torcheval.metrics.WindowedMeanSquaredError

class torcheval.metrics.WindowedMeanSquaredError(*, num_tasks: int = 1, max_num_updates: int = 100, enable_lifetime: bool = True, multioutput: str = 'uniform_average', device: device | None = None)

The windowed version of Mean Squared Error that provides both windowed and liftime values. Windowed value is calculated from the input and target of the last window_size number of update() calls. Lifetime value is calculated from all past input and target of update() calls.

\[\text{MSE} = \frac{1}{N}\sum_i^N(y_i - \hat{y_i})^2\]

Where \(y\) is a tensor of target values, and \(\hat{y}\) is a tensor of input values.

Parameters:

• num_tasks (int) – Number of tasks that need WindowedMeanSquaredError calculation. Default value is 1. WindowedMeanSquaredError for each task will be calculated independently.

• max_num_updates (int) – The max window size that can accommodate the number of updates.

• enable_lifetime (bool) – A boolean indicator whether to calculate lifetime values.

• multioutput (str, Optional) –

  • 'uniform_average' [default]: Return scores of all outputs are averaged with uniform weight.

  • 'raw_values': Return a full set of scores.

Raises:

ValueError –

• If value of multioutput is not one of (raw_values, uniform_average). - If the dimension of input or target is not 1D or 2D. - If the input and target do not have the same size. - If the first dimension of input, target and sample_weight are not the same.

Examples::

>>> metric = MeanSquaredError(max_num_updates=1, enable_lifetime=False)
>>> metric.update(torch.tensor([[0.2, 0.3], [0.4, 0.6]]), torch.tensor([[0.1, 0.3], [0.6, 0.7]]))
>>> metric.update(torch.tensor([[0.9, 0.5], [0.3, 0.5]]), torch.tensor([[0.5, 0.8], [0.2, 0.8]]))
>>> metric.compute()
tensor(0.0875)

>>> metric = MeanSquaredError(max_num_updates=1, enable_lifetime=True)
>>> metric.update(torch.tensor([[0.2, 0.3], [0.4, 0.6]]), torch.tensor([[0.1, 0.3], [0.6, 0.7]]))
>>> metric.update(torch.tensor([[0.9, 0.5], [0.3, 0.5]]), torch.tensor([[0.5, 0.8], [0.2, 0.8]]))
>>> metric.compute()
(tensor(0.0512), tensor(0.0875))

>>> metric = MeanSquaredError(max_num_updates=1, enable_lifetime=False, multioutput="raw_values")
>>> metric.update(torch.tensor([[0.2, 0.3], [0.4, 0.6]]), torch.tensor([[0.1, 0.3], [0.6, 0.7]]))
>>> metric.update(torch.tensor([[0.9, 0.5], [0.3, 0.5]]), torch.tensor([[0.5, 0.8], [0.2, 0.8]]))
>>> metric.update(input, target)
>>> metric.compute()
tensor([0.0850, 0.0900])

__init__(*, num_tasks: int = 1, max_num_updates: int = 100, enable_lifetime: bool = True, multioutput: str = 'uniform_average', device: device | None = None) → None

Initialize a metric object and its internal states.

Use self._add_state() to initialize state variables of your metric class. The state variables should be either torch.Tensor, a list of torch.Tensor, a dictionary with torch.Tensor as values, or a deque of torch.Tensor.

Methods

__init__(*[, num_tasks, max_num_updates, ...])	Initialize a metric object and its internal states.
compute()	Return the Mean Squared Error.
load_state_dict(state_dict[, strict])	Loads metric state variables from state_dict.
merge_state(metrics)	Implement this method to update the current metric's state variables to be the merged states of the current metric and input metrics.
reset()	Reset the metric state variables to their default value.
state_dict()	Save metric state variables in state_dict.
to(device, *args, **kwargs)	Move tensors in metric state variables to device.
update(input, target, *[, sample_weight])	Update states with the ground truth values and predictions.

Attributes

device

The last input device of Metric.to().