-
Notifications
You must be signed in to change notification settings - Fork 387
/
cohen_kappa.py
110 lines (90 loc) · 4.02 KB
/
cohen_kappa.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
# Copyright The PyTorch 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
import torch
from torch import Tensor
from torchmetrics.functional.classification.confusion_matrix import _confusion_matrix_compute, _confusion_matrix_update
_cohen_kappa_update = _confusion_matrix_update
def _cohen_kappa_compute(confmat: Tensor, weights: Optional[str] = None) -> Tensor:
"""Computes Cohen's kappa based on the weighting type.
Args:
confmat: Confusion matrix without normalization
weights: Weighting type to calculate the score. Choose from:
- ``None`` or ``'none'``: no weighting
- ``'linear'``: linear weighting
- ``'quadratic'``: quadratic weighting
Example:
>>> target = torch.tensor([1, 1, 0, 0])
>>> preds = torch.tensor([0, 1, 0, 0])
>>> confmat = _cohen_kappa_update(preds, target, num_classes=2)
>>> _cohen_kappa_compute(confmat)
tensor(0.5000)
"""
confmat = _confusion_matrix_compute(confmat)
confmat = confmat.float() if not confmat.is_floating_point() else confmat
n_classes = confmat.shape[0]
sum0 = confmat.sum(dim=0, keepdim=True)
sum1 = confmat.sum(dim=1, keepdim=True)
expected = sum1 @ sum0 / sum0.sum() # outer product
if weights is None:
w_mat = torch.ones_like(confmat).flatten()
w_mat[:: n_classes + 1] = 0
w_mat = w_mat.reshape(n_classes, n_classes)
elif weights in ("linear", "quadratic"):
w_mat = torch.zeros_like(confmat)
w_mat += torch.arange(n_classes, dtype=w_mat.dtype, device=w_mat.device)
if weights == "linear":
w_mat = torch.abs(w_mat - w_mat.T)
else:
w_mat = torch.pow(w_mat - w_mat.T, 2.0)
else:
raise ValueError(
f"Received {weights} for argument ``weights`` but should be either" " None, 'linear' or 'quadratic'"
)
k = torch.sum(w_mat * confmat) / torch.sum(w_mat * expected)
return 1 - k
def cohen_kappa(
preds: Tensor,
target: Tensor,
num_classes: int,
weights: Optional[str] = None,
threshold: float = 0.5,
) -> Tensor:
r"""Calculates `Cohen's kappa score`_ that measures inter-annotator agreement.
It is defined as
.. math::
\kappa = (p_o - p_e) / (1 - p_e)
where :math:`p_o` is the empirical probability of agreement and :math:`p_e` is
the expected agreement when both annotators assign labels randomly. Note that
:math:`p_e` is estimated using a per-annotator empirical prior over the
class labels.
Args:
preds: (float or long tensor), Either a ``(N, ...)`` tensor with labels or
``(N, C, ...)`` where C is the number of classes, tensor with labels/probabilities
target: ``target`` (long tensor), tensor with shape ``(N, ...)`` with ground true labels
num_classes: Number of classes in the dataset.
weights: Weighting type to calculate the score. Choose from:
- ``None`` or ``'none'``: no weighting
- ``'linear'``: linear weighting
- ``'quadratic'``: quadratic weighting
threshold: Threshold value for binary or multi-label probabilities.
Example:
>>> from torchmetrics.functional import cohen_kappa
>>> target = torch.tensor([1, 1, 0, 0])
>>> preds = torch.tensor([0, 1, 0, 0])
>>> cohen_kappa(preds, target, num_classes=2)
tensor(0.5000)
"""
confmat = _cohen_kappa_update(preds, target, num_classes, threshold)
return _cohen_kappa_compute(confmat, weights)