D2 Log Loss Score

d2-log-loss-score

Metric d2_log_loss_score from scikit-learn (sklearn.metrics.d2_log_loss_score)

When to invoke this skill

The user has predictions + ground truth and asks to evaluate with d2_log_loss_score, or mentions sklearn.metrics.d2_log_loss_score directly, or wants the standard scikit-learn implementation.

Reference signature

from sklearn.metrics import d2_log_loss_score

# d2_log_loss_score(y_true, y_pred, *, sample_weight=None, labels=None)

Library docstring

:math:`D^2` score function, fraction of log loss explained.

Best possible score is 1.0 and it can be negative (because the model can be
arbitrarily worse). A model that always predicts the per-class proportions
of `y_true`, disregarding the input features, gets a D^2 score of 0.0.

Read more in the :ref:`User Guide <d2_score_classification>`.

.. versionadded:: 1.5

Parameters
----------
y_true : array-like or label indicator matrix
    The actuals labels for the n_samples samples.

y_pred : array-like of shape (n_samples, n_classes) or (n_samples,)
    Predicted probabilities, as returned by a classifier's
    predict_proba method. If ``y_pred.shape = (n_samples,)``
    the probabilities provided are assumed to be that of the
    positive class. The labels in ``y_pred`` are assumed to be
    ordered alphabetically, as done by
    :class:`~sklearn.preprocessing.LabelBinarizer`.

sample_weight : array-like of shape (n_samples,), default=None
    Sample weights.

labels : array-like, default=None
    If not provided, labels will be inferred from y_true. If ``labels``
    is ``None`` and ``y_pred`` has shape (n_samples,) the labels are
    assumed to be binary and are inferred from ``y_true``.

Returns
-------
d2 : float or ndarray of floats
    The D^2 score.

Notes
-----
This is not a symmetric function.

Like R^2, D^2 score may be negative (it need not actually be the square of
a quantity D).

This metric is not well-defined for a single sample and will return a NaN
value if n_samples is less than two.

Quick recipe

import sklearn.metrics as _m
score = _m.d2_log_loss_score(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).