Precision At Fixed Recall

Module Interface

class torchmetrics.PrecisionAtFixedRecall(**kwargs)

Compute the highest possible recall value given the minimum precision thresholds provided.

This is done by first calculating the precision-recall curve for different thresholds and the find the recall for a given precision level.

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 BinaryPrecisionAtFixedRecall, MulticlassPrecisionAtFixedRecall and MultilabelPrecisionAtFixedRecall for the specific details of each argument influence and examples.

static __new__(cls, task, min_recall, thresholds=None, num_classes=None, num_labels=None, ignore_index=None, validate_args=True, **kwargs)

Initialize task metric.

Return type:

Metric

BinaryPrecisionAtFixedRecall

class torchmetrics.classification.BinaryPrecisionAtFixedRecall(min_recall, thresholds=None, ignore_index=None, validate_args=True, **kwargs)

Compute the highest possible precision value given the minimum recall thresholds provided.

This is done by first calculating the precision-recall curve for different thresholds and the find the precision for a given recall level.

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

• preds (Tensor): A float tensor of shape (N, ...). Preds should be a tensor containing probabilities or logits for each observation. If preds has values outside [0,1] range we consider the input to be logits and will auto apply sigmoid per element.

• target (Tensor): An int tensor of shape (N, ...). Target should be a tensor containing ground truth labels, and therefore only contain {0,1} values (except if ignore_index is specified). The value 1 always encodes the positive class.

Tip

Additional dimension ... will be flattened into the batch dimension.

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

• precision (Tensor): A scalar tensor with the maximum precision for the given recall level

• threshold (Tensor): A scalar tensor with the corresponding threshold level

Note

The implementation both supports calculating the metric in a non-binned but accurate version and a binned version that is less accurate but more memory efficient. Setting the thresholds argument to None will activate the non-binned version that uses memory of size \(\mathcal{O}(n_{samples})\) whereas setting the thresholds argument to either an integer, list or a 1d tensor will use a binned version that uses memory of size \(\mathcal{O}(n_{thresholds})\) (constant memory).

Parameters:

• min_recall (float) – float value specifying minimum recall threshold.

• thresholds (Union[int, List[float], Tensor, None]) –

  Can be one of:

  • If set to None, will use a non-binned approach where thresholds are dynamically calculated from all the data. Most accurate but also most memory consuming approach.

  • If set to an int (larger than 1), will use that number of thresholds linearly spaced from 0 to 1 as bins for the calculation.

  • If set to an list of floats, will use the indicated thresholds in the list as bins for the calculation

  • If set to an 1d Tensor of floats, will use the indicated thresholds in the tensor as bins for the calculation.

• validate_args (bool) – bool indicating if input arguments and tensors should be validated for correctness. Set to False for faster computations.

• kwargs (Any) – Additional keyword arguments, see Advanced metric settings for more info.

Example

>>> from torch import tensor
>>> from torchmetrics.classification import BinaryPrecisionAtFixedRecall
>>> preds = tensor([0, 0.5, 0.7, 0.8])
>>> target = tensor([0, 1, 1, 0])
>>> metric = BinaryPrecisionAtFixedRecall(min_recall=0.5, thresholds=None)
>>> metric(preds, target)
(tensor(0.6667), tensor(0.5000))
>>> metric = BinaryPrecisionAtFixedRecall(min_recall=0.5, thresholds=5)
>>> metric(preds, target)
(tensor(0.6667), tensor(0.5000))

plot(val=None, ax=None)

Plot a single or multiple values from the metric.

Parameters:

• val (Union[Tensor, Sequence[Tensor], None]) – 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 (Optional[Axes]) – An matplotlib axis object. If provided will add plot to that axis

Return type:

Tuple[Figure, Union[Axes, ndarray]]

Returns:

Figure object and Axes object

Raises:

ModuleNotFoundError – If matplotlib is not installed

>>> from torch import rand, randint
>>> # Example plotting a single value
>>> from torchmetrics.classification import BinaryPrecisionAtFixedRecall
>>> metric = BinaryPrecisionAtFixedRecall(min_recall=0.5)
>>> metric.update(rand(10), randint(2,(10,)))
>>> fig_, ax_ = metric.plot()  # the returned plot only shows the maximum recall value by default

>>> from torch import rand, randint
>>> # Example plotting multiple values
>>> from torchmetrics.classification import BinaryPrecisionAtFixedRecall
>>> metric = BinaryPrecisionAtFixedRecall(min_recall=0.5)
>>> values = [ ]
>>> for _ in range(10):
...     # we index by 0 such that only the maximum recall value is plotted
...     values.append(metric(rand(10), randint(2,(10,)))[0])
>>> fig_, ax_ = metric.plot(values)

MulticlassPrecisionAtFixedRecall

class torchmetrics.classification.MulticlassPrecisionAtFixedRecall(num_classes, min_recall, thresholds=None, ignore_index=None, validate_args=True, **kwargs)

Compute the highest possible precision value given the minimum recall thresholds provided.

This is done by first calculating the precision-recall curve for different thresholds and the find the precision for a given recall level.

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

• preds (Tensor): A float tensor of shape (N, C, ...). Preds should be a tensor containing probabilities or logits for each observation. If preds has values outside [0,1] range we consider the input to be logits and will auto apply softmax per sample.

• target (Tensor): An int tensor of shape (N, ...). Target should be a tensor containing ground truth labels, and therefore only contain values in the [0, n_classes-1] range (except if ignore_index is specified).

Tip

Additional dimension ... will be flattened into the batch dimension.

As output to forward and compute the metric returns a tuple of either 2 tensors or 2 lists containing:

• precision (Tensor): A 1d tensor of size (n_classes, ) with the maximum precision for the given recall level per class

• threshold (Tensor): A 1d tensor of size (n_classes, ) with the corresponding threshold level per class

Note

The implementation both supports calculating the metric in a non-binned but accurate version and a binned version that is less accurate but more memory efficient. Setting the thresholds argument to None will activate the non-binned version that uses memory of size \(\mathcal{O}(n_{samples})\) whereas setting the thresholds argument to either an integer, list or a 1d tensor will use a binned version that uses memory of size \(\mathcal{O}(n_{thresholds} \times n_{classes})\) (constant memory).

Parameters:

• num_classes (int) – Integer specifying the number of classes

• min_recall (float) – float value specifying minimum recall threshold.

• thresholds (Union[int, List[float], Tensor, None]) –

  Can be one of:

  • If set to None, will use a non-binned approach where thresholds are dynamically calculated from all the data. Most accurate but also most memory consuming approach.

  • If set to an int (larger than 1), will use that number of thresholds linearly spaced from 0 to 1 as bins for the calculation.

  • If set to an list of floats, will use the indicated thresholds in the list as bins for the calculation

  • If set to an 1d Tensor of floats, will use the indicated thresholds in the tensor as bins for the calculation.

• validate_args (bool) – bool indicating if input arguments and tensors should be validated for correctness. Set to False for faster computations.

• kwargs (Any) – Additional keyword arguments, see Advanced metric settings for more info.

Example

>>> from torch import tensor
>>> from torchmetrics.classification import MulticlassPrecisionAtFixedRecall
>>> preds = tensor([[0.75, 0.05, 0.05, 0.05, 0.05],
...                 [0.05, 0.75, 0.05, 0.05, 0.05],
...                 [0.05, 0.05, 0.75, 0.05, 0.05],
...                 [0.05, 0.05, 0.05, 0.75, 0.05]])
>>> target = tensor([0, 1, 3, 2])
>>> metric = MulticlassPrecisionAtFixedRecall(num_classes=5, min_recall=0.5, thresholds=None)
>>> metric(preds, target)  
(tensor([1.0000, 1.0000, 0.2500, 0.2500, 0.0000]),
 tensor([7.5000e-01, 7.5000e-01, 5.0000e-02, 5.0000e-02, 1.0000e+06]))
>>> mcrafp = MulticlassPrecisionAtFixedRecall(num_classes=5, min_recall=0.5, thresholds=5)
>>> mcrafp(preds, target)  
(tensor([1.0000, 1.0000, 0.2500, 0.2500, 0.0000]),
 tensor([7.5000e-01, 7.5000e-01, 0.0000e+00, 0.0000e+00, 1.0000e+06]))

plot(val=None, ax=None)

Plot a single or multiple values from the metric.

Parameters:

• val (Union[Tensor, Sequence[Tensor], None]) – 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 (Optional[Axes]) – An matplotlib axis object. If provided will add plot to that axis

Return type:

Tuple[Figure, Union[Axes, ndarray]]

Returns:

Figure object and Axes object

Raises:

ModuleNotFoundError – If matplotlib is not installed

>>> from torch import rand, randint
>>> # Example plotting a single value per class
>>> from torchmetrics.classification import MulticlassPrecisionAtFixedRecall
>>> metric = MulticlassPrecisionAtFixedRecall(num_classes=3, min_recall=0.5)
>>> metric.update(rand(20, 3).softmax(dim=-1), randint(3, (20,)))
>>> fig_, ax_ = metric.plot()  # the returned plot only shows the maximum recall value by default

>>> from torch import rand, randint
>>> # Example plotting a multiple values per class
>>> from torchmetrics.classification import MulticlassPrecisionAtFixedRecall
>>> metric = MulticlassPrecisionAtFixedRecall(num_classes=3, min_recall=0.5)
>>> values = []
>>> for _ in range(20):
...     # we index by 0 such that only the maximum recall value is plotted
...     values.append(metric(rand(20, 3).softmax(dim=-1), randint(3, (20,)))[0])
>>> fig_, ax_ = metric.plot(values)

MultilabelPrecisionAtFixedRecall

class torchmetrics.classification.MultilabelPrecisionAtFixedRecall(num_labels, min_recall, thresholds=None, ignore_index=None, validate_args=True, **kwargs)

Compute the highest possible precision value given the minimum recall thresholds provided.

This is done by first calculating the precision-recall curve for different thresholds and the find the precision for a given recall level.

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

• preds (Tensor): A float tensor of shape (N, C, ...). Preds should be a tensor containing probabilities or logits for each observation. If preds has values outside [0,1] range we consider the input to be logits and will auto apply sigmoid per element.

• target (Tensor): An int tensor of shape (N, ...). Target should be a tensor containing ground truth labels, and therefore only contain {0,1} values (except if ignore_index is specified). The value 1 always encodes the positive class.

Tip

Additional dimension ... will be flattened into the batch dimension.

As output to forward and compute the metric returns a tuple of either 2 tensors or 2 lists containing:

• precision (Tensor): A 1d tensor of size (n_classes, ) with the maximum precision for the given recall level per class

• threshold (Tensor): A 1d tensor of size (n_classes, ) with the corresponding threshold level per class

Note

The implementation both supports calculating the metric in a non-binned but accurate version and a binned version that is less accurate but more memory efficient. Setting the thresholds argument to None will activate the non-binned version that uses memory of size \(\mathcal{O}(n_{samples})\) whereas setting the thresholds argument to either an integer, list or a 1d tensor will use a binned version that uses memory of size \(\mathcal{O}(n_{thresholds} \times n_{labels})\) (constant memory).

Parameters:

• num_labels (int) – Integer specifying the number of labels

• min_recall (float) – float value specifying minimum recall threshold.

• thresholds (Union[int, List[float], Tensor, None]) –

  Can be one of:

  • If set to None, will use a non-binned approach where thresholds are dynamically calculated from all the data. Most accurate but also most memory consuming approach.

  • If set to an int (larger than 1), will use that number of thresholds linearly spaced from 0 to 1 as bins for the calculation.

  • If set to an list of floats, will use the indicated thresholds in the list as bins for the calculation

  • If set to an 1d Tensor of floats, will use the indicated thresholds in the tensor as bins for the calculation.

• validate_args (bool) – bool indicating if input arguments and tensors should be validated for correctness. Set to False for faster computations.

• kwargs (Any) – Additional keyword arguments, see Advanced metric settings for more info.

Example

>>> from torch import tensor
>>> from torchmetrics.classification import MultilabelPrecisionAtFixedRecall
>>> preds = tensor([[0.75, 0.05, 0.35],
...                 [0.45, 0.75, 0.05],
...                 [0.05, 0.55, 0.75],
...                 [0.05, 0.65, 0.05]])
>>> target = tensor([[1, 0, 1],
...                  [0, 0, 0],
...                  [0, 1, 1],
...                  [1, 1, 1]])
>>> metric = MultilabelPrecisionAtFixedRecall(num_labels=3, min_recall=0.5, thresholds=None)
>>> metric(preds, target)
(tensor([1.0000, 0.6667, 1.0000]), tensor([0.7500, 0.5500, 0.3500]))
>>> mlrafp = MultilabelPrecisionAtFixedRecall(num_labels=3, min_recall=0.5, thresholds=5)
>>> mlrafp(preds, target)
(tensor([1.0000, 0.6667, 1.0000]), tensor([0.7500, 0.5000, 0.2500]))

plot(val=None, ax=None)

Plot a single or multiple values from the metric.

Parameters:

• val (Union[Tensor, Sequence[Tensor], None]) – 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 (Optional[Axes]) – An matplotlib axis object. If provided will add plot to that axis

Return type:

Tuple[Figure, Union[Axes, ndarray]]

Returns:

Figure object and Axes object

Raises:

ModuleNotFoundError – If matplotlib is not installed

>>> from torch import rand, randint
>>> # Example plotting a single value
>>> from torchmetrics.classification import MultilabelPrecisionAtFixedRecall
>>> metric = MultilabelPrecisionAtFixedRecall(num_labels=3, min_recall=0.5)
>>> metric.update(rand(20, 3), randint(2, (20, 3)))
>>> fig_, ax_ = metric.plot()  # the returned plot only shows the maximum recall value by default

>>> from torch import rand, randint
>>> # Example plotting multiple values
>>> from torchmetrics.classification import MultilabelPrecisionAtFixedRecall
>>> metric = MultilabelPrecisionAtFixedRecall(num_labels=3, min_recall=0.5)
>>> values = [ ]
>>> for _ in range(10):
...     # we index by 0 such that only the maximum recall value is plotted
...     values.append(metric(rand(20, 3), randint(2, (20, 3)))[0])
>>> fig_, ax_ = metric.plot(values)

Functional Interface

binary_precision_at_fixed_recall

torchmetrics.functional.classification.binary_precision_at_fixed_recall(preds, target, min_recall, thresholds=None, ignore_index=None, validate_args=True)

Compute the highest possible precision value given the minimum recall thresholds provided for binary tasks.

This is done by first calculating the precision-recall curve for different thresholds and the find the precision for a given recall level.

Accepts the following input tensors:

• preds (float tensor): (N, ...). Preds should be a tensor containing probabilities or logits for each observation. If preds has values outside [0,1] range we consider the input to be logits and will auto apply sigmoid per element.

• target (int tensor): (N, ...). Target should be a tensor containing ground truth labels, and therefore only contain {0,1} values (except if ignore_index is specified). The value 1 always encodes the positive class.

Additional dimension ... will be flattened into the batch dimension.

The implementation both supports calculating the metric in a non-binned but accurate version and a binned version that is less accurate but more memory efficient. Setting the thresholds argument to None will activate the non-binned version that uses memory of size \(\mathcal{O}(n_{samples})\) whereas setting the thresholds argument to either an integer, list or a 1d tensor will use a binned version that uses memory of size \(\mathcal{O}(n_{thresholds})\) (constant memory).

Parameters:

• preds (Tensor) – Tensor with predictions

• target (Tensor) – Tensor with true labels

• min_recall (float) – float value specifying minimum recall threshold.

• thresholds (Union[int, List[float], Tensor, None]) –

  Can be one of:

  • If set to None, will use a non-binned approach where thresholds are dynamically calculated from all the data. Most accurate but also most memory consuming approach.

  • If set to an int (larger than 1), will use that number of thresholds linearly spaced from 0 to 1 as bins for the calculation.

  • If set to an list of floats, will use the indicated thresholds in the list as bins for the calculation

  • If set to an 1d Tensor of floats, will use the indicated thresholds in the tensor as bins for the calculation.

• ignore_index (Optional[int]) – Specifies a target value that is ignored and does not contribute to the metric calculation

• validate_args (bool) – bool indicating if input arguments and tensors should be validated for correctness. Set to False for faster computations.

Returns:

a tuple of 2 tensors containing:

• precision: an scalar tensor with the maximum precision for the given precision level

• threshold: an scalar tensor with the corresponding threshold level

Return type:

(tuple)

Example

>>> from torchmetrics.functional.classification import binary_precision_at_fixed_recall
>>> preds = torch.tensor([0, 0.5, 0.7, 0.8])
>>> target = torch.tensor([0, 1, 1, 0])
>>> binary_precision_at_fixed_recall(preds, target, min_recall=0.5, thresholds=None)
(tensor(0.6667), tensor(0.5000))
>>> binary_precision_at_fixed_recall(preds, target, min_recall=0.5, thresholds=5)
(tensor(0.6667), tensor(0.5000))

multiclass_precision_at_fixed_recall

torchmetrics.functional.classification.multiclass_precision_at_fixed_recall(preds, target, num_classes, min_recall, thresholds=None, ignore_index=None, validate_args=True)

Compute the highest possible precision value given the minimum recall thresholds provided for multiclass tasks.

This is done by first calculating the precision-recall curve for different thresholds and the find the precision for a given recall level.

Accepts the following input tensors:

• preds (float tensor): (N, C, ...). Preds should be a tensor containing probabilities or logits for each observation. If preds has values outside [0,1] range we consider the input to be logits and will auto apply softmax per sample.

• target (int tensor): (N, ...). Target should be a tensor containing ground truth labels, and therefore only contain values in the [0, n_classes-1] range (except if ignore_index is specified).

Additional dimension ... will be flattened into the batch dimension.

The implementation both supports calculating the metric in a non-binned but accurate version and a binned version that is less accurate but more memory efficient. Setting the thresholds argument to None will activate the non-binned version that uses memory of size \(\mathcal{O}(n_{samples})\) whereas setting the thresholds argument to either an integer, list or a 1d tensor will use a binned version that uses memory of size \(\mathcal{O}(n_{thresholds} \times n_{classes})\) (constant memory).

Parameters:

• preds (Tensor) – Tensor with predictions

• target (Tensor) – Tensor with true labels

• num_classes (int) – Integer specifying the number of classes

• min_recall (float) – float value specifying minimum recall threshold.

• thresholds (Union[int, List[float], Tensor, None]) –

  Can be one of:

  • If set to None, will use a non-binned approach where thresholds are dynamically calculated from all the data. Most accurate but also most memory consuming approach.

  • If set to an int (larger than 1), will use that number of thresholds linearly spaced from 0 to 1 as bins for the calculation.

  • If set to an list of floats, will use the indicated thresholds in the list as bins for the calculation

  • If set to an 1d Tensor of floats, will use the indicated thresholds in the tensor as bins for the calculation.

• ignore_index (Optional[int]) – Specifies a target value that is ignored and does not contribute to the metric calculation

• validate_args (bool) – bool indicating if input arguments and tensors should be validated for correctness. Set to False for faster computations.

Returns:

a tuple of either 2 tensors or 2 lists containing

• precision: an 1d tensor of size (n_classes, ) with the maximum precision for the given recall level per class

• thresholds: an 1d tensor of size (n_classes, ) with the corresponding threshold level per class

Return type:

(tuple)

Example

>>> from torchmetrics.functional.classification import multiclass_precision_at_fixed_recall
>>> preds = torch.tensor([[0.75, 0.05, 0.05, 0.05, 0.05],
...                       [0.05, 0.75, 0.05, 0.05, 0.05],
...                       [0.05, 0.05, 0.75, 0.05, 0.05],
...                       [0.05, 0.05, 0.05, 0.75, 0.05]])
>>> target = torch.tensor([0, 1, 3, 2])
>>> multiclass_precision_at_fixed_recall(  
...     preds, target, num_classes=5, min_recall=0.5, thresholds=None)
(tensor([1.0000, 1.0000, 0.2500, 0.2500, 0.0000]),
 tensor([7.5000e-01, 7.5000e-01, 5.0000e-02, 5.0000e-02, 1.0000e+06]))
>>> multiclass_precision_at_fixed_recall(  
...     preds, target, num_classes=5, min_recall=0.5, thresholds=5)
(tensor([1.0000, 1.0000, 0.2500, 0.2500, 0.0000]),
 tensor([7.5000e-01, 7.5000e-01, 0.0000e+00, 0.0000e+00, 1.0000e+06]))

multilabel_precision_at_fixed_recall

torchmetrics.functional.classification.multilabel_precision_at_fixed_recall(preds, target, num_labels, min_recall, thresholds=None, ignore_index=None, validate_args=True)

Compute the highest possible precision value given the minimum recall thresholds provided for multilabel tasks.

This is done by first calculating the precision-recall curve for different thresholds and the find the precision for a given recall level.

Accepts the following input tensors:

• preds (float tensor): (N, C, ...). Preds should be a tensor containing probabilities or logits for each observation. If preds has values outside [0,1] range we consider the input to be logits and will auto apply sigmoid per element.

• target (int tensor): (N, C, ...). Target should be a tensor containing ground truth labels, and therefore only contain {0,1} values (except if ignore_index is specified).

Additional dimension ... will be flattened into the batch dimension.

The implementation both supports calculating the metric in a non-binned but accurate version and a binned version that is less accurate but more memory efficient. Setting the thresholds argument to None will activate the non-binned version that uses memory of size \(\mathcal{O}(n_{samples})\) whereas setting the thresholds argument to either an integer, list or a 1d tensor will use a binned version that uses memory of size \(\mathcal{O}(n_{thresholds} \times n_{labels})\) (constant memory).

Parameters:

• preds (Tensor) – Tensor with predictions

• target (Tensor) – Tensor with true labels

• num_labels (int) – Integer specifying the number of labels

• min_recall (float) – float value specifying minimum recall threshold.

• thresholds (Union[int, List[float], Tensor, None]) –

  Can be one of:

  • If set to None, will use a non-binned approach where thresholds are dynamically calculated from all the data. Most accurate but also most memory consuming approach.

  • If set to an int (larger than 1), will use that number of thresholds linearly spaced from 0 to 1 as bins for the calculation.

  • If set to an list of floats, will use the indicated thresholds in the list as bins for the calculation

  • If set to an 1d Tensor of floats, will use the indicated thresholds in the tensor as bins for the calculation.

• ignore_index (Optional[int]) – Specifies a target value that is ignored and does not contribute to the metric calculation

• validate_args (bool) – bool indicating if input arguments and tensors should be validated for correctness. Set to False for faster computations.

Returns:

a tuple of either 2 tensors or 2 lists containing

• precision: an 1d tensor of size (n_classes, ) with the maximum precision for the given recall level per class

• thresholds: an 1d tensor of size (n_classes, ) with the corresponding threshold level per class

Return type:

(tuple)

Example

>>> from torchmetrics.functional.classification import multilabel_precision_at_fixed_recall
>>> preds = torch.tensor([[0.75, 0.05, 0.35],
...                       [0.45, 0.75, 0.05],
...                       [0.05, 0.55, 0.75],
...                       [0.05, 0.65, 0.05]])
>>> target = torch.tensor([[1, 0, 1],
...                        [0, 0, 0],
...                        [0, 1, 1],
...                        [1, 1, 1]])
>>> multilabel_precision_at_fixed_recall(preds, target, num_labels=3, min_recall=0.5, thresholds=None)
(tensor([1.0000, 0.6667, 1.0000]), tensor([0.7500, 0.5500, 0.3500]))
>>> multilabel_precision_at_fixed_recall(preds, target, num_labels=3, min_recall=0.5, thresholds=5)
(tensor([1.0000, 0.6667, 1.0000]), tensor([0.7500, 0.5000, 0.2500]))