Meanabsoluteerror

meanabsoluteerror

Metric MeanAbsoluteError from torchmetrics (torchmetrics.MeanAbsoluteError)

When to invoke this skill

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

Reference signature

from torchmetrics import MeanAbsoluteError

# MeanAbsoluteError(num_outputs: int = 1, **kwargs: Any) -> None

Library docstring

`Compute Mean Absolute Error`_ (MAE).

.. math:: \text{MAE} = \frac{1}{N}\sum_i^N | y_i - \hat{y_i} |

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 values

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

- ``mean_absolute_error`` (:class:`~torch.Tensor`): A tensor with the mean absolute error over the state

Args:
    num_outputs: Number of outputs in multioutput setting
    kwargs: Additional keyword arguments, see :ref:`Metric kwargs` for more info.

Example:
    >>> from torch import tensor
    >>> from torchmetrics.regression import MeanAbsoluteError
    >>> target = tensor([3.0, -0.5, 2.0, 7.0])
    >>> preds = tensor([2.5, 0.0, 2.0, 8.0])
    >>> mean_absolute_error = MeanAbsoluteError()
    >>> mean_absolute_error(preds, target)
    tensor(0.5000)

Example::
    Multioutput mse computation:

    >>> from torch import tensor
    >>> from torchmetrics.regression import MeanAbsoluteError
    >>> target = tensor([[0.0, 0.0, 0.0], [0.0, 0.0, 0.0]])
    >>> preds = tensor([[1.0, 2.0, 3.0], [1.0, 2.0, 3.0]])
    >>> mean_absolute_error = MeanAbsoluteError(num_outputs=3)
    >>> mean_absolute_error(preds, target)
    tensor([1., 2., 3.])

Quick recipe

import torchmetrics as _m
score = _m.MeanAbsoluteError(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).