-
Notifications
You must be signed in to change notification settings - Fork 387
/
specificity.py
394 lines (328 loc) · 18.1 KB
/
specificity.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
# Copyright The Lightning team.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from typing import Optional
from torch import Tensor
from typing_extensions import Literal
from torchmetrics.functional.classification.stat_scores import (
_binary_stat_scores_arg_validation,
_binary_stat_scores_format,
_binary_stat_scores_tensor_validation,
_binary_stat_scores_update,
_multiclass_stat_scores_arg_validation,
_multiclass_stat_scores_format,
_multiclass_stat_scores_tensor_validation,
_multiclass_stat_scores_update,
_multilabel_stat_scores_arg_validation,
_multilabel_stat_scores_format,
_multilabel_stat_scores_tensor_validation,
_multilabel_stat_scores_update,
)
from torchmetrics.utilities.compute import _adjust_weights_safe_divide, _safe_divide
from torchmetrics.utilities.enums import ClassificationTask
def _specificity_reduce(
tp: Tensor,
fp: Tensor,
tn: Tensor,
fn: Tensor,
average: Optional[Literal["binary", "micro", "macro", "weighted", "none"]],
multidim_average: Literal["global", "samplewise"] = "global",
multilabel: bool = False,
) -> Tensor:
if average == "binary":
return _safe_divide(tn, tn + fp)
if average == "micro":
tn = tn.sum(dim=0 if multidim_average == "global" else 1)
fp = fp.sum(dim=0 if multidim_average == "global" else 1)
return _safe_divide(tn, tn + fp)
specificity_score = _safe_divide(tn, tn + fp)
return _adjust_weights_safe_divide(specificity_score, average, multilabel, tp, fp, fn)
def binary_specificity(
preds: Tensor,
target: Tensor,
threshold: float = 0.5,
multidim_average: Literal["global", "samplewise"] = "global",
ignore_index: Optional[int] = None,
validate_args: bool = True,
) -> Tensor:
r"""Compute `Specificity`_ for binary tasks.
.. math:: \text{Specificity} = \frac{\text{TN}}{\text{TN} + \text{FP}}
Where :math:`\text{TN}` and :math:`\text{FP}` represent the number of true negatives and
false positives respecitively.
Accepts the following input tensors:
- ``preds`` (int or float tensor): ``(N, ...)``. If preds is a floating point tensor with values outside
[0,1] range we consider the input to be logits and will auto apply sigmoid per element. Addtionally,
we convert to int tensor with thresholding using the value in ``threshold``.
- ``target`` (int tensor): ``(N, ...)``
Args:
preds: Tensor with predictions
target: Tensor with true labels
threshold: Threshold for transforming probability to binary {0,1} predictions
multidim_average:
Defines how additionally dimensions ``...`` should be handled. Should be one of the following:
- ``global``: Additional dimensions are flatted along the batch dimension
- ``samplewise``: Statistic will be calculated independently for each sample on the ``N`` axis.
The statistics in this case are calculated over the additional dimensions.
ignore_index:
Specifies a target value that is ignored and does not contribute to the metric calculation
validate_args: bool indicating if input arguments and tensors should be validated for correctness.
Set to ``False`` for faster computations.
Returns:
If ``multidim_average`` is set to ``global``, the metric returns a scalar value. If ``multidim_average``
is set to ``samplewise``, the metric returns ``(N,)`` vector consisting of a scalar value per sample.
Example (preds is int tensor):
>>> from torch import tensor
>>> from torchmetrics.functional.classification import binary_specificity
>>> target = tensor([0, 1, 0, 1, 0, 1])
>>> preds = tensor([0, 0, 1, 1, 0, 1])
>>> binary_specificity(preds, target)
tensor(0.6667)
Example (preds is float tensor):
>>> from torchmetrics.functional.classification import binary_specificity
>>> target = tensor([0, 1, 0, 1, 0, 1])
>>> preds = tensor([0.11, 0.22, 0.84, 0.73, 0.33, 0.92])
>>> binary_specificity(preds, target)
tensor(0.6667)
Example (multidim tensors):
>>> from torchmetrics.functional.classification import binary_specificity
>>> target = tensor([[[0, 1], [1, 0], [0, 1]], [[1, 1], [0, 0], [1, 0]]])
>>> preds = tensor([[[0.59, 0.91], [0.91, 0.99], [0.63, 0.04]],
... [[0.38, 0.04], [0.86, 0.780], [0.45, 0.37]]])
>>> binary_specificity(preds, target, multidim_average='samplewise')
tensor([0.0000, 0.3333])
"""
if validate_args:
_binary_stat_scores_arg_validation(threshold, multidim_average, ignore_index)
_binary_stat_scores_tensor_validation(preds, target, multidim_average, ignore_index)
preds, target = _binary_stat_scores_format(preds, target, threshold, ignore_index)
tp, fp, tn, fn = _binary_stat_scores_update(preds, target, multidim_average)
return _specificity_reduce(tp, fp, tn, fn, average="binary", multidim_average=multidim_average)
def multiclass_specificity(
preds: Tensor,
target: Tensor,
num_classes: int,
average: Optional[Literal["micro", "macro", "weighted", "none"]] = "macro",
top_k: int = 1,
multidim_average: Literal["global", "samplewise"] = "global",
ignore_index: Optional[int] = None,
validate_args: bool = True,
) -> Tensor:
r"""Compute `Specificity`_ for multiclass tasks.
.. math:: \text{Specificity} = \frac{\text{TN}}{\text{TN} + \text{FP}}
Where :math:`\text{TN}` and :math:`\text{FP}` represent the number of true negatives and
false positives respecitively.
Accepts the following input tensors:
- ``preds``: ``(N, ...)`` (int tensor) or ``(N, C, ..)`` (float tensor). If preds is a floating point
we apply ``torch.argmax`` along the ``C`` dimension to automatically convert probabilities/logits into
an int tensor.
- ``target`` (int tensor): ``(N, ...)``
Args:
preds: Tensor with predictions
target: Tensor with true labels
num_classes: Integer specifing the number of classes
average:
Defines the reduction that is applied over labels. Should be one of the following:
- ``micro``: Sum statistics over all labels
- ``macro``: Calculate statistics for each label and average them
- ``weighted``: calculates statistics for each label and computes weighted average using their support
- ``"none"`` or ``None``: calculates statistic for each label and applies no reduction
top_k:
Number of highest probability or logit score predictions considered to find the correct label.
Only works when ``preds`` contain probabilities/logits.
multidim_average:
Defines how additionally dimensions ``...`` should be handled. Should be one of the following:
- ``global``: Additional dimensions are flatted along the batch dimension
- ``samplewise``: Statistic will be calculated independently for each sample on the ``N`` axis.
The statistics in this case are calculated over the additional dimensions.
ignore_index:
Specifies a target value that is ignored and does not contribute to the metric calculation
validate_args: bool indicating if input arguments and tensors should be validated for correctness.
Set to ``False`` for faster computations.
Returns:
The returned shape depends on the ``average`` and ``multidim_average`` arguments:
- If ``multidim_average`` is set to ``global``:
- If ``average='micro'/'macro'/'weighted'``, the output will be a scalar tensor
- If ``average=None/'none'``, the shape will be ``(C,)``
- If ``multidim_average`` is set to ``samplewise``:
- If ``average='micro'/'macro'/'weighted'``, the shape will be ``(N,)``
- If ``average=None/'none'``, the shape will be ``(N, C)``
Example (preds is int tensor):
>>> from torch import tensor
>>> from torchmetrics.functional.classification import multiclass_specificity
>>> target = tensor([2, 1, 0, 0])
>>> preds = tensor([2, 1, 0, 1])
>>> multiclass_specificity(preds, target, num_classes=3)
tensor(0.8889)
>>> multiclass_specificity(preds, target, num_classes=3, average=None)
tensor([1.0000, 0.6667, 1.0000])
Example (preds is float tensor):
>>> from torchmetrics.functional.classification import multiclass_specificity
>>> target = tensor([2, 1, 0, 0])
>>> preds = tensor([[0.16, 0.26, 0.58],
... [0.22, 0.61, 0.17],
... [0.71, 0.09, 0.20],
... [0.05, 0.82, 0.13]])
>>> multiclass_specificity(preds, target, num_classes=3)
tensor(0.8889)
>>> multiclass_specificity(preds, target, num_classes=3, average=None)
tensor([1.0000, 0.6667, 1.0000])
Example (multidim tensors):
>>> from torchmetrics.functional.classification import multiclass_specificity
>>> target = tensor([[[0, 1], [2, 1], [0, 2]], [[1, 1], [2, 0], [1, 2]]])
>>> preds = tensor([[[0, 2], [2, 0], [0, 1]], [[2, 2], [2, 1], [1, 0]]])
>>> multiclass_specificity(preds, target, num_classes=3, multidim_average='samplewise')
tensor([0.7500, 0.6556])
>>> multiclass_specificity(preds, target, num_classes=3, multidim_average='samplewise', average=None)
tensor([[0.7500, 0.7500, 0.7500],
[0.8000, 0.6667, 0.5000]])
"""
if validate_args:
_multiclass_stat_scores_arg_validation(num_classes, top_k, average, multidim_average, ignore_index)
_multiclass_stat_scores_tensor_validation(preds, target, num_classes, multidim_average, ignore_index)
preds, target = _multiclass_stat_scores_format(preds, target, top_k)
tp, fp, tn, fn = _multiclass_stat_scores_update(
preds, target, num_classes, top_k, average, multidim_average, ignore_index
)
return _specificity_reduce(tp, fp, tn, fn, average=average, multidim_average=multidim_average)
def multilabel_specificity(
preds: Tensor,
target: Tensor,
num_labels: int,
threshold: float = 0.5,
average: Optional[Literal["micro", "macro", "weighted", "none"]] = "macro",
multidim_average: Literal["global", "samplewise"] = "global",
ignore_index: Optional[int] = None,
validate_args: bool = True,
) -> Tensor:
r"""Compute `Specificity`_ for multilabel tasks.
.. math:: \text{Specificity} = \frac{\text{TN}}{\text{TN} + \text{FP}}
Where :math:`\text{TN}` and :math:`\text{FP}` represent the number of true negatives and
false positives respecitively.
Accepts the following input tensors:
- ``preds`` (int or float tensor): ``(N, C, ...)``. If preds is a floating point tensor with values outside
[0,1] range we consider the input to be logits and will auto apply sigmoid per element. Addtionally,
we convert to int tensor with thresholding using the value in ``threshold``.
- ``target`` (int tensor): ``(N, C, ...)``
Args:
preds: Tensor with predictions
target: Tensor with true labels
num_labels: Integer specifing the number of labels
threshold: Threshold for transforming probability to binary (0,1) predictions
average:
Defines the reduction that is applied over labels. Should be one of the following:
- ``micro``: Sum statistics over all labels
- ``macro``: Calculate statistics for each label and average them
- ``weighted``: calculates statistics for each label and computes weighted average using their support
- ``"none"`` or ``None``: calculates statistic for each label and applies no reduction
multidim_average:
Defines how additionally dimensions ``...`` should be handled. Should be one of the following:
- ``global``: Additional dimensions are flatted along the batch dimension
- ``samplewise``: Statistic will be calculated independently for each sample on the ``N`` axis.
The statistics in this case are calculated over the additional dimensions.
ignore_index:
Specifies a target value that is ignored and does not contribute to the metric calculation
validate_args: bool indicating if input arguments and tensors should be validated for correctness.
Set to ``False`` for faster computations.
Returns:
The returned shape depends on the ``average`` and ``multidim_average`` arguments:
- If ``multidim_average`` is set to ``global``:
- If ``average='micro'/'macro'/'weighted'``, the output will be a scalar tensor
- If ``average=None/'none'``, the shape will be ``(C,)``
- If ``multidim_average`` is set to ``samplewise``:
- If ``average='micro'/'macro'/'weighted'``, the shape will be ``(N,)``
- If ``average=None/'none'``, the shape will be ``(N, C)``
Example (preds is int tensor):
>>> from torch import tensor
>>> from torchmetrics.functional.classification import multilabel_specificity
>>> target = tensor([[0, 1, 0], [1, 0, 1]])
>>> preds = tensor([[0, 0, 1], [1, 0, 1]])
>>> multilabel_specificity(preds, target, num_labels=3)
tensor(0.6667)
>>> multilabel_specificity(preds, target, num_labels=3, average=None)
tensor([1., 1., 0.])
Example (preds is float tensor):
>>> from torchmetrics.functional.classification import multilabel_specificity
>>> target = tensor([[0, 1, 0], [1, 0, 1]])
>>> preds = tensor([[0.11, 0.22, 0.84], [0.73, 0.33, 0.92]])
>>> multilabel_specificity(preds, target, num_labels=3)
tensor(0.6667)
>>> multilabel_specificity(preds, target, num_labels=3, average=None)
tensor([1., 1., 0.])
Example (multidim tensors):
>>> from torchmetrics.functional.classification import multilabel_specificity
>>> target = tensor([[[0, 1], [1, 0], [0, 1]], [[1, 1], [0, 0], [1, 0]]])
>>> preds = tensor([[[0.59, 0.91], [0.91, 0.99], [0.63, 0.04]],
... [[0.38, 0.04], [0.86, 0.780], [0.45, 0.37]]])
>>> multilabel_specificity(preds, target, num_labels=3, multidim_average='samplewise')
tensor([0.0000, 0.3333])
>>> multilabel_specificity(preds, target, num_labels=3, multidim_average='samplewise', average=None)
tensor([[0., 0., 0.],
[0., 0., 1.]])
"""
if validate_args:
_multilabel_stat_scores_arg_validation(num_labels, threshold, average, multidim_average, ignore_index)
_multilabel_stat_scores_tensor_validation(preds, target, num_labels, multidim_average, ignore_index)
preds, target = _multilabel_stat_scores_format(preds, target, num_labels, threshold, ignore_index)
tp, fp, tn, fn = _multilabel_stat_scores_update(preds, target, multidim_average)
return _specificity_reduce(tp, fp, tn, fn, average=average, multidim_average=multidim_average, multilabel=True)
def specificity(
preds: Tensor,
target: Tensor,
task: Literal["binary", "multiclass", "multilabel"],
threshold: float = 0.5,
num_classes: Optional[int] = None,
num_labels: Optional[int] = None,
average: Optional[Literal["micro", "macro", "weighted", "none"]] = "micro",
multidim_average: Optional[Literal["global", "samplewise"]] = "global",
top_k: Optional[int] = 1,
ignore_index: Optional[int] = None,
validate_args: bool = True,
) -> Tensor:
r"""Compute `Specificity`_.
.. math:: \text{Specificity} = \frac{\text{TN}}{\text{TN} + \text{FP}}
Where :math:`\text{TN}` and :math:`\text{FP}` represent the number of true negatives and
false positives respecitively.
This function is a simple wrapper to get the task specific versions of this metric, which is done by setting the
``task`` argument to either ``'binary'``, ``'multiclass'`` or ``multilabel``. See the documentation of
:func:`~torchmetrics.functional.classification.binary_specificity`,
:func:`~torchmetrics.functional.classification.multiclass_specificity` and
:func:`~torchmetrics.functional.classification.multilabel_specificity` for the specific
details of each argument influence and examples.
LegacyExample:
>>> from torch import tensor
>>> preds = tensor([2, 0, 2, 1])
>>> target = tensor([1, 1, 2, 0])
>>> specificity(preds, target, task="multiclass", average='macro', num_classes=3)
tensor(0.6111)
>>> specificity(preds, target, task="multiclass", average='micro', num_classes=3)
tensor(0.6250)
"""
task = ClassificationTask.from_str(task)
assert multidim_average is not None # noqa: S101 # needed for mypy
if task == ClassificationTask.BINARY:
return binary_specificity(preds, target, threshold, multidim_average, ignore_index, validate_args)
if task == ClassificationTask.MULTICLASS:
if not isinstance(num_classes, int):
raise ValueError(f"`num_classes` is expected to be `int` but `{type(num_classes)} was passed.`")
if not isinstance(top_k, int):
raise ValueError(f"`top_k` is expected to be `int` but `{type(top_k)} was passed.`")
return multiclass_specificity(
preds, target, num_classes, average, top_k, multidim_average, ignore_index, validate_args
)
if task == ClassificationTask.MULTILABEL:
if not isinstance(num_labels, int):
raise ValueError(f"`num_labels` is expected to be `int` but `{type(num_labels)} was passed.`")
return multilabel_specificity(
preds, target, num_labels, threshold, average, multidim_average, ignore_index, validate_args
)
raise ValueError(f"Not handled value: {task}")