[DOC] solves several small raised issues (#266)

* solves #258, #261, and #256

Author: RobinVogel

doc/getting_started.rst

@@ -29,11 +29,14 @@ Quick start
 This example loads the iris dataset, and evaluates a k-nearest neighbors
 algorithm on an embedding space learned with `NCA`.
 
->>> from metric_learn import NCA
->>> from sklearn.datasets import load_iris
->>> from sklearn.model_selection import cross_val_score
->>> from sklearn.pipeline import make_pipeline
->>>
->>> X, y = load_iris(return_X_y=True)
->>> clf = make_pipeline(NCA(), KNeighborsClassifier())
->>> cross_val_score(clf, X, y)
+::
+
+    from metric_learn import NCA
+    from sklearn.datasets import load_iris
+    from sklearn.model_selection import cross_val_score
+    from sklearn.pipeline import make_pipeline
+    from sklearn.neighbors import KNeighborsClassifier
+    
+    X, y = load_iris(return_X_y=True)
+    clf = make_pipeline(NCA(), KNeighborsClassifier())
+    cross_val_score(clf, X, y)

doc/supervised.rst

@@ -374,7 +374,12 @@ Supervised versions of weakly-supervised algorithms
 Each :ref:`weakly-supervised algorithm <weakly_supervised_section>`
 has a supervised version of the form `*_Supervised` where similarity tuples are
 randomly generated from the labels information and passed to the underlying
-algorithm.
+algorithm. 
+
+.. warning::
+    Supervised versions of weakly-supervised algorithms interpret label -1
+    (or any negative label) as a point with unknown label.
+    Those points are discarded in the learning process.
 
 For pairs learners (see :ref:`learning_on_pairs`), pairs (tuple of two points
 from the dataset), and pair labels (`int` indicating whether the two points

metric_learn/itml.py

@@ -273,6 +273,7 @@ class ITML_Supervised(_BaseITML, TransformerMixin):
            be removed in 0.6.0.
   num_constraints: int, optional
       number of constraints to generate
+      (`20 * num_classes**2` constraints by default)
   bounds : Not used
          .. deprecated:: 0.5.0
         `bounds` was deprecated in version 0.5.0 and will

metric_learn/lsml.py

@@ -276,6 +276,7 @@ class LSML_Supervised(_BaseLSML, TransformerMixin):
        be removed in 0.6.0.
   num_constraints: int, optional
       number of constraints to generate
+      (`20 * num_classes**2` constraints by default)
   weights : (m,) array of floats, optional
       scale factor for each constraint
   verbose : bool, optional

metric_learn/mmc.py

@@ -498,6 +498,7 @@ class MMC_Supervised(_BaseMMC, TransformerMixin):
        be removed in 0.6.0.
   num_constraints: int, optional
       number of constraints to generate
+      (`20 * num_classes**2` constraints by default)
   init : None, string or numpy array, optional (default=None)
       Initialization of the Mahalanobis matrix. Possible options are
       'identity', 'covariance', 'random', and a numpy array of

metric_learn/sdml.py

@@ -302,6 +302,7 @@ class SDML_Supervised(_BaseSDML, TransformerMixin):
        be removed in 0.6.0.
   num_constraints : int, optional (default=None)
       number of constraints to generate
+      (`20 * num_classes**2` constraints by default)
   verbose : bool, optional (default=False)
       if True, prints information while learning
   preprocessor : array-like, shape=(n_samples, n_features) or callable