-
Notifications
You must be signed in to change notification settings - Fork 400
/
tschuprows.py
154 lines (121 loc) · 6.16 KB
/
tschuprows.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
# 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 Any, Optional, Sequence, Union
import torch
from torch import Tensor
from typing_extensions import Literal
from torchmetrics.functional.nominal.tschuprows import _tschuprows_t_compute, _tschuprows_t_update
from torchmetrics.functional.nominal.utils import _nominal_input_validation
from torchmetrics.metric import Metric
from torchmetrics.utilities.imports import _MATPLOTLIB_AVAILABLE
from torchmetrics.utilities.plot import _AX_TYPE, _PLOT_OUT_TYPE
if not _MATPLOTLIB_AVAILABLE:
__doctest_skip__ = ["TschuprowsT.plot"]
class TschuprowsT(Metric):
r"""Compute `Tschuprow's T`_ statistic measuring the association between two categorical (nominal) data series.
.. math::
T = \sqrt{\frac{\chi^2 / n}{\sqrt{(r - 1) * (k - 1)}}}
where
.. math::
\chi^2 = \sum_{i,j} \ frac{\left(n_{ij} - \frac{n_{i.} n_{.j}}{n}\right)^2}{\frac{n_{i.} n_{.j}}{n}}
where :math:`n_{ij}` denotes the number of times the values :math:`(A_i, B_j)` are observed with :math:`A_i, B_j`
represent frequencies of values in ``preds`` and ``target``, respectively. Tschuprow's T is a symmetric coefficient,
i.e. :math:`T(preds, target) = T(target, preds)`, so order of input arguments does not matter. The output values
lies in [0, 1] with 1 meaning the perfect association.
As input to ``forward`` and ``update`` the metric accepts the following input:
- ``preds`` (:class:`~torch.Tensor`): Either 1D or 2D tensor of categorical (nominal) data from the first data
series with shape ``(batch_size,)`` or ``(batch_size, num_classes)``, respectively.
- ``target`` (:class:`~torch.Tensor`): Either 1D or 2D tensor of categorical (nominal) data from the second data
series with shape ``(batch_size,)`` or ``(batch_size, num_classes)``, respectively.
As output of ``forward`` and ``compute`` the metric returns the following output:
- ``tschuprows_t`` (:class:`~torch.Tensor`): Scalar tensor containing the Tschuprow's T statistic.
Args:
num_classes: Integer specifying the number of classes
bias_correction: Indication of whether to use bias correction.
nan_strategy: Indication of whether to replace or drop ``NaN`` values
nan_replace_value: Value to replace ``NaN``s when ``nan_strategy = 'replace'``
kwargs: Additional keyword arguments, see :ref:`Metric kwargs` for more info.
Raises:
ValueError:
If `nan_strategy` is not one of `'replace'` and `'drop'`
ValueError:
If `nan_strategy` is equal to `'replace'` and `nan_replace_value` is not an `int` or `float`
Example::
>>> from torch import randint
>>> from torchmetrics.nominal import TschuprowsT
>>> preds = randint(0, 4, (100,))
>>> target = (preds + torch.randn(100)).round().clamp(0, 4)
>>> tschuprows_t = TschuprowsT(num_classes=5)
>>> tschuprows_t(preds, target)
tensor(0.4930)
"""
full_state_update: bool = False
is_differentiable: bool = False
higher_is_better: bool = True
plot_lower_bound: float = 0.0
plot_upper_bound: float = 1.0
confmat: Tensor
def __init__(
self,
num_classes: int,
bias_correction: bool = True,
nan_strategy: Literal["replace", "drop"] = "replace",
nan_replace_value: Optional[float] = 0.0,
**kwargs: Any,
) -> None:
super().__init__(**kwargs)
self.num_classes = num_classes
self.bias_correction = bias_correction
_nominal_input_validation(nan_strategy, nan_replace_value)
self.nan_strategy = nan_strategy
self.nan_replace_value = nan_replace_value
self.add_state("confmat", torch.zeros(num_classes, num_classes), dist_reduce_fx="sum")
def update(self, preds: Tensor, target: Tensor) -> None:
"""Update state with predictions and targets."""
confmat = _tschuprows_t_update(preds, target, self.num_classes, self.nan_strategy, self.nan_replace_value)
self.confmat += confmat
def compute(self) -> Tensor:
"""Compute Tschuprow's T statistic."""
return _tschuprows_t_compute(self.confmat, self.bias_correction)
def plot(self, val: Union[Tensor, Sequence[Tensor], None] = None, ax: Optional[_AX_TYPE] = None) -> _PLOT_OUT_TYPE:
"""Plot a single or multiple values from the metric.
Args:
val: Either a single result from calling `metric.forward` or `metric.compute` or a list of these results.
If no value is provided, will automatically call `metric.compute` and plot that result.
ax: An matplotlib axis object. If provided will add plot to that axis
Returns:
Figure and Axes object
Raises:
ModuleNotFoundError:
If `matplotlib` is not installed
.. plot::
:scale: 75
>>> # Example plotting a single value
>>> import torch
>>> from torchmetrics.nominal import TschuprowsT
>>> metric = TschuprowsT(num_classes=5)
>>> metric.update(torch.randint(0, 4, (100,)), torch.randint(0, 4, (100,)))
>>> fig_, ax_ = metric.plot()
.. plot::
:scale: 75
>>> # Example plotting multiple values
>>> import torch
>>> from torchmetrics.nominal import TschuprowsT
>>> metric = TschuprowsT(num_classes=5)
>>> values = [ ]
>>> for _ in range(10):
... values.append(metric(torch.randint(0, 4, (100,)), torch.randint(0, 4, (100,))))
>>> fig_, ax_ = metric.plot(values)
"""
return self._plot(val, ax)