Implementation:Scikit learn Scikit learn Make Scorer

Metadata

• Domains: Statistics, Model_Evaluation
• Source File: sklearn/metrics/_scorer.py
• Last Updated: 2026-02-08 15:00 GMT

Overview

Concrete tool for converting metric functions into scorer callables provided by scikit-learn. This implementation covers two complementary APIs: make_scorer, which constructs custom scorer objects from arbitrary metric or loss functions, and get_scorer_names, which lists all pre-defined scorer names available in the registry.

API Signatures

make_scorer

from sklearn.metrics import make_scorer

make_scorer(
    score_func,
    *,
    response_method="predict",
    greater_is_better=True,
    **kwargs
)

Parameters:

• score_func (callable) -- Score function (or loss function) with signature score_func(y, y_pred, **kwargs).
• response_method (str or list/tuple of str, default="predict") -- Specifies which estimator method to use to obtain predictions. Valid options: "predict", "predict_proba", "decision_function". If a list or tuple, the first available method on the estimator is used. (Added in version 1.4.)
• greater_is_better (bool, default=True) -- Whether score_func is a score function (high is good) or a loss function (low is good). When False, the scorer object will negate the outcome of score_func.
• **kwargs -- Additional parameters to be passed to score_func at every invocation.

Returns:

• scorer (callable) -- A callable object with signature scorer(estimator, X, y_true, **kwargs) that returns a scalar score where greater is better.

Example -- Custom F-beta scorer:

from sklearn.metrics import fbeta_score, make_scorer

# Create an F2 scorer (beta=2 emphasizes recall)
ftwo_scorer = make_scorer(fbeta_score, beta=2)

# Use with GridSearchCV
from sklearn.model_selection import GridSearchCV
from sklearn.svm import LinearSVC

grid = GridSearchCV(
    LinearSVC(), param_grid={'C': [1, 10]},
    scoring=ftwo_scorer
)

Example -- Loss function with sign flipping:

from sklearn.metrics import mean_squared_error, make_scorer

# mean_squared_error is a loss (lower is better), so negate it
neg_mse_scorer = make_scorer(mean_squared_error, greater_is_better=False)

Example -- Probability-based scorer:

from sklearn.metrics import log_loss, make_scorer

# log_loss needs predict_proba and is a loss function
neg_log_loss_scorer = make_scorer(
    log_loss,
    response_method="predict_proba",
    greater_is_better=False,
)

get_scorer_names

from sklearn.metrics import get_scorer_names

get_scorer_names()

Parameters: None.

Returns:

• list of str -- Sorted list of all available pre-defined scorer names. These names can be passed as strings to any scoring parameter in scikit-learn or to get_scorer() to retrieve the corresponding scorer object.

Example:

from sklearn.metrics import get_scorer_names

all_scorers = get_scorer_names()
print(type(all_scorers))   # <class 'list'>
print(all_scorers[:3])     # ['accuracy', 'adjusted_mutual_info_score', 'adjusted_rand_score']
print("roc_auc" in all_scorers)  # True

Sign Convention Summary

Metric Type	greater_is_better	Pre-defined Name Example	Effect
Score (higher is better)	True	'accuracy', 'r2'	Score passed through unchanged
Loss (lower is better)	False	'neg_mean_squared_error'	Score is negated (multiplied by -1)

When using pre-defined scorer names with the neg_ prefix, the sign has already been flipped. When constructing a custom scorer with make_scorer, you must set greater_is_better=False for loss functions to achieve the same effect.

Related Pages

• Principle:Scikit_learn_Scikit_learn_Scoring_Configuration