Minkowskidistance

minkowskidistance

Metric MinkowskiDistance from torchmetrics (torchmetrics.MinkowskiDistance)

When to invoke this skill

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

Reference signature

from torchmetrics import MinkowskiDistance

# MinkowskiDistance(p: float, **kwargs: Any) -> None

Library docstring

Compute `Minkowski Distance`_.

.. math::
    d_{\text{Minkowski}} = \sum_{i}^N (| y_i - \hat{y_i} |^p)^\frac{1}{p}

where
    :math: `y` is a tensor of target values,
    :math: `\hat{y}` is a tensor of predictions,
    :math: `\p` is a non-negative integer or floating-point number

This metric can be seen as generalized version of the standard euclidean distance which corresponds to minkowski
distance with p=2.

Args:
    p: int or float larger than 1, exponent to which the difference between preds and target is to be raised
    kwargs: Additional keyword arguments, see :ref:`Metric kwargs` for more info.

Example:
    >>> from torchmetrics.regression import MinkowskiDistance
    >>> target = tensor([1.0, 2.8, 3.5, 4.5])
    >>> preds = tensor([6.1, 2.11, 3.1, 5.6])
    >>> minkowski_distance = MinkowskiDistance(3)
    >>> minkowski_distance(preds, target)
    tensor(5.1220)

Quick recipe

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