Metrics¶

Module implementing metrics for torch tensors

borch.metrics.accuracy(rv)¶

Calculates the accuracy, i.e. how much agreement between two long tensors. It will return values between 0 and 1. :param rv: an observed RandomVariable. :type rv: borch.RandomVariable

Returns: tensor, with the calculated accuracy

Examples

>>> from borch import RandomVariable, distributions
>>> import torch
>>> rv = distributions.Categorical(logits=torch.randn(4))
>>> acc = accuracy(rv)

Notes

This function does not support gradient trough it

borch.metrics.all_metrics(rv)¶

Calculates all valid performance metrics of an observed RandomVariable,: rv (borch.RandomVariable): an observed RandomVariable.

Returns: dict, with performance measures

Notes

If no performance measures is defined for the support of the distribution, an empty dict will be returned.

Examples

>>> import torch
>>> from borch import distributions
>>> rv = distributions.Normal(0, 1)
>>> met = all_metrics(rv)

borch.metrics.mean_squared_error(rv)¶

Measures the averaged element-wise mean squared error of an observed RandomVariable :param rv: an observed RandomVariable. :type rv: borch.RandomVariable

Returns: tensor, with the mean squared error

Examples

>>> from borch import RandomVariable, distributions
>>> import torch
>>> rv = distributions.Normal(torch.randn(10), torch.randn(10).exp())
>>> mse = mean_squared_error(rv)