diff --git a/doc/modules/classes.rst b/doc/modules/classes.rst
index 46f8c4017bc47d3c925f28cc597e9739dda008dc..dd73a52078188c91715807acd21f1b6ad2bb26b4 100644
--- a/doc/modules/classes.rst
+++ b/doc/modules/classes.rst
@@ -783,7 +783,6 @@ details.
metrics.classification_report
metrics.cohen_kappa_score
metrics.confusion_matrix
- metrics.dcg_score
metrics.f1_score
metrics.fbeta_score
metrics.hamming_loss
@@ -791,7 +790,6 @@ details.
metrics.jaccard_similarity_score
metrics.log_loss
metrics.matthews_corrcoef
- metrics.ndcg_score
metrics.precision_recall_curve
metrics.precision_recall_fscore_support
metrics.precision_score
diff --git a/doc/modules/model_evaluation.rst b/doc/modules/model_evaluation.rst
index c5384d49fa6582eca4e9d5e214798bb4ca155ae3..0f5f0937a751d840f2175ec94e3d6f46a4549b2f 100644
--- a/doc/modules/model_evaluation.rst
+++ b/doc/modules/model_evaluation.rst
@@ -308,14 +308,6 @@ Some also work in the multilabel case:
recall_score
zero_one_loss
-Some are typically used for ranking:
-
-.. autosummary::
- :template: function.rst
-
- dcg_score
- ndcg_score
-
And some work with binary and multilabel (but not multiclass) problems:
.. autosummary::
diff --git a/sklearn/metrics/__init__.py b/sklearn/metrics/__init__.py
index 93d21a146619a00f582fa0490f7bb697236a7412..eb7cf3c01d115a32b29b895c02b5a426d423c485 100644
--- a/sklearn/metrics/__init__.py
+++ b/sklearn/metrics/__init__.py
@@ -12,8 +12,6 @@ from .ranking import label_ranking_loss
from .ranking import precision_recall_curve
from .ranking import roc_auc_score
from .ranking import roc_curve
-from .ranking import dcg_score
-from .ranking import ndcg_score
from .classification import accuracy_score
from .classification import classification_report
@@ -118,6 +116,4 @@ __all__ = [
'v_measure_score',
'zero_one_loss',
'brier_score_loss',
- 'dcg_score',
- 'ndcg_score'
]
diff --git a/sklearn/metrics/ranking.py b/sklearn/metrics/ranking.py
index 62d8db89e6a5b242b0825efef8477b67be0e2ddc..b7867ce2605c458938d0af259401559f25d5e846 100644
--- a/sklearn/metrics/ranking.py
+++ b/sklearn/metrics/ranking.py
@@ -26,7 +26,7 @@ from scipy.stats import rankdata
from ..utils import assert_all_finite
from ..utils import check_consistent_length
-from ..utils import column_or_1d, check_array, check_X_y
+from ..utils import column_or_1d, check_array
from ..utils.multiclass import type_of_target
from ..utils.extmath import stable_cumsum
from ..utils.sparsefuncs import count_nonzero
@@ -788,91 +788,3 @@ def label_ranking_loss(y_true, y_score, sample_weight=None):
loss[np.logical_or(n_positives == 0, n_positives == n_labels)] = 0.
return np.average(loss, weights=sample_weight)
-
-
-def dcg_score(y_true, y_score, k=5):
- """Discounted cumulative gain (DCG) at rank K.
-
- Parameters
- ----------
- y_true : array, shape = [n_samples]
- Ground truth (true relevance labels).
- y_score : array, shape = [n_samples]
- Predicted scores.
- k : int
- Rank.
-
- Returns
- -------
- score : float
-
- References
- ----------
- .. [1] `Wikipedia entry for the Discounted Cumulative Gain
- <https://en.wikipedia.org/wiki/Discounted_cumulative_gain>`_
- """
- order = np.argsort(y_score)[::-1]
- y_true = np.take(y_true, order[:k])
-
- gain = 2 ** y_true - 1
-
- discounts = np.log2(np.arange(len(y_true)) + 2)
- return np.sum(gain / discounts)
-
-
-def ndcg_score(y_true, y_score, k=5):
- """Normalized discounted cumulative gain (NDCG) at rank K.
-
- Normalized Discounted Cumulative Gain (NDCG) measures the performance of a
- recommendation system based on the graded relevance of the recommended
- entities. It varies from 0.0 to 1.0, with 1.0 representing the ideal
- ranking of the entities.
-
- Parameters
- ----------
- y_true : array, shape = [n_samples]
- Ground truth (true labels represended as integers).
- y_score : array, shape = [n_samples, n_classes]
- Predicted probabilities.
- k : int
- Rank.
-
- Returns
- -------
- score : float
-
- Examples
- --------
- >>> y_true = [1, 0, 2]
- >>> y_score = [[0.15, 0.55, 0.2], [0.7, 0.2, 0.1], [0.06, 0.04, 0.9]]
- >>> ndcg_score(y_true, y_score, k=2)
- 1.0
- >>> y_score = [[0.9, 0.5, 0.8], [0.7, 0.2, 0.1], [0.06, 0.04, 0.9]]
- >>> ndcg_score(y_true, y_score, k=2)
- 0.66666666666666663
-
- References
- ----------
- .. [1] `Kaggle entry for the Normalized Discounted Cumulative Gain
- <https://www.kaggle.com/wiki/NormalizedDiscountedCumulativeGain>`_
- """
- y_score, y_true = check_X_y(y_score, y_true)
-
- # Make sure we use all the labels (max between the length and the higher
- # number in the array)
- lb = LabelBinarizer()
- lb.fit(np.arange(max(np.max(y_true) + 1, len(y_true))))
- binarized_y_true = lb.transform(y_true)
-
- if binarized_y_true.shape != y_score.shape:
- raise ValueError("y_true and y_score have different value ranges")
-
- scores = []
-
- # Iterate over each y_value_true and compute the DCG score
- for y_value_true, y_value_score in zip(binarized_y_true, y_score):
- actual = dcg_score(y_value_true, y_value_score, k)
- best = dcg_score(y_value_true, y_value_true, k)
- scores.append(actual / best)
-
- return np.mean(scores)
diff --git a/sklearn/metrics/tests/test_ranking.py b/sklearn/metrics/tests/test_ranking.py
index db80691663606b3946295e88df2ace846a16e699..951992132f3e3d6029cba8446fbdc0c18525fada 100644
--- a/sklearn/metrics/tests/test_ranking.py
+++ b/sklearn/metrics/tests/test_ranking.py
@@ -29,7 +29,6 @@ from sklearn.metrics import precision_recall_curve
from sklearn.metrics import label_ranking_loss
from sklearn.metrics import roc_auc_score
from sklearn.metrics import roc_curve
-from sklearn.metrics import ndcg_score
from sklearn.exceptions import UndefinedMetricWarning
@@ -738,38 +737,6 @@ def check_zero_or_all_relevant_labels(lrap_score):
[[0.5], [0.5], [0.5], [0.5]]), 1.)
-def test_ndcg_score():
- # Check perfect ranking
- y_true = [1, 0, 2]
- y_score = [
- [0.15, 0.55, 0.2],
- [0.7, 0.2, 0.1],
- [0.06, 0.04, 0.9]
- ]
- perfect = ndcg_score(y_true, y_score)
- assert_equal(perfect, 1.0)
-
- # Check bad ranking with a small K
- y_true = [0, 2, 1]
- y_score = [
- [0.15, 0.55, 0.2],
- [0.7, 0.2, 0.1],
- [0.06, 0.04, 0.9]
- ]
- short_k = ndcg_score(y_true, y_score, k=1)
- assert_equal(short_k, 0.0)
-
- # Check a random scoring
- y_true = [2, 1, 0]
- y_score = [
- [0.15, 0.55, 0.2],
- [0.7, 0.2, 0.1],
- [0.06, 0.04, 0.9]
- ]
- average_ranking = ndcg_score(y_true, y_score, k=2)
- assert_almost_equal(average_ranking, 0.63092975)
-
-
def check_lrap_error_raised(lrap_score):
# Raise value error if not appropriate format
assert_raises(ValueError, lrap_score,