scikit-learn · KaikeWesleyReis · Dec 5, 2020 · Dec 5, 2020 · Dec 5, 2020
diff --git a/sklearn/metrics/__init__.py b/sklearn/metrics/__init__.py
@@ -35,6 +35,7 @@
 from ._classification import zero_one_loss
 from ._classification import brier_score_loss
 from ._classification import multilabel_confusion_matrix
+from ._classification import spherical_payoff_score
 
 from . import cluster
 from .cluster import adjusted_mutual_info_score

diff --git a/sklearn/metrics/_classification.py b/sklearn/metrics/_classification.py
@@ -19,6 +19,7 @@
 #          Saurabh Jha <saurabh.jhaa@gmail.com>
 #          Bernardo Stein <bernardovstein@gmail.com>
 #          Shangwu Yao <shangwuyao@gmail.com>
+#          Kaike Wesley Reis <kaikewesley@hotmail.com>
 # License: BSD 3 clause
 
 
@@ -2504,3 +2505,69 @@ def brier_score_loss(y_true, y_prob, *, sample_weight=None, pos_label=None):
             raise
     y_true = np.array(y_true == pos_label, int)
     return np.average((y_true - y_prob) ** 2, weights=sample_weight)
+
+def spherical_payoff_score(y_true, y_prob, *, sample_weight=None):
+    """Compute the Spherical Payoff.
+    The Spherical Payoff works for binary and multiclass classification 
+    and meausures the model's confidence to predict the correct category.
+    The Spherical Payoff have a defined interval: [0, 1]. Best possible
+    score is 1.0, and the worst is 0.0. It's calculated as the average
+    over all samples.
+
+    Parameters
+    ----------
+    y_true : array, shape (n_samples,)
+        True targets.
+    y_prob : array, shape (n_samples,)
+        Probabilities of the positive class.
+    sample_weight : array-like of shape (n_samples,), default=None
+        Sample weights.
+    Returns
+    -------
+    score : float
+        Spherical Payoff
+
+    Examples
+    --------
+    >>> import numpy as np
+    >>> from sklearn.datasets import load_iris
+    >>> from sklearn.metrics import spherical_payoff_score
+    >>> from sklearn.ensemble import RandomForestClassifier
+    >>> from sklearn.model_selection import train_test_split
+    >>> X, y = load_iris(return_X_y=True)
+    >>> x_train, x_test, y_train, y_test = train_test_split(X, y, test_size=0.10, random_state=1206)
+    >>> rfc = RandomForestClassifier(random_state=1206).fit(x_train, y_train)
+    >>> y_prob = rfc.predict_proba(x_test)
+    >>> spherical_payoff_score(y_test, y_prob)
+    0.983...
+
+    References
+    ----------
+    .. [1] `Wikipedia entry for Scoring Rule, including spherical payoff
+            <https://en.wikipedia.org/wiki/Scoring_rule>`
+    .. [2] `Netica Tutorial for scoring rules
+            <https://www.norsys.com/tutorials/netica/secD/tut_D2.htm>`
+    .. [3] `Guideline for developing and updating Bayesian Belief Networks
+            applied to ecological modelling and conservation
+            <DOI: 10.1139/x06-135>`
+    """
+    y_true = column_or_1d(y_true)
+    assert_all_finite(y_true)
+    assert_all_finite(y_prob)
+    check_consistent_length(y_true, y_prob, sample_weight)
+
+    if y_prob.max() > 1:
+        raise ValueError("y_prob contains values greater than 1.")
+    if y_prob.min() < 0:
+        raise ValueError("y_prob contains values less than 0.")
+
+    # Loop to change the category values to default index
+    categories = np.unique(y_true)
+    for category, idx in zip(categories, np.arange(0, len(categories))):
+        y_true[y_true==category] = idx    
+
+    # Spherical Payoff
+    correct_prob = y_prob[np.arange(y_prob.shape[0]), y_true]
+    sqrt_all_prob = np.sqrt(np.power(y_prob, 2).sum(axis=1))
+
+    return np.average(correct_prob/sqrt_all_prob, weights=sample_weight)