Spaces:
Running
on
CPU Upgrade
Running
on
CPU Upgrade
Delete deepscreen/models/metrics
Browse files- deepscreen/models/metrics/__init__.py +0 -0
- deepscreen/models/metrics/__pycache__/__init__.cpython-311.pyc +0 -0
- deepscreen/models/metrics/__pycache__/__init__.cpython-39.pyc +0 -0
- deepscreen/models/metrics/__pycache__/bedroc.cpython-311.pyc +0 -0
- deepscreen/models/metrics/__pycache__/bedroc.cpython-39.pyc +0 -0
- deepscreen/models/metrics/__pycache__/ci.cpython-311.pyc +0 -0
- deepscreen/models/metrics/__pycache__/ef.cpython-311.pyc +0 -0
- deepscreen/models/metrics/__pycache__/hit_rate.cpython-311.pyc +0 -0
- deepscreen/models/metrics/__pycache__/hit_rate.cpython-39.pyc +0 -0
- deepscreen/models/metrics/__pycache__/rie.cpython-311.pyc +0 -0
- deepscreen/models/metrics/__pycache__/rie.cpython-39.pyc +0 -0
- deepscreen/models/metrics/__pycache__/sensitivity.cpython-311.pyc +0 -0
- deepscreen/models/metrics/bedroc.py +0 -45
- deepscreen/models/metrics/ci.py +0 -39
- deepscreen/models/metrics/ef.py +0 -34
- deepscreen/models/metrics/hit_rate.py +0 -36
- deepscreen/models/metrics/rie.py +0 -44
- deepscreen/models/metrics/sensitivity.py +0 -337
deepscreen/models/metrics/__init__.py
DELETED
File without changes
|
deepscreen/models/metrics/__pycache__/__init__.cpython-311.pyc
DELETED
Binary file (179 Bytes)
|
|
deepscreen/models/metrics/__pycache__/__init__.cpython-39.pyc
DELETED
Binary file (171 Bytes)
|
|
deepscreen/models/metrics/__pycache__/bedroc.cpython-311.pyc
DELETED
Binary file (2.87 kB)
|
|
deepscreen/models/metrics/__pycache__/bedroc.cpython-39.pyc
DELETED
Binary file (1.58 kB)
|
|
deepscreen/models/metrics/__pycache__/ci.cpython-311.pyc
DELETED
Binary file (3.22 kB)
|
|
deepscreen/models/metrics/__pycache__/ef.cpython-311.pyc
DELETED
Binary file (2.38 kB)
|
|
deepscreen/models/metrics/__pycache__/hit_rate.cpython-311.pyc
DELETED
Binary file (2.37 kB)
|
|
deepscreen/models/metrics/__pycache__/hit_rate.cpython-39.pyc
DELETED
Binary file (1.42 kB)
|
|
deepscreen/models/metrics/__pycache__/rie.cpython-311.pyc
DELETED
Binary file (2.82 kB)
|
|
deepscreen/models/metrics/__pycache__/rie.cpython-39.pyc
DELETED
Binary file (1.58 kB)
|
|
deepscreen/models/metrics/__pycache__/sensitivity.cpython-311.pyc
DELETED
Binary file (17.6 kB)
|
|
deepscreen/models/metrics/bedroc.py
DELETED
@@ -1,45 +0,0 @@
|
|
1 |
-
import torch
|
2 |
-
from torch import Tensor
|
3 |
-
from torchmetrics.retrieval.base import RetrievalMetric
|
4 |
-
from torchmetrics.utilities.checks import _check_retrieval_functional_inputs
|
5 |
-
|
6 |
-
from deepscreen.models.metrics.rie import calc_rie
|
7 |
-
|
8 |
-
|
9 |
-
class BEDROC(RetrievalMetric):
|
10 |
-
is_differentiable: bool = False
|
11 |
-
higher_is_better: bool = True
|
12 |
-
full_state_update: bool = False
|
13 |
-
|
14 |
-
def __init__(
|
15 |
-
self,
|
16 |
-
alpha: float = 80.5,
|
17 |
-
):
|
18 |
-
super().__init__()
|
19 |
-
self.alpha = alpha
|
20 |
-
|
21 |
-
def _metric(self, preds: Tensor, target: Tensor) -> Tensor:
|
22 |
-
preds, target = _check_retrieval_functional_inputs(preds, target)
|
23 |
-
|
24 |
-
n_total = target.size(0)
|
25 |
-
n_actives = target.sum()
|
26 |
-
|
27 |
-
if n_actives == 0:
|
28 |
-
return torch.tensor(0.0, device=preds.device)
|
29 |
-
elif n_actives == n_total:
|
30 |
-
return torch.tensor(1.0, device=preds.device)
|
31 |
-
|
32 |
-
r_a = n_actives / n_total
|
33 |
-
exp_a = torch.exp(torch.tensor(self.alpha))
|
34 |
-
|
35 |
-
idx = torch.argsort(preds, descending=True, stable=True)
|
36 |
-
active_ranks = torch.take(target, idx).nonzero() + 1
|
37 |
-
|
38 |
-
rie = calc_rie(n_total, active_ranks, r_a, exp_a)
|
39 |
-
rie_min = (1 - exp_a ** r_a) / (r_a * (1 - exp_a))
|
40 |
-
rie_max = (1 - exp_a ** (-r_a)) / (r_a * (1 - exp_a ** (-1)))
|
41 |
-
|
42 |
-
return (rie - rie_min) / (rie_max - rie_min)
|
43 |
-
|
44 |
-
def plot(self, val=None, ax=None):
|
45 |
-
return self._plot(val, ax)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
deepscreen/models/metrics/ci.py
DELETED
@@ -1,39 +0,0 @@
|
|
1 |
-
import torch
|
2 |
-
from torchmetrics import Metric
|
3 |
-
from torchmetrics.utilities.checks import _check_same_shape
|
4 |
-
from torchmetrics.utilities.imports import _MATPLOTLIB_AVAILABLE
|
5 |
-
|
6 |
-
if not _MATPLOTLIB_AVAILABLE:
|
7 |
-
__doctest_skip__ = ["ConcordanceIndex.plot"]
|
8 |
-
|
9 |
-
|
10 |
-
class ConcordanceIndex(Metric):
|
11 |
-
is_differentiable: bool = False
|
12 |
-
higher_is_better: bool = True
|
13 |
-
full_state_update: bool = False
|
14 |
-
plot_lower_bound: float = 0.5
|
15 |
-
plot_upper_bound: float = 1.0
|
16 |
-
|
17 |
-
def __init__(self, dist_sync_on_step=False):
|
18 |
-
super().__init__(dist_sync_on_step=dist_sync_on_step)
|
19 |
-
|
20 |
-
self.add_state("num_concordant", default=torch.tensor(0), dist_reduce_fx="sum")
|
21 |
-
self.add_state("num_valid", default=torch.tensor(0), dist_reduce_fx="sum")
|
22 |
-
|
23 |
-
def update(self, preds: torch.Tensor, target: torch.Tensor):
|
24 |
-
_check_same_shape(preds, target)
|
25 |
-
|
26 |
-
g = preds.unsqueeze(-1) - preds
|
27 |
-
g = (g == 0) * 0.5 + (g > 0)
|
28 |
-
|
29 |
-
f = (target.unsqueeze(-1) - target) > 0
|
30 |
-
f = torch.tril(f, diagonal=0)
|
31 |
-
|
32 |
-
self.num_concordant += torch.sum(torch.mul(g, f)).long()
|
33 |
-
self.num_valid += torch.sum(f).long()
|
34 |
-
|
35 |
-
def compute(self):
|
36 |
-
return torch.where(self.num_valid == 0, 0.0, self.num_concordant / self.num_valid)
|
37 |
-
|
38 |
-
def plot(self, val=None, ax=None):
|
39 |
-
return self._plot(val, ax)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
deepscreen/models/metrics/ef.py
DELETED
@@ -1,34 +0,0 @@
|
|
1 |
-
import math
|
2 |
-
|
3 |
-
from torch import Tensor, topk
|
4 |
-
from torchmetrics.retrieval.base import RetrievalMetric
|
5 |
-
from torchmetrics.utilities.checks import _check_retrieval_functional_inputs
|
6 |
-
|
7 |
-
|
8 |
-
class EnrichmentFactor(RetrievalMetric):
|
9 |
-
is_differentiable: bool = False
|
10 |
-
higher_is_better: bool = True
|
11 |
-
full_state_update: bool = False
|
12 |
-
|
13 |
-
def __init__(
|
14 |
-
self,
|
15 |
-
alpha: float,
|
16 |
-
):
|
17 |
-
super().__init__()
|
18 |
-
if alpha <= 0 or alpha > 1:
|
19 |
-
raise ValueError(f"Argument ``alpha`` has to be in interval (0, 1] but got {alpha}")
|
20 |
-
self.alpha = alpha
|
21 |
-
|
22 |
-
def _metric(self, preds: Tensor, target: Tensor) -> Tensor:
|
23 |
-
preds, target = _check_retrieval_functional_inputs(preds, target)
|
24 |
-
|
25 |
-
n_total = target.size(0)
|
26 |
-
n_sampled = math.ceil(n_total * self.alpha)
|
27 |
-
_, idx = topk(preds, n_sampled)
|
28 |
-
hits_sampled = target[idx].sum()
|
29 |
-
hits_total = target.sum()
|
30 |
-
|
31 |
-
return hits_sampled / (hits_total * self.alpha)
|
32 |
-
|
33 |
-
def plot(self, val=None, ax=None):
|
34 |
-
return self._plot(val, ax)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
deepscreen/models/metrics/hit_rate.py
DELETED
@@ -1,36 +0,0 @@
|
|
1 |
-
import math
|
2 |
-
|
3 |
-
from torch import Tensor, topk
|
4 |
-
from torchmetrics.retrieval.base import RetrievalMetric
|
5 |
-
from torchmetrics.utilities.checks import _check_retrieval_functional_inputs
|
6 |
-
|
7 |
-
|
8 |
-
class HitRate(RetrievalMetric):
|
9 |
-
"""
|
10 |
-
Computes hit rate for virtual screening.
|
11 |
-
"""
|
12 |
-
is_differentiable: bool = False
|
13 |
-
higher_is_better: bool = True
|
14 |
-
full_state_update: bool = False
|
15 |
-
|
16 |
-
def __init__(
|
17 |
-
self,
|
18 |
-
alpha: float = 0.01,
|
19 |
-
):
|
20 |
-
super().__init__()
|
21 |
-
if alpha <= 0 or alpha > 1:
|
22 |
-
raise ValueError(f"Argument ``alpha`` has to be in interval (0, 1] but got {alpha}")
|
23 |
-
self.alpha = alpha
|
24 |
-
|
25 |
-
def _metric(self, preds: Tensor, target: Tensor) -> Tensor:
|
26 |
-
preds, target = _check_retrieval_functional_inputs(preds, target)
|
27 |
-
|
28 |
-
n_total = target.size(0)
|
29 |
-
n_sampled = math.ceil(n_total * self.alpha)
|
30 |
-
_, idx = topk(preds, n_sampled)
|
31 |
-
hits_sampled = target[idx].sum()
|
32 |
-
|
33 |
-
return hits_sampled / n_sampled
|
34 |
-
|
35 |
-
def plot(self, val=None, ax=None):
|
36 |
-
return self._plot(val, ax)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
deepscreen/models/metrics/rie.py
DELETED
@@ -1,44 +0,0 @@
|
|
1 |
-
import torch
|
2 |
-
from torch import Tensor
|
3 |
-
from torchmetrics.retrieval.base import RetrievalMetric
|
4 |
-
from torchmetrics.utilities.checks import _check_retrieval_functional_inputs
|
5 |
-
|
6 |
-
|
7 |
-
def calc_rie(n_total, active_ranks, r_a, exp_a):
|
8 |
-
numerator = (exp_a ** (- active_ranks / n_total)).sum()
|
9 |
-
denominator = (1 - exp_a ** (-1)) / (exp_a ** (1 / n_total) - 1)
|
10 |
-
|
11 |
-
return numerator / (r_a * denominator)
|
12 |
-
|
13 |
-
|
14 |
-
class RIE(RetrievalMetric):
|
15 |
-
is_differentiable: bool = False
|
16 |
-
higher_is_better: bool = True
|
17 |
-
full_state_update: bool = False
|
18 |
-
|
19 |
-
def __init__(
|
20 |
-
self,
|
21 |
-
alpha: float = 80.5,
|
22 |
-
):
|
23 |
-
super().__init__()
|
24 |
-
self.alpha = alpha
|
25 |
-
|
26 |
-
def _metric(self, preds: Tensor, target: Tensor) -> Tensor:
|
27 |
-
preds, target = _check_retrieval_functional_inputs(preds, target)
|
28 |
-
|
29 |
-
n_total = target.size(0)
|
30 |
-
n_actives = target.sum()
|
31 |
-
|
32 |
-
if n_actives == 0:
|
33 |
-
return torch.tensor(0.0, device=preds.device)
|
34 |
-
|
35 |
-
r_a = n_actives / n_total
|
36 |
-
exp_a = torch.exp(torch.tensor(-self.alpha))
|
37 |
-
|
38 |
-
idx = torch.argsort(preds, descending=True, stable=True)
|
39 |
-
active_ranks = torch.take(target, idx).nonzero() + 1
|
40 |
-
|
41 |
-
return calc_rie(n_total, active_ranks, r_a, exp_a)
|
42 |
-
|
43 |
-
def plot(self, val=None, ax=None):
|
44 |
-
return self._plot(val, ax)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
deepscreen/models/metrics/sensitivity.py
DELETED
@@ -1,337 +0,0 @@
|
|
1 |
-
# Copyright The Lightning team.
|
2 |
-
#
|
3 |
-
# Licensed under the Apache License, Version 2.0 (the "License");
|
4 |
-
# you may not use this file except in compliance with the License.
|
5 |
-
# You may obtain a copy of the License at
|
6 |
-
#
|
7 |
-
# http://www.apache.org/licenses/LICENSE-2.0
|
8 |
-
#
|
9 |
-
# Unless required by applicable law or agreed to in writing, software
|
10 |
-
# distributed under the License is distributed on an "AS IS" BASIS,
|
11 |
-
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
12 |
-
# See the License for the specific language governing permissions and
|
13 |
-
# limitations under the License.
|
14 |
-
from typing import Any, Optional, Sequence, Union
|
15 |
-
|
16 |
-
from torch import Tensor
|
17 |
-
from torchmetrics.utilities.compute import _safe_divide, _adjust_weights_safe_divide
|
18 |
-
from typing_extensions import Literal
|
19 |
-
|
20 |
-
from torchmetrics.classification.base import _ClassificationTaskWrapper
|
21 |
-
from torchmetrics.classification.stat_scores import BinaryStatScores, MulticlassStatScores, MultilabelStatScores
|
22 |
-
from torchmetrics.metric import Metric
|
23 |
-
from torchmetrics.utilities.enums import ClassificationTask
|
24 |
-
from torchmetrics.utilities.imports import _MATPLOTLIB_AVAILABLE
|
25 |
-
from torchmetrics.utilities.plot import _AX_TYPE, _PLOT_OUT_TYPE
|
26 |
-
|
27 |
-
if not _MATPLOTLIB_AVAILABLE:
|
28 |
-
__doctest_skip__ = ["BinarySensitivity.plot", "MulticlassSensitivity.plot", "MultilabelSensitivity.plot"]
|
29 |
-
|
30 |
-
|
31 |
-
class BinarySensitivity(BinaryStatScores):
|
32 |
-
r"""Compute `Sensitivity`_ for binary tasks.
|
33 |
-
|
34 |
-
.. math:: \text{Sensitivity} = \frac{\text{TN}}{\text{TN} + \text{FP}}
|
35 |
-
|
36 |
-
Where :math:`\text{TN}` and :math:`\text{FP}` represent the number of true negatives and false positives
|
37 |
-
respectively. The metric is only proper defined when :math:`\text{TN} + \text{FP} \neq 0`. If this case is
|
38 |
-
encountered a score of 0 is returned.
|
39 |
-
|
40 |
-
As input to ``forward`` and ``update`` the metric accepts the following input:
|
41 |
-
|
42 |
-
- ``preds`` (:class:`~torch.Tensor`): An int or float tensor of shape ``(N, ...)``. If preds is a floating point
|
43 |
-
tensor with values outside [0,1] range we consider the input to be logits and will auto apply sigmoid per
|
44 |
-
element. Addtionally, we convert to int tensor with thresholding using the value in ``threshold``.
|
45 |
-
- ``target`` (:class:`~torch.Tensor`): An int tensor of shape ``(N, ...)``
|
46 |
-
|
47 |
-
As output to ``forward`` and ``compute`` the metric returns the following output:
|
48 |
-
|
49 |
-
- ``bs`` (:class:`~torch.Tensor`): If ``multidim_average`` is set to ``global``, the metric returns a scalar value.
|
50 |
-
If ``multidim_average`` is set to ``samplewise``, the metric returns ``(N,)`` vector consisting of a scalar value
|
51 |
-
per sample.
|
52 |
-
|
53 |
-
Args:
|
54 |
-
threshold: Threshold for transforming probability to binary {0,1} predictions
|
55 |
-
multidim_average:
|
56 |
-
Defines how additionally dimensions ``...`` should be handled. Should be one of the following:
|
57 |
-
|
58 |
-
- ``global``: Additional dimensions are flatted along the batch dimension
|
59 |
-
- ``samplewise``: Statistic will be calculated independently for each sample on the ``N`` axis.
|
60 |
-
The statistics in this case are calculated over the additional dimensions.
|
61 |
-
|
62 |
-
ignore_index:
|
63 |
-
Specifies a target value that is ignored and does not contribute to the metric calculation
|
64 |
-
validate_args: bool indicating if input arguments and tensors should be validated for correctness.
|
65 |
-
Set to ``False`` for faster computations.
|
66 |
-
"""
|
67 |
-
plot_lower_bound: float = 0.0
|
68 |
-
plot_upper_bound: float = 1.0
|
69 |
-
|
70 |
-
def compute(self) -> Tensor:
|
71 |
-
"""Compute metric."""
|
72 |
-
tp, fp, tn, fn = self._final_state()
|
73 |
-
return _sensitivity_reduce(tp, fp, tn, fn, average="binary", multidim_average=self.multidim_average)
|
74 |
-
|
75 |
-
def plot(
|
76 |
-
self, val: Optional[Union[Tensor, Sequence[Tensor]]] = None, ax: Optional[_AX_TYPE] = None
|
77 |
-
) -> _PLOT_OUT_TYPE:
|
78 |
-
"""Plot a single or multiple values from the metric.
|
79 |
-
|
80 |
-
Args:
|
81 |
-
val: Either a single result from calling `metric.forward` or `metric.compute` or a list of these results.
|
82 |
-
If no value is provided, will automatically call `metric.compute` and plot that result.
|
83 |
-
ax: An matplotlib axis object. If provided will add plot to that axis
|
84 |
-
|
85 |
-
Returns:
|
86 |
-
Figure object and Axes object
|
87 |
-
|
88 |
-
Raises:
|
89 |
-
ModuleNotFoundError:
|
90 |
-
If `matplotlib` is not installed
|
91 |
-
"""
|
92 |
-
return self._plot(val, ax)
|
93 |
-
|
94 |
-
|
95 |
-
class MulticlassSensitivity(MulticlassStatScores):
|
96 |
-
r"""Compute `Sensitivity`_ for multiclass tasks.
|
97 |
-
|
98 |
-
.. math:: \text{Sensitivity} = \frac{\text{TN}}{\text{TN} + \text{FP}}
|
99 |
-
|
100 |
-
Where :math:`\text{TN}` and :math:`\text{FP}` represent the number of true negatives and false positives
|
101 |
-
respectively. The metric is only proper defined when :math:`\text{TN} + \text{FP} \neq 0`. If this case is
|
102 |
-
encountered for any class, the metric for that class will be set to 0 and the overall metric may therefore be
|
103 |
-
affected in turn.
|
104 |
-
|
105 |
-
As input to ``forward`` and ``update`` the metric accepts the following input:
|
106 |
-
|
107 |
-
- ``preds`` (:class:`~torch.Tensor`): An int tensor of shape ``(N, ...)`` or float tensor of shape ``(N, C, ..)``.
|
108 |
-
If preds is a floating point we apply ``torch.argmax`` along the ``C`` dimension to automatically convert
|
109 |
-
probabilities/logits into an int tensor.
|
110 |
-
- ``target`` (:class:`~torch.Tensor`): An int tensor of shape ``(N, ...)``
|
111 |
-
|
112 |
-
As output to ``forward`` and ``compute`` the metric returns the following output:
|
113 |
-
|
114 |
-
- ``mcs`` (:class:`~torch.Tensor`): The returned shape depends on the ``average`` and ``multidim_average``
|
115 |
-
arguments:
|
116 |
-
|
117 |
-
- If ``multidim_average`` is set to ``global``:
|
118 |
-
|
119 |
-
- If ``average='micro'/'macro'/'weighted'``, the output will be a scalar tensor
|
120 |
-
- If ``average=None/'none'``, the shape will be ``(C,)``
|
121 |
-
|
122 |
-
- If ``multidim_average`` is set to ``samplewise``:
|
123 |
-
|
124 |
-
- If ``average='micro'/'macro'/'weighted'``, the shape will be ``(N,)``
|
125 |
-
- If ``average=None/'none'``, the shape will be ``(N, C)``
|
126 |
-
|
127 |
-
Args:
|
128 |
-
num_classes: Integer specifing the number of classes
|
129 |
-
average:
|
130 |
-
Defines the reduction that is applied over labels. Should be one of the following:
|
131 |
-
|
132 |
-
- ``micro``: Sum statistics over all labels
|
133 |
-
- ``macro``: Calculate statistics for each label and average them
|
134 |
-
- ``weighted``: calculates statistics for each label and computes weighted average using their support
|
135 |
-
- ``"none"`` or ``None``: calculates statistic for each label and applies no reduction
|
136 |
-
|
137 |
-
top_k:
|
138 |
-
Number of highest probability or logit score predictions considered to find the correct label.
|
139 |
-
Only works when ``preds`` contain probabilities/logits.
|
140 |
-
multidim_average:
|
141 |
-
Defines how additionally dimensions ``...`` should be handled. Should be one of the following:
|
142 |
-
|
143 |
-
- ``global``: Additional dimensions are flatted along the batch dimension
|
144 |
-
- ``samplewise``: Statistic will be calculated independently for each sample on the ``N`` axis.
|
145 |
-
The statistics in this case are calculated over the additional dimensions.
|
146 |
-
|
147 |
-
ignore_index:
|
148 |
-
Specifies a target value that is ignored and does not contribute to the metric calculation
|
149 |
-
validate_args: bool indicating if input arguments and tensors should be validated for correctness.
|
150 |
-
Set to ``False`` for faster computations.
|
151 |
-
"""
|
152 |
-
plot_lower_bound: float = 0.0
|
153 |
-
plot_upper_bound: float = 1.0
|
154 |
-
plot_legend_name: str = "Class"
|
155 |
-
|
156 |
-
def compute(self) -> Tensor:
|
157 |
-
"""Compute metric."""
|
158 |
-
tp, fp, tn, fn = self._final_state()
|
159 |
-
return _sensitivity_reduce(tp, fp, tn, fn, average=self.average, multidim_average=self.multidim_average)
|
160 |
-
|
161 |
-
def plot(
|
162 |
-
self, val: Optional[Union[Tensor, Sequence[Tensor]]] = None, ax: Optional[_AX_TYPE] = None
|
163 |
-
) -> _PLOT_OUT_TYPE:
|
164 |
-
"""Plot a single or multiple values from the metric.
|
165 |
-
|
166 |
-
Args:
|
167 |
-
val: Either a single result from calling `metric.forward` or `metric.compute` or a list of these results.
|
168 |
-
If no value is provided, will automatically call `metric.compute` and plot that result.
|
169 |
-
ax: An matplotlib axis object. If provided will add plot to that axis
|
170 |
-
|
171 |
-
Returns:
|
172 |
-
Figure object and Axes object
|
173 |
-
|
174 |
-
Raises:
|
175 |
-
ModuleNotFoundError:
|
176 |
-
If `matplotlib` is not installed
|
177 |
-
"""
|
178 |
-
return self._plot(val, ax)
|
179 |
-
|
180 |
-
|
181 |
-
class MultilabelSensitivity(MultilabelStatScores):
|
182 |
-
r"""Compute `Sensitivity`_ for multilabel tasks.
|
183 |
-
|
184 |
-
.. math:: \text{Sensitivity} = \frac{\text{TN}}{\text{TN} + \text{FP}}
|
185 |
-
|
186 |
-
Where :math:`\text{TN}` and :math:`\text{FP}` represent the number of true negatives and false positives
|
187 |
-
respectively. The metric is only proper defined when :math:`\text{TN} + \text{FP} \neq 0`. If this case is
|
188 |
-
encountered for any label, the metric for that label will be set to 0 and the overall metric may therefore be
|
189 |
-
affected in turn.
|
190 |
-
|
191 |
-
As input to ``forward`` and ``update`` the metric accepts the following input:
|
192 |
-
|
193 |
-
- ``preds`` (:class:`~torch.Tensor`): An int or float tensor of shape ``(N, C, ...)``. If preds is a floating
|
194 |
-
point tensor with values outside [0,1] range we consider the input to be logits and will auto apply sigmoid
|
195 |
-
per element. Addtionally, we convert to int tensor with thresholding using the value in ``threshold``.
|
196 |
-
- ``target`` (:class:`~torch.Tensor`): An int tensor of shape ``(N, C, ...)``
|
197 |
-
|
198 |
-
|
199 |
-
As output to ``forward`` and ``compute`` the metric returns the following output:
|
200 |
-
|
201 |
-
- ``mls`` (:class:`~torch.Tensor`): The returned shape depends on the ``average`` and ``multidim_average``
|
202 |
-
arguments:
|
203 |
-
|
204 |
-
- If ``multidim_average`` is set to ``global``
|
205 |
-
|
206 |
-
- If ``average='micro'/'macro'/'weighted'``, the output will be a scalar tensor
|
207 |
-
- If ``average=None/'none'``, the shape will be ``(C,)``
|
208 |
-
|
209 |
-
- If ``multidim_average`` is set to ``samplewise``
|
210 |
-
|
211 |
-
- If ``average='micro'/'macro'/'weighted'``, the shape will be ``(N,)``
|
212 |
-
- If ``average=None/'none'``, the shape will be ``(N, C)``
|
213 |
-
|
214 |
-
Args:
|
215 |
-
num_labels: Integer specifing the number of labels
|
216 |
-
threshold: Threshold for transforming probability to binary (0,1) predictions
|
217 |
-
average:
|
218 |
-
Defines the reduction that is applied over labels. Should be one of the following:
|
219 |
-
|
220 |
-
- ``micro``: Sum statistics over all labels
|
221 |
-
- ``macro``: Calculate statistics for each label and average them
|
222 |
-
- ``weighted``: calculates statistics for each label and computes weighted average using their support
|
223 |
-
- ``"none"`` or ``None``: calculates statistic for each label and applies no reduction
|
224 |
-
|
225 |
-
multidim_average: Defines how additionally dimensions ``...`` should be handled. Should be one of the following:
|
226 |
-
|
227 |
-
- ``global``: Additional dimensions are flatted along the batch dimension
|
228 |
-
- ``samplewise``: Statistic will be calculated independently for each sample on the ``N`` axis.
|
229 |
-
The statistics in this case are calculated over the additional dimensions.
|
230 |
-
|
231 |
-
ignore_index:
|
232 |
-
Specifies a target value that is ignored and does not contribute to the metric calculation
|
233 |
-
validate_args: bool indicating if input arguments and tensors should be validated for correctness.
|
234 |
-
Set to ``False`` for faster computations.
|
235 |
-
"""
|
236 |
-
plot_lower_bound: float = 0.0
|
237 |
-
plot_upper_bound: float = 1.0
|
238 |
-
plot_legend_name: str = "Label"
|
239 |
-
|
240 |
-
def compute(self) -> Tensor:
|
241 |
-
"""Compute metric."""
|
242 |
-
tp, fp, tn, fn = self._final_state()
|
243 |
-
return _sensitivity_reduce(
|
244 |
-
tp, fp, tn, fn, average=self.average, multidim_average=self.multidim_average, multilabel=True
|
245 |
-
)
|
246 |
-
|
247 |
-
def plot(
|
248 |
-
self, val: Optional[Union[Tensor, Sequence[Tensor]]] = None, ax: Optional[_AX_TYPE] = None
|
249 |
-
) -> _PLOT_OUT_TYPE:
|
250 |
-
"""Plot a single or multiple values from the metric.
|
251 |
-
|
252 |
-
Args:
|
253 |
-
val: Either a single result from calling `metric.forward` or `metric.compute` or a list of these results.
|
254 |
-
If no value is provided, will automatically call `metric.compute` and plot that result.
|
255 |
-
ax: An matplotlib axis object. If provided will add plot to that axis
|
256 |
-
|
257 |
-
Returns:
|
258 |
-
Figure object and Axes object
|
259 |
-
|
260 |
-
Raises:
|
261 |
-
ModuleNotFoundError:
|
262 |
-
If `matplotlib` is not installed
|
263 |
-
"""
|
264 |
-
return self._plot(val, ax)
|
265 |
-
|
266 |
-
|
267 |
-
class Sensitivity(_ClassificationTaskWrapper):
|
268 |
-
r"""Compute `Sensitivity`_.
|
269 |
-
|
270 |
-
.. math:: \text{Sensitivity} = \frac{\text{TN}}{\text{TN} + \text{FP}}
|
271 |
-
|
272 |
-
Where :math:`\text{TN}` and :math:`\text{FP}` represent the number of true negatives and false positives
|
273 |
-
respectively. The metric is only proper defined when :math:`\text{TP} + \text{FP} \neq 0`. If this case is
|
274 |
-
encountered for any class/label, the metric for that class/label will be set to 0 and the overall metric may
|
275 |
-
therefore be affected in turn.
|
276 |
-
|
277 |
-
This function is a simple wrapper to get the task specific versions of this metric, which is done by setting the
|
278 |
-
``task`` argument to either ``'binary'``, ``'multiclass'`` or ``multilabel``. See the documentation of
|
279 |
-
:class:`~torchmetrics.classification.BinarySensitivity`, :class:`~torchmetrics.classification.MulticlassSensitivity`
|
280 |
-
and :class:`~torchmetrics.classification.MultilabelSensitivity` for the specific details of each argument influence
|
281 |
-
and examples.
|
282 |
-
|
283 |
-
Legacy Example:
|
284 |
-
"""
|
285 |
-
|
286 |
-
def __new__( # type: ignore[misc]
|
287 |
-
cls,
|
288 |
-
task: Literal["binary", "multiclass", "multilabel"],
|
289 |
-
threshold: float = 0.5,
|
290 |
-
num_classes: Optional[int] = None,
|
291 |
-
num_labels: Optional[int] = None,
|
292 |
-
average: Optional[Literal["micro", "macro", "weighted", "none"]] = "micro",
|
293 |
-
multidim_average: Optional[Literal["global", "samplewise"]] = "global",
|
294 |
-
top_k: Optional[int] = 1,
|
295 |
-
ignore_index: Optional[int] = None,
|
296 |
-
validate_args: bool = True,
|
297 |
-
**kwargs: Any,
|
298 |
-
) -> Metric:
|
299 |
-
"""Initialize task metric."""
|
300 |
-
task = ClassificationTask.from_str(task)
|
301 |
-
assert multidim_average is not None # noqa: S101 # needed for mypy
|
302 |
-
kwargs.update(
|
303 |
-
{"multidim_average": multidim_average, "ignore_index": ignore_index, "validate_args": validate_args}
|
304 |
-
)
|
305 |
-
if task == ClassificationTask.BINARY:
|
306 |
-
return BinarySensitivity(threshold, **kwargs)
|
307 |
-
if task == ClassificationTask.MULTICLASS:
|
308 |
-
if not isinstance(num_classes, int):
|
309 |
-
raise ValueError(f"`num_classes` is expected to be `int` but `{type(num_classes)} was passed.`")
|
310 |
-
if not isinstance(top_k, int):
|
311 |
-
raise ValueError(f"`top_k` is expected to be `int` but `{type(top_k)} was passed.`")
|
312 |
-
return MulticlassSensitivity(num_classes, top_k, average, **kwargs)
|
313 |
-
if task == ClassificationTask.MULTILABEL:
|
314 |
-
if not isinstance(num_labels, int):
|
315 |
-
raise ValueError(f"`num_labels` is expected to be `int` but `{type(num_labels)} was passed.`")
|
316 |
-
return MultilabelSensitivity(num_labels, threshold, average, **kwargs)
|
317 |
-
raise ValueError(f"Task {task} not supported!")
|
318 |
-
|
319 |
-
|
320 |
-
def _sensitivity_reduce(
|
321 |
-
tp: Tensor,
|
322 |
-
fp: Tensor,
|
323 |
-
tn: Tensor,
|
324 |
-
fn: Tensor,
|
325 |
-
average: Optional[Literal["binary", "micro", "macro", "weighted", "none"]],
|
326 |
-
multidim_average: Literal["global", "samplewise"] = "global",
|
327 |
-
multilabel: bool = False,
|
328 |
-
) -> Tensor:
|
329 |
-
if average == "binary":
|
330 |
-
return _safe_divide(tp, tp + fn)
|
331 |
-
if average == "micro":
|
332 |
-
tp = tp.sum(dim=0 if multidim_average == "global" else 1)
|
333 |
-
fn = fn.sum(dim=0 if multidim_average == "global" else 1)
|
334 |
-
return _safe_divide(tp, tp + fn)
|
335 |
-
|
336 |
-
sensitivity_score = _safe_divide(tp, tp + fn)
|
337 |
-
return _adjust_weights_safe_divide(sensitivity_score, average, multilabel, tp, fp, fn)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|