- class torchmetrics.MultioutputWrapper(base_metric, num_outputs, output_dim=- 1, remove_nans=True, squeeze_outputs=True)[source]
Wrap a base metric to enable it to support multiple outputs.
Several torchmetrics metrics, such as
torchmetrics.regression.spearman.SpearmanCorrcoeflack support for multioutput mode. This class wraps such metrics to support computing one metric per output. Unlike specific torchmetric metrics, it doesn’t support any aggregation across outputs. This means if you set
.compute()will return a Tensor of dimension
...represents the dimensions the metric returns when not wrapped.
In addition to enabling multioutput support for metrics that lack it, this class also supports, albeit in a crude fashion, dealing with missing labels (or other data). When
remove_nansis passed, the class will remove the intersection of NaN containing “rows” upon each update for each output. For example, suppose a user uses MultioutputWrapper to wrap
torchmetrics.regression.r2.R2Scorewith 2 outputs, one of which occasionally has missing labels for classes like
R2Scoreis that this class supports removing
remove_nans) on a per-output basis. When
remove_nansis passed the wrapper will remove all rows
int) – Expected dimensionality of the output dimension. This parameter is used to determine the number of distinct metrics we need to track.
int) – Dimension on which output is expected. Note that while this provides some flexibility, the output dimension must be the same for all inputs to update. This applies even for metrics such as Accuracy where the labels can have a different number of dimensions than the predictions. This can be worked around if the output dimension can be set to -1 for both, even if -1 corresponds to different dimensions in different inputs.
bool) – Whether to remove the intersection of rows containing NaNs from the values passed through to each underlying metric. Proper operation requires all tensors passed to update to have dimension
(N, ...)where N represents the length of the batch or dataset being passed in.
bool) – If
True, will squeeze the 1-item dimensions left after
index_selectis applied. This is sometimes unnecessary but harmless for metrics such as R2Score but useful for certain classification metrics that can’t handle additional 1-item dimensions.
>>> # Mimic R2Score in `multioutput`, `raw_values` mode: >>> import torch >>> from torchmetrics import MultioutputWrapper, R2Score >>> target = torch.tensor([[0.5, 1], [-1, 1], [7, -6]]) >>> preds = torch.tensor([[0, 2], [-1, 2], [8, -5]]) >>> r2score = MultioutputWrapper(R2Score(), 2) >>> r2score(preds, target) [tensor(0.9654), tensor(0.9082)]
Initializes internal Module state, shared by both nn.Module and ScriptModule.
- forward(*args, **kwargs)[source]
Call underlying forward methods and aggregate the results if they’re non-null.
We override this method to ensure that state variables get copied over on the underlying metrics.
- Return type