Weightedmeanabsolutepercentageerror

weightedmeanabsolutepercentageerror

Metric WeightedMeanAbsolutePercentageError from torchmetrics (torchmetrics.WeightedMeanAbsolutePercentageError)

When to invoke this skill

The user has predictions + ground truth and asks to evaluate with WeightedMeanAbsolutePercentageError, or mentions torchmetrics.WeightedMeanAbsolutePercentageError directly, or wants the standard torchmetrics implementation.

Reference signature

from torchmetrics import WeightedMeanAbsolutePercentageError

# WeightedMeanAbsolutePercentageError(**kwargs: Any) -> None

Library docstring

Compute weighted mean absolute percentage error (`WMAPE`_).

The output of WMAPE metric is a non-negative floating point, where the optimal value is 0. It is computes as:

.. math::
    \text{WMAPE} = \frac{\sum_{t=1}^n | y_t - \hat{y}_t | }{\sum_{t=1}^n |y_t| }

Where :math:`y` is a tensor of target values, and :math:`\hat{y}` is a tensor of predictions.

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

- ``preds`` (:class:`~torch.Tensor`): Predictions from model
- ``target`` (:class:`~torch.Tensor`): Ground truth float tensor with shape ``(N,d)``

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

- ``wmape`` (:class:`~torch.Tensor`): A tensor with non-negative floating point wmape value between 0 and 1

Args:
    kwargs: Additional keyword arguments, see :ref:`Metric kwargs` for more info.

Example:
    >>> from torch import randn
    >>> preds = randn(20,)
    >>> target = randn(20,)
    >>> wmape = WeightedMeanAbsolutePercentageError()
    >>> wmape(preds, target)
    tensor(1.3967)

Quick recipe

import torchmetrics as _m
score = _m.WeightedMeanAbsolutePercentageError(y_true, y_pred)

Don'ts

• Don't reimplement when the library version handles edge cases (NaN, ties, empty inputs) better than a hand-rolled formula.
• Always check the library version's argument order — sklearn is (y_true, y_pred) while torchmetrics is (preds, target).