array, a list, or a sparse matrix
|
| 1744 |
+
of length ``D_chunk.shape[0]``; or a tuple of such objects.
|
| 1745 |
+
Returning None is useful for in-place operations, rather than
|
| 1746 |
+
reductions.
|
| 1747 |
+
|
| 1748 |
+
If None, pairwise_distances_chunked returns a generator of vertical
|
| 1749 |
+
chunks of the distance matrix.
|
| 1750 |
+
|
| 1751 |
+
metric : str or callable, default='euclidean'
|
| 1752 |
+
The metric to use when calculating distance between instances in a
|
| 1753 |
+
feature array. If metric is a string, it must be one of the options
|
| 1754 |
+
allowed by scipy.spatial.distance.pdist for its metric parameter,
|
| 1755 |
+
or a metric listed in pairwise.PAIRWISE_DISTANCE_FUNCTIONS.
|
| 1756 |
+
If metric is "precomputed", X is assumed to be a distance matrix.
|
| 1757 |
+
Alternatively, if metric is a callable function, it is called on
|
| 1758 |
+
each pair of instances (rows) and the resulting value recorded.
|
| 1759 |
+
The callable should take two arrays from X as input and return a
|
| 1760 |
+
value indicating the distance between them.
|
| 1761 |
+
|
| 1762 |
+
n_jobs : int, default=None
|
| 1763 |
+
The number of jobs to use for the computation. This works by
|
| 1764 |
+
breaking down the pairwise matrix into n_jobs even slices and
|
| 1765 |
+
computing them in parallel.
|
| 1766 |
+
|
| 1767 |
+
``None`` means 1 unless in a :obj:`joblib.parallel_backend` context.
|
| 1768 |
+
``-1`` means using all processors. See :term:`Glossary <n_jobs>`
|
| 1769 |
+
for more details.
|
| 1770 |
+
|
| 1771 |
+
working_memory : int, default=None
|
| 1772 |
+
The sought maximum memory for temporary distance matrix chunks.
|
| 1773 |
+
When None (default), the value of
|
| 1774 |
+
``sklearn.get_config()['working_memory']`` is used.
|
| 1775 |
+
|
| 1776 |
+
**kwds : optional keyword parameters
|
| 1777 |
+
Any further parameters are passed directly to the distance function.
|
| 1778 |
+
If using a scipy.spatial.distance metric, the parameters are still
|
| 1779 |
+
metric dependent. See the scipy docs for usage examples.
|
| 1780 |
+
|
| 1781 |
+
Yields
|
| 1782 |
+
------
|
| 1783 |
+
D_chunk : {ndarray, sparse matrix}
|
| 1784 |
+
A contiguous slice of distance matrix, optionally processed by
|
| 1785 |
+
``reduce_func``.
|
| 1786 |
+
|
| 1787 |
+
Examples
|
| 1788 |
+
--------
|
| 1789 |
+
Without reduce_func:
|
| 1790 |
+
|
| 1791 |
+
>>> import numpy as np
|
| 1792 |
+
>>> from sklearn.metrics import pairwise_distances_chunked
|
| 1793 |
+
>>> X = np.random.RandomState(0).rand(5, 3)
|
| 1794 |
+
>>> D_chunk = next(pairwise_distances_chunked(X))
|
| 1795 |
+
>>> D_chunk
|
| 1796 |
+
array([[0. ..., 0.29..., 0.41..., 0.19..., 0.57...],
|
| 1797 |
+
[0.29..., 0. ..., 0.57..., 0.41..., 0.76...],
|
| 1798 |
+
[0.41..., 0.57..., 0. ..., 0.44..., 0.90...],
|
| 1799 |
+
[0.19..., 0.41..., 0.44..., 0. ..., 0.51...],
|
| 1800 |
+
[0.57..., 0.76..., 0.90..., 0.51..., 0. ...]])
|
| 1801 |
+
|
| 1802 |
+
Retrieve all neighbors and average distance within radius r:
|
| 1803 |
+
|
| 1804 |
+
>>> r = .2
|
| 1805 |
+
>>> def reduce_func(D_chunk, start):
|
| 1806 |
+
... neigh = [np.flatnonzero(d < r) for d in D_chunk]
|
| 1807 |
+
... avg_dist = (D_chunk * (D_chunk < r)).mean(axis=1)
|
| 1808 |
+
... return neigh, avg_dist
|
| 1809 |
+
>>> gen = pairwise_distances_chunked(X, reduce_func=reduce_func)
|
| 1810 |
+
>>> neigh, avg_dist = next(gen)
|
| 1811 |
+
>>> neigh
|
| 1812 |
+
[array([0, 3]), array([1]), array([2]), array([0, 3]), array([4])]
|
| 1813 |
+
>>> avg_dist
|
| 1814 |
+
array([0.039..., 0. , 0. , 0.039..., 0. ])
|
| 1815 |
+
|
| 1816 |
+
Where r is defined per sample, we need to make use of ``start``:
|
| 1817 |
+
|
| 1818 |
+
>>> r = [.2, .4, .4, .3, .1]
|
| 1819 |
+
>>> def reduce_func(D_chunk, start):
|
| 1820 |
+
... neigh = [np.flatnonzero(d < r[i])
|
| 1821 |
+
... for i, d in enumerate(D_chunk, start)]
|
| 1822 |
+
... return neigh
|
| 1823 |
+
>>> neigh = next(pairwise_distances_chunked(X, reduce_func=reduce_func))
|
| 1824 |
+
>>> neigh
|
| 1825 |
+
[array([0, 3]), array([0, 1]), array([2]), array([0, 3]), array([4])]
|
| 1826 |
+
|
| 1827 |
+
Force row-by-row generation by reducing ``working_memory``:
|
| 1828 |
+
|
| 1829 |
+
>>> gen = pairwise_distances_chunked(X, reduce_func=reduce_func,
|
| 1830 |
+
... working_memory=0)
|
| 1831 |
+
>>> next(gen)
|
| 1832 |
+
[array([0, 3])]
|
| 1833 |
+
>>> next(gen)
|
| 1834 |
+
[array([0, 1])]
|
| 1835 |
+
"""
|
| 1836 |
+
n_samples_X = _num_samples(X)
|
| 1837 |
+
if metric == "precomputed":
|
| 1838 |
+
slices = (slice(0, n_samples_X),)
|
| 1839 |
+
else:
|
| 1840 |
+
if Y is None:
|
| 1841 |
+
Y = X
|
| 1842 |
+
# We get as many rows as possible within our working_memory budget to
|
| 1843 |
+
# store len(Y) distances in each row of output.
|
| 1844 |
+
#
|
| 1845 |
+
# Note:
|
| 1846 |
+
# - this will get at least 1 row, even if 1 row of distances will
|
| 1847 |
+
# exceed working_memory.
|
| 1848 |
+
# - this does not account for any temporary memory usage while
|
| 1849 |
+
# calculating distances (e.g. difference of vectors in manhattan
|
| 1850 |
+
# distance.
|
| 1851 |
+
chunk_n_rows = get_chunk_n_rows(
|
| 1852 |
+
row_bytes=8 * _num_samples(Y),
|
| 1853 |
+
max_n_rows=n_samples_X,
|
| 1854 |
+
working_memory=working_memory,
|
| 1855 |
+
)
|
| 1856 |
+
slices = gen_batches(n_samples_X, chunk_n_rows)
|
| 1857 |
+
|
| 1858 |
+
# precompute data-derived metric params
|
| 1859 |
+
params = _precompute_metric_params(X, Y, metric=metric, **kwds)
|
| 1860 |
+
kwds.update(**params)
|
| 1861 |
+
|
| 1862 |
+
for sl in slices:
|
| 1863 |
+
if sl.start == 0 and sl.stop == n_samples_X:
|
| 1864 |
+
X_chunk = X # enable optimised paths for X is Y
|
| 1865 |
+
else:
|
| 1866 |
+
X_chunk = X[sl]
|
| 1867 |
+
D_chunk = pairwise_distances(X_chunk, Y, metric=metric, n_jobs=n_jobs, **kwds)
|
| 1868 |
+
if (X is Y or Y is None) and PAIRWISE_DISTANCE_FUNCTIONS.get(
|
| 1869 |
+
metric, None
|
| 1870 |
+
) is euclidean_distances:
|
| 1871 |
+
# zeroing diagonal, taking care of aliases of "euclidean",
|
| 1872 |
+
# i.e. "l2"
|
| 1873 |
+
D_chunk.flat[sl.start :: _num_samples(X) + 1] = 0
|
| 1874 |
+
if reduce_func is not None:
|
| 1875 |
+
chunk_size = D_chunk.shape[0]
|
| 1876 |
+
D_chunk = reduce_func(D_chunk, sl.start)
|
| 1877 |
+
_check_chunk_size(D_chunk, chunk_size)
|
| 1878 |
+
yield D_chunk
|
| 1879 |
+
|
| 1880 |
+
|
| 1881 |
+
def pairwise_distances(
|
| 1882 |
+
X, Y=None, metric="euclidean", *, n_jobs=None, force_all_finite=True, **kwds
|
| 1883 |
+
):
|
| 1884 |
+
"""Compute the distance matrix from a vector array X and optional Y.
|
| 1885 |
+
|
| 1886 |
+
This method takes either a vector array or a distance matrix, and returns
|
| 1887 |
+
a distance matrix. If the input is a vector array, the distances are
|
| 1888 |
+
computed. If the input is a distances matrix, it is returned instead.
|
| 1889 |
+
|
| 1890 |
+
This method provides a safe way to take a distance matrix as input, while
|
| 1891 |
+
preserving compatibility with many other algorithms that take a vector
|
| 1892 |
+
array.
|
| 1893 |
+
|
| 1894 |
+
If Y is given (default is None), then the returned matrix is the pairwise
|
| 1895 |
+
distance between the arrays from both X and Y.
|
| 1896 |
+
|
| 1897 |
+
Valid values for metric are:
|
| 1898 |
+
|
| 1899 |
+
- From scikit-learn: ['cityblock', 'cosine', 'euclidean', 'l1', 'l2',
|
| 1900 |
+
'manhattan']. These metrics support sparse matrix
|
| 1901 |
+
inputs.
|
| 1902 |
+
['nan_euclidean'] but it does not yet support sparse matrices.
|
| 1903 |
+
|
| 1904 |
+
- From scipy.spatial.distance: ['braycurtis', 'canberra', 'chebyshev',
|
| 1905 |
+
'correlation', 'dice', 'hamming', 'jaccard', 'kulsinski', 'mahalanobis',
|
| 1906 |
+
'minkowski', 'rogerstanimoto', 'russellrao', 'seuclidean',
|
| 1907 |
+
'sokalmichener', 'sokalsneath', 'sqeuclidean', 'yule']
|
| 1908 |
+
See the documentation for scipy.spatial.distance for details on these
|
| 1909 |
+
metrics. These metrics do not support sparse matrix inputs.
|
| 1910 |
+
|
| 1911 |
+
Note that in the case of 'cityblock', 'cosine' and 'euclidean' (which are
|
| 1912 |
+
valid scipy.spatial.distance metrics), the scikit-learn implementation
|
| 1913 |
+
will be used, which is faster and has support for sparse matrices (except
|
| 1914 |
+
for 'cityblock'). For a verbose description of the metrics from
|
| 1915 |
+
scikit-learn, see :func:`sklearn.metrics.pairwise.distance_metrics`
|
| 1916 |
+
function.
|
| 1917 |
+
|
| 1918 |
+
Read more in the :ref:`User Guide <metrics>`.
|
| 1919 |
+
|
| 1920 |
+
Parameters
|
| 1921 |
+
----------
|
| 1922 |
+
X : ndarray of shape (n_samples_X, n_samples_X) or \
|
| 1923 |
+
(n_samples_X, n_features)
|
| 1924 |
+
Array of pairwise distances between samples, or a feature array.
|
| 1925 |
+
The shape of the array should be (n_samples_X, n_samples_X) if
|
| 1926 |
+
metric == "precomputed" and (n_samples_X, n_features) otherwise.
|
| 1927 |
+
|
| 1928 |
+
Y : ndarray of shape (n_samples_Y, n_features), default=None
|
| 1929 |
+
An optional second feature array. Only allowed if
|
| 1930 |
+
metric != "precomputed".
|
| 1931 |
+
|
| 1932 |
+
metric : str or callable, default='euclidean'
|
| 1933 |
+
The metric to use when calculating distance between instances in a
|
| 1934 |
+
feature array. If metric is a string, it must be one of the options
|
| 1935 |
+
allowed by scipy.spatial.distance.pdist for its metric parameter, or
|
| 1936 |
+
a metric listed in ``pairwise.PAIRWISE_DISTANCE_FUNCTIONS``.
|
| 1937 |
+
If metric is "precomputed", X is assumed to be a distance matrix.
|
| 1938 |
+
Alternatively, if metric is a callable function, it is called on each
|
| 1939 |
+
pair of instances (rows) and the resulting value recorded. The callable
|
| 1940 |
+
should take two arrays from X as input and return a value indicating
|
| 1941 |
+
the distance between them.
|
| 1942 |
+
|
| 1943 |
+
n_jobs : int, default=None
|
| 1944 |
+
The number of jobs to use for the computation. This works by breaking
|
| 1945 |
+
down the pairwise matrix into n_jobs even slices and computing them in
|
| 1946 |
+
parallel.
|
| 1947 |
+
|
| 1948 |
+
``None`` means 1 unless in a :obj:`joblib.parallel_backend` context.
|
| 1949 |
+
``-1`` means using all processors. See :term:`Glossary <n_jobs>`
|
| 1950 |
+
for more details.
|
| 1951 |
+
|
| 1952 |
+
force_all_finite : bool or 'allow-nan', default=True
|
| 1953 |
+
Whether to raise an error on np.inf, np.nan, pd.NA in array. Ignored
|
| 1954 |
+
for a metric listed in ``pairwise.PAIRWISE_DISTANCE_FUNCTIONS``. The
|
| 1955 |
+
possibilities are:
|
| 1956 |
+
|
| 1957 |
+
- True: Force all values of array to be finite.
|
| 1958 |
+
- False: accepts np.inf, np.nan, pd.NA in array.
|
| 1959 |
+
- 'allow-nan': accepts only np.nan and pd.NA values in array. Values
|
| 1960 |
+
cannot be infinite.
|
| 1961 |
+
|
| 1962 |
+
.. versionadded:: 0.22
|
| 1963 |
+
``force_all_finite`` accepts the string ``'allow-nan'``.
|
| 1964 |
+
|
| 1965 |
+
.. versionchanged:: 0.23
|
| 1966 |
+
Accepts `pd.NA` and converts it into `np.nan`.
|
| 1967 |
+
|
| 1968 |
+
**kwds : optional keyword parameters
|
| 1969 |
+
Any further parameters are passed directly to the distance function.
|
| 1970 |
+
If using a scipy.spatial.distance metric, the parameters are still
|
| 1971 |
+
metric dependent. See the scipy docs for usage examples.
|
| 1972 |
+
|
| 1973 |
+
Returns
|
| 1974 |
+
-------
|
| 1975 |
+
D : ndarray of shape (n_samples_X, n_samples_X) or \
|
| 1976 |
+
(n_samples_X, n_samples_Y)
|
| 1977 |
+
A distance matrix D such that D_{i, j} is the distance between the
|
| 1978 |
+
ith and jth vectors of the given matrix X, if Y is None.
|
| 1979 |
+
If Y is not None, then D_{i, j} is the distance between the ith array
|
| 1980 |
+
from X and the jth array from Y.
|
| 1981 |
+
|
| 1982 |
+
See Also
|
| 1983 |
+
--------
|
| 1984 |
+
pairwise_distances_chunked : Performs the same calculation as this
|
| 1985 |
+
function, but returns a generator of chunks of the distance matrix, in
|
| 1986 |
+
order to limit memory usage.
|
| 1987 |
+
paired_distances : Computes the distances between corresponding elements
|
| 1988 |
+
of two arrays.
|
| 1989 |
+
"""
|
| 1990 |
+
if (
|
| 1991 |
+
metric not in _VALID_METRICS
|
| 1992 |
+
and not callable(metric)
|
| 1993 |
+
and metric != "precomputed"
|
| 1994 |
+
):
|
| 1995 |
+
raise ValueError(
|
| 1996 |
+
"Unknown metric %s. Valid metrics are %s, or 'precomputed', or a callable"
|
| 1997 |
+
% (metric, _VALID_METRICS)
|
| 1998 |
+
)
|
| 1999 |
+
|
| 2000 |
+
if metric == "precomputed":
|
| 2001 |
+
X, _ = check_pairwise_arrays(
|
| 2002 |
+
X, Y, precomputed=True, force_all_finite=force_all_finite
|
| 2003 |
+
)
|
| 2004 |
+
|
| 2005 |
+
whom = (
|
| 2006 |
+
"`pairwise_distances`. Precomputed distance "
|
| 2007 |
+
" need to have non-negative values."
|
| 2008 |
+
)
|
| 2009 |
+
check_non_negative(X, whom=whom)
|
| 2010 |
+
return X
|
| 2011 |
+
elif metric in PAIRWISE_DISTANCE_FUNCTIONS:
|
| 2012 |
+
func = PAIRWISE_DISTANCE_FUNCTIONS[metric]
|
| 2013 |
+
elif callable(metric):
|
| 2014 |
+
func = partial(
|
| 2015 |
+
_pairwise_callable, metric=metric, force_all_finite=force_all_finite, **kwds
|
| 2016 |
+
)
|
| 2017 |
+
else:
|
| 2018 |
+
if issparse(X) or issparse(Y):
|
| 2019 |
+
raise TypeError("scipy distance metrics do not support sparse matrices.")
|
| 2020 |
+
|
| 2021 |
+
dtype = bool if metric in PAIRWISE_BOOLEAN_FUNCTIONS else None
|
| 2022 |
+
|
| 2023 |
+
if dtype == bool and (X.dtype != bool or (Y is not None and Y.dtype != bool)):
|
| 2024 |
+
msg = "Data was converted to boolean for metric %s" % metric
|
| 2025 |
+
warnings.warn(msg, DataConversionWarning)
|
| 2026 |
+
|
| 2027 |
+
X, Y = check_pairwise_arrays(
|
| 2028 |
+
X, Y, dtype=dtype, force_all_finite=force_all_finite
|
| 2029 |
+
)
|
| 2030 |
+
|
| 2031 |
+
# precompute data-derived metric params
|
| 2032 |
+
params = _precompute_metric_params(X, Y, metric=metric, **kwds)
|
| 2033 |
+
kwds.update(**params)
|
| 2034 |
+
|
| 2035 |
+
if effective_n_jobs(n_jobs) == 1 and X is Y:
|
| 2036 |
+
return distance.squareform(distance.pdist(X, metric=metric, **kwds))
|
| 2037 |
+
func = partial(distance.cdist, metric=metric, **kwds)
|
| 2038 |
+
|
| 2039 |
+
return _parallel_pairwise(X, Y, func, n_jobs, **kwds)
|
| 2040 |
+
|
| 2041 |
+
|
| 2042 |
+
# These distances require boolean arrays, when using scipy.spatial.distance
|
| 2043 |
+
PAIRWISE_BOOLEAN_FUNCTIONS = [
|
| 2044 |
+
"dice",
|
| 2045 |
+
"jaccard",
|
| 2046 |
+
"kulsinski",
|
| 2047 |
+
"matching",
|
| 2048 |
+
"rogerstanimoto",
|
| 2049 |
+
"russellrao",
|
| 2050 |
+
"sokalmichener",
|
| 2051 |
+
"sokalsneath",
|
| 2052 |
+
"yule",
|
| 2053 |
+
]
|
| 2054 |
+
|
| 2055 |
+
# Helper functions - distance
|
| 2056 |
+
PAIRWISE_KERNEL_FUNCTIONS = {
|
| 2057 |
+
# If updating this dictionary, update the doc in both distance_metrics()
|
| 2058 |
+
# and also in pairwise_distances()!
|
| 2059 |
+
"additive_chi2": additive_chi2_kernel,
|
| 2060 |
+
"chi2": chi2_kernel,
|
| 2061 |
+
"linear": linear_kernel,
|
| 2062 |
+
"polynomial": polynomial_kernel,
|
| 2063 |
+
"poly": polynomial_kernel,
|
| 2064 |
+
"rbf": rbf_kernel,
|
| 2065 |
+
"laplacian": laplacian_kernel,
|
| 2066 |
+
"sigmoid": sigmoid_kernel,
|
| 2067 |
+
"cosine": cosine_similarity,
|
| 2068 |
+
}
|
| 2069 |
+
|
| 2070 |
+
|
| 2071 |
+
def kernel_metrics():
|
| 2072 |
+
"""Valid metrics for pairwise_kernels.
|
| 2073 |
+
|
| 2074 |
+
This function simply returns the valid pairwise distance metrics.
|
| 2075 |
+
It exists, however, to allow for a verbose description of the mapping for
|
| 2076 |
+
each of the valid strings.
|
| 2077 |
+
|
| 2078 |
+
The valid distance metrics, and the function they map to, are:
|
| 2079 |
+
=============== ========================================
|
| 2080 |
+
metric Function
|
| 2081 |
+
=============== ========================================
|
| 2082 |
+
'additive_chi2' sklearn.pairwise.additive_chi2_kernel
|
| 2083 |
+
'chi2' sklearn.pairwise.chi2_kernel
|
| 2084 |
+
'linear' sklearn.pairwise.linear_kernel
|
| 2085 |
+
'poly' sklearn.pairwise.polynomial_kernel
|
| 2086 |
+
'polynomial' sklearn.pairwise.polynomial_kernel
|
| 2087 |
+
'rbf' sklearn.pairwise.rbf_kernel
|
| 2088 |
+
'laplacian' sklearn.pairwise.laplacian_kernel
|
| 2089 |
+
'sigmoid' sklearn.pairwise.sigmoid_kernel
|
| 2090 |
+
'cosine' sklearn.pairwise.cosine_similarity
|
| 2091 |
+
=============== ========================================
|
| 2092 |
+
|
| 2093 |
+
Read more in the :ref:`User Guide <metrics>`.
|
| 2094 |
+
|
| 2095 |
+
Returns
|
| 2096 |
+
-------
|
| 2097 |
+
kernal_metrics : dict
|
| 2098 |
+
Returns valid metrics for pairwise_kernels.
|
| 2099 |
+
"""
|
| 2100 |
+
return PAIRWISE_KERNEL_FUNCTIONS
|
| 2101 |
+
|
| 2102 |
+
|
| 2103 |
+
KERNEL_PARAMS = {
|
| 2104 |
+
"additive_chi2": (),
|
| 2105 |
+
"chi2": frozenset(["gamma"]),
|
| 2106 |
+
"cosine": (),
|
| 2107 |
+
"linear": (),
|
| 2108 |
+
"poly": frozenset(["gamma", "degree", "coef0"]),
|
| 2109 |
+
"polynomial": frozenset(["gamma", "degree", "coef0"]),
|
| 2110 |
+
"rbf": frozenset(["gamma"]),
|
| 2111 |
+
"laplacian": frozenset(["gamma"]),
|
| 2112 |
+
"sigmoid": frozenset(["gamma", "coef0"]),
|
| 2113 |
+
}
|
| 2114 |
+
|
| 2115 |
+
|
| 2116 |
+
def pairwise_kernels(
|
| 2117 |
+
X, Y=None, metric="linear", *, filter_params=False, n_jobs=None, **kwds
|
| 2118 |
+
):
|
| 2119 |
+
"""Compute the kernel between arrays X and optional array Y.
|
| 2120 |
+
|
| 2121 |
+
This method takes either a vector array or a kernel matrix, and returns
|
| 2122 |
+
a kernel matrix. If the input is a vector array, the kernels are
|
| 2123 |
+
computed. If the input is a kernel matrix, it is returned instead.
|
| 2124 |
+
|
| 2125 |
+
This method provides a safe way to take a kernel matrix as input, while
|
| 2126 |
+
preserving compatibility with many other algorithms that take a vector
|
| 2127 |
+
array.
|
| 2128 |
+
|
| 2129 |
+
If Y is given (default is None), then the returned matrix is the pairwise
|
| 2130 |
+
kernel between the arrays from both X and Y.
|
| 2131 |
+
|
| 2132 |
+
Valid values for metric are:
|
| 2133 |
+
['additive_chi2', 'chi2', 'linear', 'poly', 'polynomial', 'rbf',
|
| 2134 |
+
'laplacian', 'sigmoid', 'cosine']
|
| 2135 |
+
|
| 2136 |
+
Read more in the :ref:`User Guide <metrics>`.
|
| 2137 |
+
|
| 2138 |
+
Parameters
|
| 2139 |
+
----------
|
| 2140 |
+
X : ndarray of shape (n_samples_X, n_samples_X) or (n_samples_X, n_features)
|
| 2141 |
+
Array of pairwise kernels between samples, or a feature array.
|
| 2142 |
+
The shape of the array should be (n_samples_X, n_samples_X) if
|
| 2143 |
+
metric == "precomputed" and (n_samples_X, n_features) otherwise.
|
| 2144 |
+
|
| 2145 |
+
Y : ndarray of shape (n_samples_Y, n_features), default=None
|
| 2146 |
+
A second feature array only if X has shape (n_samples_X, n_features).
|
| 2147 |
+
|
| 2148 |
+
metric : str or callable, default="linear"
|
| 2149 |
+
The metric to use when calculating kernel between instances in a
|
| 2150 |
+
feature array. If metric is a string, it must be one of the metrics
|
| 2151 |
+
in pairwise.PAIRWISE_KERNEL_FUNCTIONS.
|
| 2152 |
+
If metric is "precomputed", X is assumed to be a kernel matrix.
|
| 2153 |
+
Alternatively, if metric is a callable function, it is called on each
|
| 2154 |
+
pair of instances (rows) and the resulting value recorded. The callable
|
| 2155 |
+
should take two rows from X as input and return the corresponding
|
| 2156 |
+
kernel value as a single number. This means that callables from
|
| 2157 |
+
:mod:`sklearn.metrics.pairwise` are not allowed, as they operate on
|
| 2158 |
+
matrices, not single samples. Use the string identifying the kernel
|
| 2159 |
+
instead.
|
| 2160 |
+
|
| 2161 |
+
filter_params : bool, default=False
|
| 2162 |
+
Whether to filter invalid parameters or not.
|
| 2163 |
+
|
| 2164 |
+
n_jobs : int, default=None
|
| 2165 |
+
The number of jobs to use for the computation. This works by breaking
|
| 2166 |
+
down the pairwise matrix into n_jobs even slices and computing them in
|
| 2167 |
+
parallel.
|
| 2168 |
+
|
| 2169 |
+
``None`` means 1 unless in a :obj:`joblib.parallel_backend` context.
|
| 2170 |
+
``-1`` means using all processors. See :term:`Glossary <n_jobs>`
|
| 2171 |
+
for more details.
|
| 2172 |
+
|
| 2173 |
+
**kwds : optional keyword parameters
|
| 2174 |
+
Any further parameters are passed directly to the kernel function.
|
| 2175 |
+
|
| 2176 |
+
Returns
|
| 2177 |
+
-------
|
| 2178 |
+
K : ndarray of shape (n_samples_X, n_samples_X) or (n_samples_X, n_samples_Y)
|
| 2179 |
+
A kernel matrix K such that K_{i, j} is the kernel between the
|
| 2180 |
+
ith and jth vectors of the given matrix X, if Y is None.
|
| 2181 |
+
If Y is not None, then K_{i, j} is the kernel between the ith array
|
| 2182 |
+
from X and the jth array from Y.
|
| 2183 |
+
|
| 2184 |
+
Notes
|
| 2185 |
+
-----
|
| 2186 |
+
If metric is 'precomputed', Y is ignored and X is returned.
|
| 2187 |
+
"""
|
| 2188 |
+
# import GPKernel locally to prevent circular imports
|
| 2189 |
+
from ..gaussian_process.kernels import Kernel as GPKernel
|
| 2190 |
+
|
| 2191 |
+
if metric == "precomputed":
|
| 2192 |
+
X, _ = check_pairwise_arrays(X, Y, precomputed=True)
|
| 2193 |
+
return X
|
| 2194 |
+
elif isinstance(metric, GPKernel):
|
| 2195 |
+
func = metric.__call__
|
| 2196 |
+
elif metric in PAIRWISE_KERNEL_FUNCTIONS:
|
| 2197 |
+
if filter_params:
|
| 2198 |
+
kwds = {k: kwds[k] for k in kwds if k in KERNEL_PARAMS[metric]}
|
| 2199 |
+
func = PAIRWISE_KERNEL_FUNCTIONS[metric]
|
| 2200 |
+
elif callable(metric):
|
| 2201 |
+
func = partial(_pairwise_callable, metric=metric, **kwds)
|
| 2202 |
+
else:
|
| 2203 |
+
raise ValueError("Unknown kernel %r" % metric)
|
| 2204 |
+
|
| 2205 |
+
return _parallel_pairwise(X, Y, func, n_jobs, **kwds)
|
videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/__init__.cpython-310.pyc
ADDED
|
Binary file (31.6 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/_arpack.cpython-310.pyc
ADDED
|
Binary file (1.33 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/_array_api.cpython-310.pyc
ADDED
|
Binary file (8.09 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/_bunch.cpython-310.pyc
ADDED
|
Binary file (1.58 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/_estimator_html_repr.cpython-310.pyc
ADDED
|
Binary file (10.9 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/_joblib.cpython-310.pyc
ADDED
|
Binary file (730 Bytes). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/_param_validation.cpython-310.pyc
ADDED
|
Binary file (28.8 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/_pprint.cpython-310.pyc
ADDED
|
Binary file (11.5 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/class_weight.cpython-310.pyc
ADDED
|
Binary file (5.35 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/discovery.cpython-310.pyc
ADDED
|
Binary file (6.19 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/metaestimators.cpython-310.pyc
ADDED
|
Binary file (7.76 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/parallel.cpython-310.pyc
ADDED
|
Binary file (4.44 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/random.cpython-310.pyc
ADDED
|
Binary file (2.53 kB). View file
|
|
|
videollama2/lib/python3.10/site-packages/sklearn/utils/__pycache__/sparsefuncs.cpython-310.pyc
ADDED
|
Binary file (16.9 kB). View file
|
|
|
vllm/lib/python3.10/site-packages/torch/include/ATen/ops/_assert_async_cpu_dispatch.h
ADDED
|
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#pragma once
|
| 2 |
+
// @generated by torchgen/gen.py from DispatchKeyFunction.h
|
| 3 |
+
|
| 4 |
+
// NB: The implementing C++ file is RegisterDispatchKey.cpp
|
| 5 |
+
|
| 6 |
+
// The only #includes we need are for custom classes that have defaults in the C++ API
|
| 7 |
+
#include <c10/core/MemoryFormat.h>
|
| 8 |
+
#include <c10/core/Scalar.h>
|
| 9 |
+
#include <ATen/core/Reduction.h>
|
| 10 |
+
|
| 11 |
+
// Forward declarations of any types needed in the operator signatures.
|
| 12 |
+
// We can't directly include these classes because it will cause circular include dependencies.
|
| 13 |
+
// This file is included by TensorBody.h, which defines the Tensor class.
|
| 14 |
+
#include <ATen/core/ATen_fwd.h>
|
| 15 |
+
|
| 16 |
+
namespace at {
|
| 17 |
+
|
| 18 |
+
namespace cpu {
|
| 19 |
+
|
| 20 |
+
TORCH_API void _assert_async(const at::Tensor & self);
|
| 21 |
+
TORCH_API void _assert_async(const at::Tensor & self, c10::string_view assert_msg);
|
| 22 |
+
|
| 23 |
+
} // namespace cpu
|
| 24 |
+
} // namespace at
|
vllm/lib/python3.10/site-packages/torch/include/ATen/ops/_cast_Char_compositeimplicitautograd_dispatch.h
ADDED
|
@@ -0,0 +1,23 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#pragma once
|
| 2 |
+
// @generated by torchgen/gen.py from DispatchKeyFunction.h
|
| 3 |
+
|
| 4 |
+
// NB: The implementing C++ file is RegisterDispatchKey.cpp
|
| 5 |
+
|
| 6 |
+
// The only #includes we need are for custom classes that have defaults in the C++ API
|
| 7 |
+
#include <c10/core/MemoryFormat.h>
|
| 8 |
+
#include <c10/core/Scalar.h>
|
| 9 |
+
#include <ATen/core/Reduction.h>
|
| 10 |
+
|
| 11 |
+
// Forward declarations of any types needed in the operator signatures.
|
| 12 |
+
// We can't directly include these classes because it will cause circular include dependencies.
|
| 13 |
+
// This file is included by TensorBody.h, which defines the Tensor class.
|
| 14 |
+
#include <ATen/core/ATen_fwd.h>
|
| 15 |
+
|
| 16 |
+
namespace at {
|
| 17 |
+
|
| 18 |
+
namespace compositeimplicitautograd {
|
| 19 |
+
|
| 20 |
+
TORCH_API at::Tensor _cast_Char(const at::Tensor & self, bool non_blocking=false);
|
| 21 |
+
|
| 22 |
+
} // namespace compositeimplicitautograd
|
| 23 |
+
} // namespace at
|
vllm/lib/python3.10/site-packages/torch/include/ATen/ops/_copy_from_native.h
ADDED
|
@@ -0,0 +1,21 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#pragma once
|
| 2 |
+
|
| 3 |
+
// @generated by torchgen/gen.py from NativeFunction.h
|
| 4 |
+
|
| 5 |
+
#include <c10/core/Scalar.h>
|
| 6 |
+
#include <c10/core/Storage.h>
|
| 7 |
+
#include <c10/core/TensorOptions.h>
|
| 8 |
+
#include <c10/util/Deprecated.h>
|
| 9 |
+
#include <optional>
|
| 10 |
+
#include <c10/core/QScheme.h>
|
| 11 |
+
#include <ATen/core/Reduction.h>
|
| 12 |
+
#include <ATen/core/Tensor.h>
|
| 13 |
+
#include <tuple>
|
| 14 |
+
#include <vector>
|
| 15 |
+
|
| 16 |
+
|
| 17 |
+
namespace at {
|
| 18 |
+
namespace native {
|
| 19 |
+
TORCH_API at::Tensor & _copy_from_out(const at::Tensor & self, const at::Tensor & dst, bool non_blocking, at::Tensor & out);
|
| 20 |
+
} // namespace native
|
| 21 |
+
} // namespace at
|
vllm/lib/python3.10/site-packages/torch/include/ATen/ops/_embedding_bag_ops.h
ADDED
|
@@ -0,0 +1,39 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#pragma once
|
| 2 |
+
|
| 3 |
+
// @generated by torchgen/gen.py from Operator.h
|
| 4 |
+
|
| 5 |
+
#include <tuple>
|
| 6 |
+
#include <vector>
|
| 7 |
+
|
| 8 |
+
// Forward declarations of any types needed in the operator signatures.
|
| 9 |
+
// We can't directly include these classes because it will cause circular include dependencies.
|
| 10 |
+
// This file is included by TensorBody.h, which defines the Tensor class.
|
| 11 |
+
#include <ATen/core/ATen_fwd.h>
|
| 12 |
+
|
| 13 |
+
namespace at {
|
| 14 |
+
namespace _ops {
|
| 15 |
+
|
| 16 |
+
|
| 17 |
+
struct TORCH_API _embedding_bag {
|
| 18 |
+
using schema = ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor> (const at::Tensor &, const at::Tensor &, const at::Tensor &, bool, int64_t, bool, const ::std::optional<at::Tensor> &, bool, int64_t);
|
| 19 |
+
using ptr_schema = schema*;
|
| 20 |
+
// See Note [static constexpr char* members for windows NVCC]
|
| 21 |
+
STATIC_CONSTEXPR_STR_INL_EXCEPT_WIN_CUDA(name, "aten::_embedding_bag")
|
| 22 |
+
STATIC_CONSTEXPR_STR_INL_EXCEPT_WIN_CUDA(overload_name, "")
|
| 23 |
+
STATIC_CONSTEXPR_STR_INL_EXCEPT_WIN_CUDA(schema_str, "_embedding_bag(Tensor weight, Tensor indices, Tensor offsets, bool scale_grad_by_freq=False, int mode=0, bool sparse=False, Tensor? per_sample_weights=None, bool include_last_offset=False, int padding_idx=-1) -> (Tensor, Tensor, Tensor, Tensor)")
|
| 24 |
+
static ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor> call(const at::Tensor & weight, const at::Tensor & indices, const at::Tensor & offsets, bool scale_grad_by_freq, int64_t mode, bool sparse, const ::std::optional<at::Tensor> & per_sample_weights, bool include_last_offset, int64_t padding_idx);
|
| 25 |
+
static ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor> redispatch(c10::DispatchKeySet dispatchKeySet, const at::Tensor & weight, const at::Tensor & indices, const at::Tensor & offsets, bool scale_grad_by_freq, int64_t mode, bool sparse, const ::std::optional<at::Tensor> & per_sample_weights, bool include_last_offset, int64_t padding_idx);
|
| 26 |
+
};
|
| 27 |
+
|
| 28 |
+
struct TORCH_API _embedding_bag_out {
|
| 29 |
+
using schema = ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &,at::Tensor &> (const at::Tensor &, const at::Tensor &, const at::Tensor &, bool, int64_t, bool, const ::std::optional<at::Tensor> &, bool, int64_t, at::Tensor &, at::Tensor &, at::Tensor &, at::Tensor &);
|
| 30 |
+
using ptr_schema = schema*;
|
| 31 |
+
// See Note [static constexpr char* members for windows NVCC]
|
| 32 |
+
STATIC_CONSTEXPR_STR_INL_EXCEPT_WIN_CUDA(name, "aten::_embedding_bag")
|
| 33 |
+
STATIC_CONSTEXPR_STR_INL_EXCEPT_WIN_CUDA(overload_name, "out")
|
| 34 |
+
STATIC_CONSTEXPR_STR_INL_EXCEPT_WIN_CUDA(schema_str, "_embedding_bag.out(Tensor weight, Tensor indices, Tensor offsets, bool scale_grad_by_freq=False, int mode=0, bool sparse=False, Tensor? per_sample_weights=None, bool include_last_offset=False, int padding_idx=-1, *, Tensor(a!) out0, Tensor(b!) out1, Tensor(c!) out2, Tensor(d!) out3) -> (Tensor(a!), Tensor(b!), Tensor(c!), Tensor(d!))")
|
| 35 |
+
static ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &,at::Tensor &> call(const at::Tensor & weight, const at::Tensor & indices, const at::Tensor & offsets, bool scale_grad_by_freq, int64_t mode, bool sparse, const ::std::optional<at::Tensor> & per_sample_weights, bool include_last_offset, int64_t padding_idx, at::Tensor & out0, at::Tensor & out1, at::Tensor & out2, at::Tensor & out3);
|
| 36 |
+
static ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &,at::Tensor &> redispatch(c10::DispatchKeySet dispatchKeySet, const at::Tensor & weight, const at::Tensor & indices, const at::Tensor & offsets, bool scale_grad_by_freq, int64_t mode, bool sparse, const ::std::optional<at::Tensor> & per_sample_weights, bool include_last_offset, int64_t padding_idx, at::Tensor & out0, at::Tensor & out1, at::Tensor & out2, at::Tensor & out3);
|
| 37 |
+
};
|
| 38 |
+
|
| 39 |
+
}} // namespace at::_ops
|
vllm/lib/python3.10/site-packages/torch/include/ATen/ops/_foreach_tanh_ops.h
ADDED
|
@@ -0,0 +1,50 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#pragma once
|
| 2 |
+
|
| 3 |
+
// @generated by torchgen/gen.py from Operator.h
|
| 4 |
+
|
| 5 |
+
#include <tuple>
|
| 6 |
+
#include <vector>
|
| 7 |
+
|
| 8 |
+
// Forward declarations of any types needed in the operator signatures.
|
| 9 |
+
// We can't directly include these classes because it will cause circular include dependencies.
|
| 10 |
+
// This file is included by TensorBody.h, which defines the Tensor class.
|
| 11 |
+
#include <ATen/core/ATen_fwd.h>
|
| 12 |
+
|
| 13 |
+
namespace at {
|
| 14 |
+
namespace _ops {
|
| 15 |
+
|
| 16 |
+
|
| 17 |
+
struct TORCH_API _foreach_tanh {
|
| 18 |
+
using schema = ::std::vector<at::Tensor> (at::TensorList);
|
| 19 |
+
using ptr_schema = schema*;
|
| 20 |
+
// See Note [static constexpr char* members for windows NVCC]
|
| 21 |
+
STATIC_CONSTEXPR_STR_INL_EXCEPT_WIN_CUDA(name, "aten::_foreach_tanh")
|
| 22 |
+
STATIC_CONSTEXPR_STR_INL_EXCEPT_WIN_CUDA(overload_name, "")
|
| 23 |
+
STATIC_CONSTEXPR_STR_INL_EXCEPT_WIN_CUDA(schema_str, "_foreach_tanh(Tensor[] self) -> Tensor[]")
|
| 24 |
+
static ::std::vector<at::Tensor> call(at::TensorList self);
|
| 25 |
+
static ::std::vector<at::Tensor> redispatch(c10::DispatchKeySet dispatchKeySet, at::TensorList self);
|
| 26 |
+
};
|
| 27 |
+
|
| 28 |
+
struct TORCH_API _foreach_tanh_ {
|
| 29 |
+
using schema = void (at::TensorList);
|
| 30 |
+
using ptr_schema = schema*;
|
| 31 |
+
// See Note [static constexpr char* members for windows NVCC]
|
| 32 |
+
STATIC_CONSTEXPR_STR_INL_EXCEPT_WIN_CUDA(name, "aten::_foreach_tanh_")
|
| 33 |
+
STATIC_CONSTEXPR_STR_INL_EXCEPT_WIN_CUDA(overload_name, "")
|
| 34 |
+
STATIC_CONSTEXPR_STR_INL_EXCEPT_WIN_CUDA(schema_str, "_foreach_tanh_(Tensor(a!)[] self) -> ()")
|
| 35 |
+
static void call(at::TensorList self);
|
| 36 |
+
static void redispatch(c10::DispatchKeySet dispatchKeySet, at::TensorList self);
|
| 37 |
+
};
|
| 38 |
+
|
| 39 |
+
struct TORCH_API _foreach_tanh_out {
|
| 40 |
+
using schema = void (at::TensorList, at::TensorList);
|
| 41 |
+
using ptr_schema = schema*;
|
| 42 |
+
// See Note [static constexpr char* members for windows NVCC]
|
| 43 |
+
STATIC_CONSTEXPR_STR_INL_EXCEPT_WIN_CUDA(name, "aten::_foreach_tanh")
|
| 44 |
+
STATIC_CONSTEXPR_STR_INL_EXCEPT_WIN_CUDA(overload_name, "out")
|
| 45 |
+
STATIC_CONSTEXPR_STR_INL_EXCEPT_WIN_CUDA(schema_str, "_foreach_tanh.out(Tensor[] self, *, Tensor(a!)[] out) -> ()")
|
| 46 |
+
static void call(at::TensorList self, at::TensorList out);
|
| 47 |
+
static void redispatch(c10::DispatchKeySet dispatchKeySet, at::TensorList self, at::TensorList out);
|
| 48 |
+
};
|
| 49 |
+
|
| 50 |
+
}} // namespace at::_ops
|
vllm/lib/python3.10/site-packages/torch/include/ATen/ops/_sobol_engine_draw_ops.h
ADDED
|
@@ -0,0 +1,28 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#pragma once
|
| 2 |
+
|
| 3 |
+
// @generated by torchgen/gen.py from Operator.h
|
| 4 |
+
|
| 5 |
+
#include <tuple>
|
| 6 |
+
#include <vector>
|
| 7 |
+
|
| 8 |
+
// Forward declarations of any types needed in the operator signatures.
|
| 9 |
+
// We can't directly include these classes because it will cause circular include dependencies.
|
| 10 |
+
// This file is included by TensorBody.h, which defines the Tensor class.
|
| 11 |
+
#include <ATen/core/ATen_fwd.h>
|
| 12 |
+
|
| 13 |
+
namespace at {
|
| 14 |
+
namespace _ops {
|
| 15 |
+
|
| 16 |
+
|
| 17 |
+
struct TORCH_API _sobol_engine_draw {
|
| 18 |
+
using schema = ::std::tuple<at::Tensor,at::Tensor> (const at::Tensor &, int64_t, const at::Tensor &, int64_t, int64_t, ::std::optional<at::ScalarType>);
|
| 19 |
+
using ptr_schema = schema*;
|
| 20 |
+
// See Note [static constexpr char* members for windows NVCC]
|
| 21 |
+
STATIC_CONSTEXPR_STR_INL_EXCEPT_WIN_CUDA(name, "aten::_sobol_engine_draw")
|
| 22 |
+
STATIC_CONSTEXPR_STR_INL_EXCEPT_WIN_CUDA(overload_name, "")
|
| 23 |
+
STATIC_CONSTEXPR_STR_INL_EXCEPT_WIN_CUDA(schema_str, "_sobol_engine_draw(Tensor quasi, int n, Tensor sobolstate, int dimension, int num_generated, ScalarType? dtype) -> (Tensor, Tensor)")
|
| 24 |
+
static ::std::tuple<at::Tensor,at::Tensor> call(const at::Tensor & quasi, int64_t n, const at::Tensor & sobolstate, int64_t dimension, int64_t num_generated, ::std::optional<at::ScalarType> dtype);
|
| 25 |
+
static ::std::tuple<at::Tensor,at::Tensor> redispatch(c10::DispatchKeySet dispatchKeySet, const at::Tensor & quasi, int64_t n, const at::Tensor & sobolstate, int64_t dimension, int64_t num_generated, ::std::optional<at::ScalarType> dtype);
|
| 26 |
+
};
|
| 27 |
+
|
| 28 |
+
}} // namespace at::_ops
|
vllm/lib/python3.10/site-packages/torch/include/ATen/ops/_sparse_semi_structured_tile_ops.h
ADDED
|
@@ -0,0 +1,28 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#pragma once
|
| 2 |
+
|
| 3 |
+
// @generated by torchgen/gen.py from Operator.h
|
| 4 |
+
|
| 5 |
+
#include <tuple>
|
| 6 |
+
#include <vector>
|
| 7 |
+
|
| 8 |
+
// Forward declarations of any types needed in the operator signatures.
|
| 9 |
+
// We can't directly include these classes because it will cause circular include dependencies.
|
| 10 |
+
// This file is included by TensorBody.h, which defines the Tensor class.
|
| 11 |
+
#include <ATen/core/ATen_fwd.h>
|
| 12 |
+
|
| 13 |
+
namespace at {
|
| 14 |
+
namespace _ops {
|
| 15 |
+
|
| 16 |
+
|
| 17 |
+
struct TORCH_API _sparse_semi_structured_tile {
|
| 18 |
+
using schema = ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor,at::Tensor> (const at::Tensor &, c10::string_view, bool);
|
| 19 |
+
using ptr_schema = schema*;
|
| 20 |
+
// See Note [static constexpr char* members for windows NVCC]
|
| 21 |
+
STATIC_CONSTEXPR_STR_INL_EXCEPT_WIN_CUDA(name, "aten::_sparse_semi_structured_tile")
|
| 22 |
+
STATIC_CONSTEXPR_STR_INL_EXCEPT_WIN_CUDA(overload_name, "")
|
| 23 |
+
STATIC_CONSTEXPR_STR_INL_EXCEPT_WIN_CUDA(schema_str, "_sparse_semi_structured_tile(Tensor input, str algorithm=\"\", bool use_cutlass=True) -> (Tensor, Tensor, Tensor, Tensor, Tensor)")
|
| 24 |
+
static ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor,at::Tensor> call(const at::Tensor & input, c10::string_view algorithm, bool use_cutlass);
|
| 25 |
+
static ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor,at::Tensor> redispatch(c10::DispatchKeySet dispatchKeySet, const at::Tensor & input, c10::string_view algorithm, bool use_cutlass);
|
| 26 |
+
};
|
| 27 |
+
|
| 28 |
+
}} // namespace at::_ops
|
vllm/lib/python3.10/site-packages/torch/include/ATen/ops/_thnn_fused_lstm_cell_cuda_dispatch.h
ADDED
|
@@ -0,0 +1,23 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#pragma once
|
| 2 |
+
// @generated by torchgen/gen.py from DispatchKeyFunction.h
|
| 3 |
+
|
| 4 |
+
// NB: The implementing C++ file is RegisterDispatchKey.cpp
|
| 5 |
+
|
| 6 |
+
// The only #includes we need are for custom classes that have defaults in the C++ API
|
| 7 |
+
#include <c10/core/MemoryFormat.h>
|
| 8 |
+
#include <c10/core/Scalar.h>
|
| 9 |
+
#include <ATen/core/Reduction.h>
|
| 10 |
+
|
| 11 |
+
// Forward declarations of any types needed in the operator signatures.
|
| 12 |
+
// We can't directly include these classes because it will cause circular include dependencies.
|
| 13 |
+
// This file is included by TensorBody.h, which defines the Tensor class.
|
| 14 |
+
#include <ATen/core/ATen_fwd.h>
|
| 15 |
+
|
| 16 |
+
namespace at {
|
| 17 |
+
|
| 18 |
+
namespace cuda {
|
| 19 |
+
|
| 20 |
+
TORCH_API ::std::tuple<at::Tensor,at::Tensor,at::Tensor> _thnn_fused_lstm_cell(const at::Tensor & input_gates, const at::Tensor & hidden_gates, const at::Tensor & cx, const ::std::optional<at::Tensor> & input_bias={}, const ::std::optional<at::Tensor> & hidden_bias={});
|
| 21 |
+
|
| 22 |
+
} // namespace cuda
|
| 23 |
+
} // namespace at
|
vllm/lib/python3.10/site-packages/torch/include/ATen/ops/arccosh_native.h
ADDED
|
@@ -0,0 +1,23 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#pragma once
|
| 2 |
+
|
| 3 |
+
// @generated by torchgen/gen.py from NativeFunction.h
|
| 4 |
+
|
| 5 |
+
#include <c10/core/Scalar.h>
|
| 6 |
+
#include <c10/core/Storage.h>
|
| 7 |
+
#include <c10/core/TensorOptions.h>
|
| 8 |
+
#include <c10/util/Deprecated.h>
|
| 9 |
+
#include <optional>
|
| 10 |
+
#include <c10/core/QScheme.h>
|
| 11 |
+
#include <ATen/core/Reduction.h>
|
| 12 |
+
#include <ATen/core/Tensor.h>
|
| 13 |
+
#include <tuple>
|
| 14 |
+
#include <vector>
|
| 15 |
+
|
| 16 |
+
|
| 17 |
+
namespace at {
|
| 18 |
+
namespace native {
|
| 19 |
+
TORCH_API at::Tensor arccosh(const at::Tensor & self);
|
| 20 |
+
TORCH_API at::Tensor & arccosh_out(const at::Tensor & self, at::Tensor & out);
|
| 21 |
+
TORCH_API at::Tensor & arccosh_(at::Tensor & self);
|
| 22 |
+
} // namespace native
|
| 23 |
+
} // namespace at
|
vllm/lib/python3.10/site-packages/torch/include/ATen/ops/batch_norm_update_stats.h
ADDED
|
@@ -0,0 +1,39 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#pragma once
|
| 2 |
+
|
| 3 |
+
// @generated by torchgen/gen.py from Function.h
|
| 4 |
+
|
| 5 |
+
#include <ATen/Context.h>
|
| 6 |
+
#include <ATen/DeviceGuard.h>
|
| 7 |
+
#include <ATen/TensorUtils.h>
|
| 8 |
+
#include <ATen/TracerMode.h>
|
| 9 |
+
#include <ATen/core/Generator.h>
|
| 10 |
+
#include <ATen/core/Reduction.h>
|
| 11 |
+
#include <ATen/core/Tensor.h>
|
| 12 |
+
#include <c10/core/Scalar.h>
|
| 13 |
+
#include <c10/core/Storage.h>
|
| 14 |
+
#include <c10/core/TensorOptions.h>
|
| 15 |
+
#include <c10/util/Deprecated.h>
|
| 16 |
+
#include <optional>
|
| 17 |
+
|
| 18 |
+
|
| 19 |
+
|
| 20 |
+
#include <ATen/ops/batch_norm_update_stats_ops.h>
|
| 21 |
+
|
| 22 |
+
namespace at {
|
| 23 |
+
|
| 24 |
+
|
| 25 |
+
// aten::batch_norm_update_stats(Tensor input, Tensor? running_mean, Tensor? running_var, float momentum) -> (Tensor, Tensor)
|
| 26 |
+
inline ::std::tuple<at::Tensor,at::Tensor> batch_norm_update_stats(const at::Tensor & input, const ::std::optional<at::Tensor> & running_mean, const ::std::optional<at::Tensor> & running_var, double momentum) {
|
| 27 |
+
return at::_ops::batch_norm_update_stats::call(input, running_mean, running_var, momentum);
|
| 28 |
+
}
|
| 29 |
+
|
| 30 |
+
// aten::batch_norm_update_stats.out(Tensor input, Tensor? running_mean, Tensor? running_var, float momentum, *, Tensor(a!) out0, Tensor(b!) out1) -> (Tensor(a!), Tensor(b!))
|
| 31 |
+
inline ::std::tuple<at::Tensor &,at::Tensor &> batch_norm_update_stats_out(at::Tensor & out0, at::Tensor & out1, const at::Tensor & input, const ::std::optional<at::Tensor> & running_mean, const ::std::optional<at::Tensor> & running_var, double momentum) {
|
| 32 |
+
return at::_ops::batch_norm_update_stats_out::call(input, running_mean, running_var, momentum, out0, out1);
|
| 33 |
+
}
|
| 34 |
+
// aten::batch_norm_update_stats.out(Tensor input, Tensor? running_mean, Tensor? running_var, float momentum, *, Tensor(a!) out0, Tensor(b!) out1) -> (Tensor(a!), Tensor(b!))
|
| 35 |
+
inline ::std::tuple<at::Tensor &,at::Tensor &> batch_norm_update_stats_outf(const at::Tensor & input, const ::std::optional<at::Tensor> & running_mean, const ::std::optional<at::Tensor> & running_var, double momentum, at::Tensor & out0, at::Tensor & out1) {
|
| 36 |
+
return at::_ops::batch_norm_update_stats_out::call(input, running_mean, running_var, momentum, out0, out1);
|
| 37 |
+
}
|
| 38 |
+
|
| 39 |
+
}
|
vllm/lib/python3.10/site-packages/torch/include/ATen/ops/binary_cross_entropy_with_logits_compositeexplicitautograd_dispatch.h
ADDED
|
@@ -0,0 +1,25 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#pragma once
|
| 2 |
+
// @generated by torchgen/gen.py from DispatchKeyFunction.h
|
| 3 |
+
|
| 4 |
+
// NB: The implementing C++ file is RegisterDispatchKey.cpp
|
| 5 |
+
|
| 6 |
+
// The only #includes we need are for custom classes that have defaults in the C++ API
|
| 7 |
+
#include <c10/core/MemoryFormat.h>
|
| 8 |
+
#include <c10/core/Scalar.h>
|
| 9 |
+
#include <ATen/core/Reduction.h>
|
| 10 |
+
|
| 11 |
+
// Forward declarations of any types needed in the operator signatures.
|
| 12 |
+
// We can't directly include these classes because it will cause circular include dependencies.
|
| 13 |
+
// This file is included by TensorBody.h, which defines the Tensor class.
|
| 14 |
+
#include <ATen/core/ATen_fwd.h>
|
| 15 |
+
|
| 16 |
+
namespace at {
|
| 17 |
+
|
| 18 |
+
namespace compositeexplicitautograd {
|
| 19 |
+
|
| 20 |
+
TORCH_API at::Tensor binary_cross_entropy_with_logits(const at::Tensor & self, const at::Tensor & target, const ::std::optional<at::Tensor> & weight={}, const ::std::optional<at::Tensor> & pos_weight={}, int64_t reduction=at::Reduction::Mean);
|
| 21 |
+
TORCH_API at::Tensor & binary_cross_entropy_with_logits_out(at::Tensor & out, const at::Tensor & self, const at::Tensor & target, const ::std::optional<at::Tensor> & weight={}, const ::std::optional<at::Tensor> & pos_weight={}, int64_t reduction=at::Reduction::Mean);
|
| 22 |
+
TORCH_API at::Tensor & binary_cross_entropy_with_logits_outf(const at::Tensor & self, const at::Tensor & target, const ::std::optional<at::Tensor> & weight, const ::std::optional<at::Tensor> & pos_weight, int64_t reduction, at::Tensor & out);
|
| 23 |
+
|
| 24 |
+
} // namespace compositeexplicitautograd
|
| 25 |
+
} // namespace at
|
vllm/lib/python3.10/site-packages/torch/include/ATen/ops/conv_transpose3d_compositeimplicitautograd_dispatch.h
ADDED
|
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#pragma once
|
| 2 |
+
// @generated by torchgen/gen.py from DispatchKeyFunction.h
|
| 3 |
+
|
| 4 |
+
// NB: The implementing C++ file is RegisterDispatchKey.cpp
|
| 5 |
+
|
| 6 |
+
// The only #includes we need are for custom classes that have defaults in the C++ API
|
| 7 |
+
#include <c10/core/MemoryFormat.h>
|
| 8 |
+
#include <c10/core/Scalar.h>
|
| 9 |
+
#include <ATen/core/Reduction.h>
|
| 10 |
+
|
| 11 |
+
// Forward declarations of any types needed in the operator signatures.
|
| 12 |
+
// We can't directly include these classes because it will cause circular include dependencies.
|
| 13 |
+
// This file is included by TensorBody.h, which defines the Tensor class.
|
| 14 |
+
#include <ATen/core/ATen_fwd.h>
|
| 15 |
+
|
| 16 |
+
namespace at {
|
| 17 |
+
|
| 18 |
+
namespace compositeimplicitautograd {
|
| 19 |
+
|
| 20 |
+
TORCH_API at::Tensor conv_transpose3d(const at::Tensor & input, const at::Tensor & weight, const ::std::optional<at::Tensor> & bias={}, at::IntArrayRef stride=1, at::IntArrayRef padding=0, at::IntArrayRef output_padding=0, int64_t groups=1, at::IntArrayRef dilation=1);
|
| 21 |
+
TORCH_API at::Tensor conv_transpose3d_symint(const at::Tensor & input, const at::Tensor & weight, const ::std::optional<at::Tensor> & bias={}, c10::SymIntArrayRef stride=c10::SymInt(1), c10::SymIntArrayRef padding=c10::SymInt(0), c10::SymIntArrayRef output_padding=c10::SymInt(0), c10::SymInt groups=1, c10::SymIntArrayRef dilation=c10::SymInt(1));
|
| 22 |
+
|
| 23 |
+
} // namespace compositeimplicitautograd
|
| 24 |
+
} // namespace at
|
vllm/lib/python3.10/site-packages/torch/include/ATen/ops/corrcoef_native.h
ADDED
|
@@ -0,0 +1,21 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#pragma once
|
| 2 |
+
|
| 3 |
+
// @generated by torchgen/gen.py from NativeFunction.h
|
| 4 |
+
|
| 5 |
+
#include <c10/core/Scalar.h>
|
| 6 |
+
#include <c10/core/Storage.h>
|
| 7 |
+
#include <c10/core/TensorOptions.h>
|
| 8 |
+
#include <c10/util/Deprecated.h>
|
| 9 |
+
#include <optional>
|
| 10 |
+
#include <c10/core/QScheme.h>
|
| 11 |
+
#include <ATen/core/Reduction.h>
|
| 12 |
+
#include <ATen/core/Tensor.h>
|
| 13 |
+
#include <tuple>
|
| 14 |
+
#include <vector>
|
| 15 |
+
|
| 16 |
+
|
| 17 |
+
namespace at {
|
| 18 |
+
namespace native {
|
| 19 |
+
TORCH_API at::Tensor corrcoef(const at::Tensor & self);
|
| 20 |
+
} // namespace native
|
| 21 |
+
} // namespace at
|
vllm/lib/python3.10/site-packages/torch/include/ATen/ops/cudnn_convolution_add_relu.h
ADDED
|
@@ -0,0 +1,91 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#pragma once
|
| 2 |
+
|
| 3 |
+
// @generated by torchgen/gen.py from Function.h
|
| 4 |
+
|
| 5 |
+
#include <ATen/Context.h>
|
| 6 |
+
#include <ATen/DeviceGuard.h>
|
| 7 |
+
#include <ATen/TensorUtils.h>
|
| 8 |
+
#include <ATen/TracerMode.h>
|
| 9 |
+
#include <ATen/core/Generator.h>
|
| 10 |
+
#include <ATen/core/Reduction.h>
|
| 11 |
+
#include <ATen/core/Tensor.h>
|
| 12 |
+
#include <c10/core/Scalar.h>
|
| 13 |
+
#include <c10/core/Storage.h>
|
| 14 |
+
#include <c10/core/TensorOptions.h>
|
| 15 |
+
#include <c10/util/Deprecated.h>
|
| 16 |
+
#include <optional>
|
| 17 |
+
|
| 18 |
+
|
| 19 |
+
|
| 20 |
+
#include <ATen/ops/cudnn_convolution_add_relu_ops.h>
|
| 21 |
+
|
| 22 |
+
namespace at {
|
| 23 |
+
|
| 24 |
+
|
| 25 |
+
// aten::cudnn_convolution_add_relu(Tensor self, Tensor weight, Tensor z, Scalar? alpha, Tensor? bias, SymInt[] stride, SymInt[] padding, SymInt[] dilation, SymInt groups) -> Tensor
|
| 26 |
+
inline at::Tensor cudnn_convolution_add_relu(const at::Tensor & self, const at::Tensor & weight, const at::Tensor & z, const ::std::optional<at::Scalar> & alpha, const ::std::optional<at::Tensor> & bias, at::IntArrayRef stride, at::IntArrayRef padding, at::IntArrayRef dilation, int64_t groups) {
|
| 27 |
+
return at::_ops::cudnn_convolution_add_relu::call(self, weight, z, alpha, bias, c10::fromIntArrayRefSlow(stride), c10::fromIntArrayRefSlow(padding), c10::fromIntArrayRefSlow(dilation), groups);
|
| 28 |
+
}
|
| 29 |
+
namespace symint {
|
| 30 |
+
template <typename T, typename = std::enable_if_t<std::is_same<T, int64_t>::value>>
|
| 31 |
+
at::Tensor cudnn_convolution_add_relu(const at::Tensor & self, const at::Tensor & weight, const at::Tensor & z, const ::std::optional<at::Scalar> & alpha, const ::std::optional<at::Tensor> & bias, at::IntArrayRef stride, at::IntArrayRef padding, at::IntArrayRef dilation, int64_t groups) {
|
| 32 |
+
return at::_ops::cudnn_convolution_add_relu::call(self, weight, z, alpha, bias, c10::fromIntArrayRefSlow(stride), c10::fromIntArrayRefSlow(padding), c10::fromIntArrayRefSlow(dilation), groups);
|
| 33 |
+
}
|
| 34 |
+
}
|
| 35 |
+
|
| 36 |
+
// aten::cudnn_convolution_add_relu(Tensor self, Tensor weight, Tensor z, Scalar? alpha, Tensor? bias, SymInt[] stride, SymInt[] padding, SymInt[] dilation, SymInt groups) -> Tensor
|
| 37 |
+
inline at::Tensor cudnn_convolution_add_relu_symint(const at::Tensor & self, const at::Tensor & weight, const at::Tensor & z, const ::std::optional<at::Scalar> & alpha, const ::std::optional<at::Tensor> & bias, c10::SymIntArrayRef stride, c10::SymIntArrayRef padding, c10::SymIntArrayRef dilation, c10::SymInt groups) {
|
| 38 |
+
return at::_ops::cudnn_convolution_add_relu::call(self, weight, z, alpha, bias, stride, padding, dilation, groups);
|
| 39 |
+
}
|
| 40 |
+
namespace symint {
|
| 41 |
+
template <typename T, typename = std::enable_if_t<std::is_same<T, c10::SymInt>::value>>
|
| 42 |
+
at::Tensor cudnn_convolution_add_relu(const at::Tensor & self, const at::Tensor & weight, const at::Tensor & z, const ::std::optional<at::Scalar> & alpha, const ::std::optional<at::Tensor> & bias, c10::SymIntArrayRef stride, c10::SymIntArrayRef padding, c10::SymIntArrayRef dilation, c10::SymInt groups) {
|
| 43 |
+
return at::_ops::cudnn_convolution_add_relu::call(self, weight, z, alpha, bias, stride, padding, dilation, groups);
|
| 44 |
+
}
|
| 45 |
+
}
|
| 46 |
+
|
| 47 |
+
// aten::cudnn_convolution_add_relu.out(Tensor self, Tensor weight, Tensor z, Scalar? alpha, Tensor? bias, SymInt[] stride, SymInt[] padding, SymInt[] dilation, SymInt groups, *, Tensor(a!) out) -> Tensor(a!)
|
| 48 |
+
inline at::Tensor & cudnn_convolution_add_relu_out(at::Tensor & out, const at::Tensor & self, const at::Tensor & weight, const at::Tensor & z, const ::std::optional<at::Scalar> & alpha, const ::std::optional<at::Tensor> & bias, at::IntArrayRef stride, at::IntArrayRef padding, at::IntArrayRef dilation, int64_t groups) {
|
| 49 |
+
return at::_ops::cudnn_convolution_add_relu_out::call(self, weight, z, alpha, bias, c10::fromIntArrayRefSlow(stride), c10::fromIntArrayRefSlow(padding), c10::fromIntArrayRefSlow(dilation), groups, out);
|
| 50 |
+
}
|
| 51 |
+
namespace symint {
|
| 52 |
+
template <typename T, typename = std::enable_if_t<std::is_same<T, int64_t>::value>>
|
| 53 |
+
at::Tensor & cudnn_convolution_add_relu_out(at::Tensor & out, const at::Tensor & self, const at::Tensor & weight, const at::Tensor & z, const ::std::optional<at::Scalar> & alpha, const ::std::optional<at::Tensor> & bias, at::IntArrayRef stride, at::IntArrayRef padding, at::IntArrayRef dilation, int64_t groups) {
|
| 54 |
+
return at::_ops::cudnn_convolution_add_relu_out::call(self, weight, z, alpha, bias, c10::fromIntArrayRefSlow(stride), c10::fromIntArrayRefSlow(padding), c10::fromIntArrayRefSlow(dilation), groups, out);
|
| 55 |
+
}
|
| 56 |
+
}
|
| 57 |
+
|
| 58 |
+
// aten::cudnn_convolution_add_relu.out(Tensor self, Tensor weight, Tensor z, Scalar? alpha, Tensor? bias, SymInt[] stride, SymInt[] padding, SymInt[] dilation, SymInt groups, *, Tensor(a!) out) -> Tensor(a!)
|
| 59 |
+
inline at::Tensor & cudnn_convolution_add_relu_outf(const at::Tensor & self, const at::Tensor & weight, const at::Tensor & z, const ::std::optional<at::Scalar> & alpha, const ::std::optional<at::Tensor> & bias, at::IntArrayRef stride, at::IntArrayRef padding, at::IntArrayRef dilation, int64_t groups, at::Tensor & out) {
|
| 60 |
+
return at::_ops::cudnn_convolution_add_relu_out::call(self, weight, z, alpha, bias, c10::fromIntArrayRefSlow(stride), c10::fromIntArrayRefSlow(padding), c10::fromIntArrayRefSlow(dilation), groups, out);
|
| 61 |
+
}
|
| 62 |
+
namespace symint {
|
| 63 |
+
template <typename T, typename = std::enable_if_t<std::is_same<T, int64_t>::value>>
|
| 64 |
+
at::Tensor & cudnn_convolution_add_relu_outf(const at::Tensor & self, const at::Tensor & weight, const at::Tensor & z, const ::std::optional<at::Scalar> & alpha, const ::std::optional<at::Tensor> & bias, at::IntArrayRef stride, at::IntArrayRef padding, at::IntArrayRef dilation, int64_t groups, at::Tensor & out) {
|
| 65 |
+
return at::_ops::cudnn_convolution_add_relu_out::call(self, weight, z, alpha, bias, c10::fromIntArrayRefSlow(stride), c10::fromIntArrayRefSlow(padding), c10::fromIntArrayRefSlow(dilation), groups, out);
|
| 66 |
+
}
|
| 67 |
+
}
|
| 68 |
+
|
| 69 |
+
// aten::cudnn_convolution_add_relu.out(Tensor self, Tensor weight, Tensor z, Scalar? alpha, Tensor? bias, SymInt[] stride, SymInt[] padding, SymInt[] dilation, SymInt groups, *, Tensor(a!) out) -> Tensor(a!)
|
| 70 |
+
inline at::Tensor & cudnn_convolution_add_relu_symint_out(at::Tensor & out, const at::Tensor & self, const at::Tensor & weight, const at::Tensor & z, const ::std::optional<at::Scalar> & alpha, const ::std::optional<at::Tensor> & bias, c10::SymIntArrayRef stride, c10::SymIntArrayRef padding, c10::SymIntArrayRef dilation, c10::SymInt groups) {
|
| 71 |
+
return at::_ops::cudnn_convolution_add_relu_out::call(self, weight, z, alpha, bias, stride, padding, dilation, groups, out);
|
| 72 |
+
}
|
| 73 |
+
namespace symint {
|
| 74 |
+
template <typename T, typename = std::enable_if_t<std::is_same<T, c10::SymInt>::value>>
|
| 75 |
+
at::Tensor & cudnn_convolution_add_relu_out(at::Tensor & out, const at::Tensor & self, const at::Tensor & weight, const at::Tensor & z, const ::std::optional<at::Scalar> & alpha, const ::std::optional<at::Tensor> & bias, c10::SymIntArrayRef stride, c10::SymIntArrayRef padding, c10::SymIntArrayRef dilation, c10::SymInt groups) {
|
| 76 |
+
return at::_ops::cudnn_convolution_add_relu_out::call(self, weight, z, alpha, bias, stride, padding, dilation, groups, out);
|
| 77 |
+
}
|
| 78 |
+
}
|
| 79 |
+
|
| 80 |
+
// aten::cudnn_convolution_add_relu.out(Tensor self, Tensor weight, Tensor z, Scalar? alpha, Tensor? bias, SymInt[] stride, SymInt[] padding, SymInt[] dilation, SymInt groups, *, Tensor(a!) out) -> Tensor(a!)
|
| 81 |
+
inline at::Tensor & cudnn_convolution_add_relu_symint_outf(const at::Tensor & self, const at::Tensor & weight, const at::Tensor & z, const ::std::optional<at::Scalar> & alpha, const ::std::optional<at::Tensor> & bias, c10::SymIntArrayRef stride, c10::SymIntArrayRef padding, c10::SymIntArrayRef dilation, c10::SymInt groups, at::Tensor & out) {
|
| 82 |
+
return at::_ops::cudnn_convolution_add_relu_out::call(self, weight, z, alpha, bias, stride, padding, dilation, groups, out);
|
| 83 |
+
}
|
| 84 |
+
namespace symint {
|
| 85 |
+
template <typename T, typename = std::enable_if_t<std::is_same<T, c10::SymInt>::value>>
|
| 86 |
+
at::Tensor & cudnn_convolution_add_relu_outf(const at::Tensor & self, const at::Tensor & weight, const at::Tensor & z, const ::std::optional<at::Scalar> & alpha, const ::std::optional<at::Tensor> & bias, c10::SymIntArrayRef stride, c10::SymIntArrayRef padding, c10::SymIntArrayRef dilation, c10::SymInt groups, at::Tensor & out) {
|
| 87 |
+
return at::_ops::cudnn_convolution_add_relu_out::call(self, weight, z, alpha, bias, stride, padding, dilation, groups, out);
|
| 88 |
+
}
|
| 89 |
+
}
|
| 90 |
+
|
| 91 |
+
}
|
vllm/lib/python3.10/site-packages/torch/include/ATen/ops/deg2rad.h
ADDED
|
@@ -0,0 +1,44 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#pragma once
|
| 2 |
+
|
| 3 |
+
// @generated by torchgen/gen.py from Function.h
|
| 4 |
+
|
| 5 |
+
#include <ATen/Context.h>
|
| 6 |
+
#include <ATen/DeviceGuard.h>
|
| 7 |
+
#include <ATen/TensorUtils.h>
|
| 8 |
+
#include <ATen/TracerMode.h>
|
| 9 |
+
#include <ATen/core/Generator.h>
|
| 10 |
+
#include <ATen/core/Reduction.h>
|
| 11 |
+
#include <ATen/core/Tensor.h>
|
| 12 |
+
#include <c10/core/Scalar.h>
|
| 13 |
+
#include <c10/core/Storage.h>
|
| 14 |
+
#include <c10/core/TensorOptions.h>
|
| 15 |
+
#include <c10/util/Deprecated.h>
|
| 16 |
+
#include <optional>
|
| 17 |
+
|
| 18 |
+
|
| 19 |
+
|
| 20 |
+
#include <ATen/ops/deg2rad_ops.h>
|
| 21 |
+
|
| 22 |
+
namespace at {
|
| 23 |
+
|
| 24 |
+
|
| 25 |
+
// aten::deg2rad(Tensor self) -> Tensor
|
| 26 |
+
inline at::Tensor deg2rad(const at::Tensor & self) {
|
| 27 |
+
return at::_ops::deg2rad::call(self);
|
| 28 |
+
}
|
| 29 |
+
|
| 30 |
+
// aten::deg2rad_(Tensor(a!) self) -> Tensor(a!)
|
| 31 |
+
inline at::Tensor & deg2rad_(at::Tensor & self) {
|
| 32 |
+
return at::_ops::deg2rad_::call(self);
|
| 33 |
+
}
|
| 34 |
+
|
| 35 |
+
// aten::deg2rad.out(Tensor self, *, Tensor(a!) out) -> Tensor(a!)
|
| 36 |
+
inline at::Tensor & deg2rad_out(at::Tensor & out, const at::Tensor & self) {
|
| 37 |
+
return at::_ops::deg2rad_out::call(self, out);
|
| 38 |
+
}
|
| 39 |
+
// aten::deg2rad.out(Tensor self, *, Tensor(a!) out) -> Tensor(a!)
|
| 40 |
+
inline at::Tensor & deg2rad_outf(const at::Tensor & self, at::Tensor & out) {
|
| 41 |
+
return at::_ops::deg2rad_out::call(self, out);
|
| 42 |
+
}
|
| 43 |
+
|
| 44 |
+
}
|