deepscreen/models/metrics/sensitivity.py

# 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

from torch import Tensor
from torchmetrics.utilities.compute import _safe_divide, _adjust_weights_safe_divide
from typing_extensions import Literal

from torchmetrics.classification.base import _ClassificationTaskWrapper
from torchmetrics.classification.stat_scores import BinaryStatScores, MulticlassStatScores, MultilabelStatScores
from torchmetrics.metric import Metric
from torchmetrics.utilities.enums import ClassificationTask
from torchmetrics.utilities.imports import _MATPLOTLIB_AVAILABLE
from torchmetrics.utilities.plot import _AX_TYPE, _PLOT_OUT_TYPE

if not _MATPLOTLIB_AVAILABLE:
    __doctest_skip__ = ["BinarySensitivity.plot", "MulticlassSensitivity.plot", "MultilabelSensitivity.plot"]


class BinarySensitivity(BinaryStatScores):
    r"""Compute `Sensitivity`_ for binary tasks.

    .. math:: \text{Sensitivity} = \frac{\text{TN}}{\text{TN} + \text{FP}}

    Where :math:`\text{TN}` and :math:`\text{FP}` represent the number of true negatives and false positives
    respectively. The metric is only proper defined when :math:`\text{TN} + \text{FP} \neq 0`. If this case is
    encountered a score of 0 is returned.

    As input to ``forward`` and ``update`` the metric accepts the following input:

    - ``preds`` (:class:`~torch.Tensor`): An int or float tensor of shape ``(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`` (:class:`~torch.Tensor`): An int tensor of shape ``(N, ...)``

    As output to ``forward`` and ``compute`` the metric returns the following output:

    - ``bs`` (:class:`~torch.Tensor`): 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.

    Args:
        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.
    """
    plot_lower_bound: float = 0.0
    plot_upper_bound: float = 1.0

    def compute(self) -> Tensor:
        """Compute metric."""
        tp, fp, tn, fn = self._final_state()
        return _sensitivity_reduce(tp, fp, tn, fn, average="binary", multidim_average=self.multidim_average)

    def plot(
            self, val: Optional[Union[Tensor, Sequence[Tensor]]] = 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 object and Axes object

        Raises:
            ModuleNotFoundError:
                If `matplotlib` is not installed
        """
        return self._plot(val, ax)


class MulticlassSensitivity(MulticlassStatScores):
    r"""Compute `Sensitivity`_ for multiclass tasks.

    .. math:: \text{Sensitivity} = \frac{\text{TN}}{\text{TN} + \text{FP}}

    Where :math:`\text{TN}` and :math:`\text{FP}` represent the number of true negatives and false positives
    respectively.  The metric is only proper defined when :math:`\text{TN} + \text{FP} \neq 0`. If this case is
    encountered for any class, the metric for that class will be set to 0 and the overall metric may therefore be
    affected in turn.

    As input to ``forward`` and ``update`` the metric accepts the following input:

    - ``preds`` (:class:`~torch.Tensor`): An int tensor of shape ``(N, ...)`` or float tensor of shape ``(N, C, ..)``.
      If preds is a floating point we apply ``torch.argmax`` along the ``C`` dimension to automatically convert
      probabilities/logits into an int tensor.
    - ``target`` (:class:`~torch.Tensor`): An int tensor of shape ``(N, ...)``

    As output to ``forward`` and ``compute`` the metric returns the following output:

    - ``mcs`` (:class:`~torch.Tensor`): 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)``

    Args:
        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.
    """
    plot_lower_bound: float = 0.0
    plot_upper_bound: float = 1.0
    plot_legend_name: str = "Class"

    def compute(self) -> Tensor:
        """Compute metric."""
        tp, fp, tn, fn = self._final_state()
        return _sensitivity_reduce(tp, fp, tn, fn, average=self.average, multidim_average=self.multidim_average)

    def plot(
            self, val: Optional[Union[Tensor, Sequence[Tensor]]] = 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 object and Axes object

        Raises:
            ModuleNotFoundError:
                If `matplotlib` is not installed
        """
        return self._plot(val, ax)


class MultilabelSensitivity(MultilabelStatScores):
    r"""Compute `Sensitivity`_ for multilabel tasks.

    .. math:: \text{Sensitivity} = \frac{\text{TN}}{\text{TN} + \text{FP}}

    Where :math:`\text{TN}` and :math:`\text{FP}` represent the number of true negatives and false positives
    respectively. The metric is only proper defined when :math:`\text{TN} + \text{FP} \neq 0`. If this case is
    encountered for any label, the metric for that label will be set to 0 and the overall metric may therefore be
    affected in turn.

    As input to ``forward`` and ``update`` the metric accepts the following input:

    - ``preds`` (:class:`~torch.Tensor`): An int or float tensor of shape ``(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`` (:class:`~torch.Tensor`): An int tensor of shape ``(N, C, ...)``


    As output to ``forward`` and ``compute`` the metric returns the following output:

    - ``mls`` (:class:`~torch.Tensor`): 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)``

    Args:
        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.
    """
    plot_lower_bound: float = 0.0
    plot_upper_bound: float = 1.0
    plot_legend_name: str = "Label"

    def compute(self) -> Tensor:
        """Compute metric."""
        tp, fp, tn, fn = self._final_state()
        return _sensitivity_reduce(
            tp, fp, tn, fn, average=self.average, multidim_average=self.multidim_average, multilabel=True
        )

    def plot(
            self, val: Optional[Union[Tensor, Sequence[Tensor]]] = 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 object and Axes object

        Raises:
            ModuleNotFoundError:
                If `matplotlib` is not installed
        """
        return self._plot(val, ax)


class Sensitivity(_ClassificationTaskWrapper):
    r"""Compute `Sensitivity`_.

    .. math:: \text{Sensitivity} = \frac{\text{TN}}{\text{TN} + \text{FP}}

    Where :math:`\text{TN}` and :math:`\text{FP}` represent the number of true negatives and false positives
    respectively. The metric is only proper defined when :math:`\text{TP} + \text{FP} \neq 0`. If this case is
    encountered for any class/label, the metric for that class/label will be set to 0 and the overall metric may
    therefore be affected in turn.

    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
    :class:`~torchmetrics.classification.BinarySensitivity`, :class:`~torchmetrics.classification.MulticlassSensitivity`
    and :class:`~torchmetrics.classification.MultilabelSensitivity` for the specific details of each argument influence
    and examples.

    Legacy Example:
    """

    def __new__(  # type: ignore[misc]
            cls,
            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,
            **kwargs: Any,
    ) -> Metric:
        """Initialize task metric."""
        task = ClassificationTask.from_str(task)
        assert multidim_average is not None  # noqa: S101  # needed for mypy
        kwargs.update(
            {"multidim_average": multidim_average, "ignore_index": ignore_index, "validate_args": validate_args}
        )
        if task == ClassificationTask.BINARY:
            return BinarySensitivity(threshold, **kwargs)
        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 MulticlassSensitivity(num_classes, top_k, average, **kwargs)
        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 MultilabelSensitivity(num_labels, threshold, average, **kwargs)
        raise ValueError(f"Task {task} not supported!")


def _sensitivity_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(tp, tp + fn)
    if average == "micro":
        tp = tp.sum(dim=0 if multidim_average == "global" else 1)
        fn = fn.sum(dim=0 if multidim_average == "global" else 1)
        return _safe_divide(tp, tp + fn)

    sensitivity_score = _safe_divide(tp, tp + fn)
    return _adjust_weights_safe_divide(sensitivity_score, average, multilabel, tp, fp, fn)