[MRG+1] Threshold for pairs learners

metric_learn/base_metric.py

@@ -374,8 +374,10 @@ def score(self, pairs, y):
  def set_default_threshold(self, pairs, y):
  """Returns a threshold that is the mean between the similar metrics
  mean, and the dissimilar metrics mean"""
- similar_threshold = np.mean(self.decision_function(pairs[y==1]))
- dissimilar_threshold = np.mean(self.decision_function(pairs[y==1]))
+ similar_threshold = np.mean(self.decision_function(
+ pairs[(y == 1).ravel()]))
+ dissimilar_threshold = np.mean(self.decision_function(
+ pairs[(y == -1).ravel()]))
  self.threshold_ = np.mean([similar_threshold, dissimilar_threshold])
 
 
@@ -458,9 +460,14 @@ def score(self, quadruplets, y=None):
  score : float
  The quadruplets score.
  """
- quadruplets = check_input(quadruplets, y, type_of_inputs='tuples',
- preprocessor=self.preprocessor_,
- estimator=self, tuple_size=self._tuple_size)
+ checked_input = check_input(quadruplets, y, type_of_inputs='tuples',
+ preprocessor=self.preprocessor_,
+ estimator=self, tuple_size=self._tuple_size)
+ # checked_input will be of the form `(checked_quadruplets, checked_y)` if
+ # `y` is not None, or just `checked_quadruplets` if `y` is None
+ quadruplets = checked_input if y is None else checked_input[0]
  if y is None:
  y = np.ones(quadruplets.shape[0])
+ else:
+ y = checked_input[1]
  return accuracy_score(y, self.predict(quadruplets))

test/test_quadruplets_classifiers.py

@@ -25,35 +25,11 @@ def test_predict_only_one_or_minus_one(estimator, build_dataset,
  assert np.isin(predictions, [-1, 1]).all()
 
 
-@pytest.mark.parametrize('with_preprocessor', [True, False])
-@pytest.mark.parametrize('estimator, build_dataset', quadruplets_learners,
- ids=ids_quadruplets_learners)
-def test_predict_monotonous(estimator, build_dataset,
- with_preprocessor):
- """Test that there is a threshold distance separating points labeled as
- similar and points labeled as dissimilar """
- input_data, labels, preprocessor, _ = build_dataset(with_preprocessor)
- estimator = clone(estimator)
- estimator.set_params(preprocessor=preprocessor)
- set_random_state(estimator)
- (quadruplets_train,
- quadruplets_test, y_train, y_test) = train_test_split(input_data, labels)
- estimator.fit(quadruplets_train, y_train)
- distances = estimator.score_quadruplets(quadruplets_test)
- predictions = estimator.predict(quadruplets_test)
- min_dissimilar = np.min(distances[predictions == -1])
- max_similar = np.max(distances[predictions == 1])
- assert max_similar <= min_dissimilar
- separator = np.mean([min_dissimilar, max_similar])
- assert (predictions[distances > separator] == -1).all()
- assert (predictions[distances < separator] == 1).all()
-
-
 @pytest.mark.parametrize('with_preprocessor', [True, False])
 @pytest.mark.parametrize('estimator, build_dataset', quadruplets_learners,
  ids=ids_quadruplets_learners)
 def test_raise_not_fitted_error_if_not_fitted(estimator, build_dataset,
- with_preprocessor):
+  with_preprocessor):
  """Test that a NotFittedError is raised if someone tries to predict and
  the metric learner has not been fitted."""
  input_data, labels, preprocessor, _ = build_dataset(with_preprocessor)