scikit-learn · ogrisel · Feb 26, 2015 · Feb 6, 2015 · ogrisel · Feb 16, 2015
diff --git a/doc/whats_new.rst b/doc/whats_new.rst
@@ -401,7 +401,9 @@ API changes summary
     - Estimators will treat input with dtype object as numeric when possible.
       By `Andreas Müller`_
 
-
+    - Estimators now raise `ValueError` consistently when fitted on empty
+      data (less than 1 sample or less than 1 feature for 2D input).
+      By `Olivier Grisel`_.
 
 .. _changes_0_15_2:
 

diff --git a/sklearn/ensemble/forest.py b/sklearn/ensemble/forest.py
@@ -191,10 +191,8 @@ def fit(self, X, y, sample_weight=None):
         self : object
             Returns self.
         """
-        # Convert data
-        # ensure_2d=False because there are actually unit test checking we fail
-        # for 1d. FIXME make this consistent in the future.
-        X = check_array(X, dtype=DTYPE, ensure_2d=False, accept_sparse="csc")
+        # Validate or convert input data
+        X = check_array(X, dtype=DTYPE, accept_sparse="csc")
         if issparse(X):
             # Pre-sort indices to avoid that each individual tree of the
             # ensemble sorts the indices.
@@ -207,7 +205,7 @@ def fit(self, X, y, sample_weight=None):
         if y.ndim == 2 and y.shape[1] == 1:
             warn("A column-vector y was passed when a 1d array was"
                  " expected. Please change the shape of y to "
-                 "(n_samples, ), for example using ravel().",
+                 "(n_samples,), for example using ravel().",
                  DataConversionWarning, stacklevel=2)
 
         if y.ndim == 1:

diff --git a/sklearn/kernel_approximation.py b/sklearn/kernel_approximation.py
@@ -437,10 +437,8 @@ def fit(self, X, y=None):
         X : array-like, shape=(n_samples, n_feature)
             Training data.
         """
-
+        X = check_array(X, accept_sparse='csr')
         rnd = check_random_state(self.random_state)
-        if not sp.issparse(X):
-            X = np.asarray(X)
         n_samples = X.shape[0]
 
         # get basis vectors
@@ -487,6 +485,7 @@ def transform(self, X):
             Transformed data.
         """
         check_is_fitted(self, 'components_')
+        X = check_array(X, accept_sparse='csr')
 
         kernel_params = self._get_kernel_params()
         embedded = pairwise_kernels(X, self.components_,

diff --git a/sklearn/linear_model/coordinate_descent.py b/sklearn/linear_model/coordinate_descent.py
@@ -984,6 +984,8 @@ def fit(self, X, y):
             Target values
         """
         y = np.asarray(y, dtype=np.float64)
+        if y.shape[0] == 0:
+            raise ValueError("y has 0 samples: %r" % y)
 
         if hasattr(self, 'l1_ratio'):
             model_str = 'ElasticNet'

diff --git a/sklearn/preprocessing/label.py b/sklearn/preprocessing/label.py
@@ -22,6 +22,7 @@
 from ..utils.fixes import in1d
 from ..utils import deprecated, column_or_1d
 from ..utils.validation import check_array
+from ..utils.validation import _num_samples
 from ..utils.multiclass import unique_labels
 from ..utils.multiclass import type_of_target
 
@@ -315,6 +316,8 @@ def fit(self, y):
         if 'multioutput' in self.y_type_:
             raise ValueError("Multioutput target data is not supported with "
                              "label binarization")
+        if _num_samples(y) == 0:
+            raise ValueError('y has 0 samples: %r' % y)
 
         self.sparse_input_ = sp.issparse(y)
         self.classes_ = unique_labels(y)
@@ -465,6 +468,9 @@ def label_binarize(y, classes, neg_label=0, pos_label=1,
         # XXX Workaround that will be removed when list of list format is
         # dropped
         y = check_array(y, accept_sparse='csr', ensure_2d=False, dtype=None)
+    else:
+        if _num_samples(y) == 0:
+            raise ValueError('y has 0 samples: %r' % y)
     if neg_label >= pos_label:
         raise ValueError("neg_label={0} must be strictly less than "
                          "pos_label={1}.".format(neg_label, pos_label))

diff --git a/sklearn/tests/test_common.py b/sklearn/tests/test_common.py
@@ -41,6 +41,7 @@
     check_regressors_pickle,
     check_transformer_pickle,
     check_transformers_unfitted,
+    check_estimators_empty_data_messages,
     check_estimators_nan_inf,
     check_estimators_unfitted,
     check_classifiers_one_label,
@@ -99,6 +100,10 @@ def test_non_meta_estimators():
             yield check_fit_score_takes_y, name, Estimator
             yield check_dtype_object, name, Estimator
 
+            # Check that all estimator yield informative messages when
+            # trained on empty datasets
+            yield check_estimators_empty_data_messages, name, Estimator
+
         if name not in CROSS_DECOMPOSITION + ['SpectralEmbedding']:
             # SpectralEmbedding is non-deterministic,
             # see issue #4236

diff --git a/sklearn/utils/estimator_checks.py b/sklearn/utils/estimator_checks.py
@@ -12,6 +12,7 @@
 
 from sklearn.externals.six.moves import zip
 from sklearn.utils.testing import assert_raises
+from sklearn.utils.testing import assert_raise_message
 from sklearn.utils.testing import assert_equal
 from sklearn.utils.testing import assert_true
 from sklearn.utils.testing import assert_false
@@ -346,6 +347,24 @@ def check_estimators_dtypes(name, Estimator):
                 pass
 
 
+def check_estimators_empty_data_messages(name, Estimator):
+    e = Estimator()
+    set_fast_parameters(e)
+    set_random_state(e, 1)
+
+    X_zero_samples = np.empty(0).reshape(0, 3)
+    # The precise message can change depending on whether X or y is
+    # validated first. Let us test the type of exception only:
+    assert_raises(ValueError, e.fit, X_zero_samples, [])
+
+    X_zero_features = np.empty(0).reshape(3, 0)
+    # the following y should be accepted by both classifiers and regressors
+    # and ignored by unsupervised models
+    y = multioutput_estimator_convert_y_2d(name, np.array([1, 0, 1]))
+    msg = "0 feature(s) (shape=(3, 0)) while a minimum of 1 is required."
+    assert_raise_message(ValueError, msg, e.fit, X_zero_features, y)
+
+
 def check_estimators_nan_inf(name, Estimator):
     rnd = np.random.RandomState(0)
     X_train_finite = rnd.uniform(size=(10, 3))

diff --git a/sklearn/utils/tests/test_validation.py b/sklearn/utils/tests/test_validation.py
@@ -237,6 +237,11 @@ def test_check_array_min_samples_and_features_messages():
     assert_raise_message(ValueError, msg, check_X_y, X, y,
                          ensure_min_samples=2)
 
+    # The same message is raised if the data has 2 dimensions even if this is
+    # not mandatory
+    assert_raise_message(ValueError, msg, check_X_y, X, y,
+                         ensure_min_samples=2, ensure_2d=False)
+
     # Simulate a model that would require at least 3 features (e.g. SelectKBest
     # with k=3)
     X = np.ones((10, 2))
@@ -245,6 +250,11 @@ def test_check_array_min_samples_and_features_messages():
     assert_raise_message(ValueError, msg, check_X_y, X, y,
                          ensure_min_features=3)
 
+    # Only the feature check is enabled whenever the number of dimensions is 2
+    # even if allow_nd is enabled:
+    assert_raise_message(ValueError, msg, check_X_y, X, y,
+                         ensure_min_features=3, allow_nd=True)
+
     # Simulate a case where a pipeline stage as trimmed all the features of a
     # 2D dataset.
     X = np.empty(0).reshape(10, 0)

diff --git a/sklearn/utils/validation.py b/sklearn/utils/validation.py
@@ -307,8 +307,9 @@ def check_array(array, accept_sparse=None, dtype="numeric", order=None, copy=Fal
     ensure_min_features : int (default=1)
         Make sure that the 2D array has some minimum number of features
         (columns). The default value of 1 rejects empty datasets.
-        This check is only enforced when ``ensure_2d`` is True and
-        ``allow_nd`` is False. Setting to 0 disables this check.
+        This check is only enforced when the input data has effectively 2
+        dimensions or is originally 1D and ``ensure_2d`` is True. Setting to 0
+        disables this check.
 
     Returns
     -------
@@ -347,7 +348,8 @@ def check_array(array, accept_sparse=None, dtype="numeric", order=None, copy=Fal
                              " minimum of %d is required."
                              % (n_samples, shape_repr, ensure_min_samples))
 
-    if ensure_min_features > 0 and ensure_2d and not allow_nd:
+
+    if ensure_min_features > 0 and array.ndim == 2:
         n_features = array.shape[1]
         if n_features < ensure_min_features:
             raise ValueError("Found array with %d feature(s) (shape=%s) while"
@@ -411,13 +413,16 @@ def check_X_y(X, y, accept_sparse=None, dtype="numeric", order=None, copy=False,
         axis (rows for a 2D array).
 
     ensure_min_features : int (default=1)
-        Make sure that the 2D X has some minimum number of features
+        Make sure that the 2D array has some minimum number of features
         (columns). The default value of 1 rejects empty datasets.
-        This check is only enforced when ``ensure_2d`` is True and
-        ``allow_nd`` is False.
+        This check is only enforced when X has effectively 2 dimensions or
+        is originally 1D and ``ensure_2d`` is True. Setting to 0 disables
+        this check.
+
     y_numeric : boolean (default=False)
         Whether to ensure that y has a numeric type. If dtype of y is object,
-        it is converted to float64. Should only be used for regression algorithms.
+        it is converted to float64. Should only be used for regression
+        algorithms.
 
     Returns
     -------
@@ -428,7 +433,8 @@ def check_X_y(X, y, accept_sparse=None, dtype="numeric", order=None, copy=False,
                     ensure_2d, allow_nd, ensure_min_samples,
                     ensure_min_features)
     if multi_output:
-        y = check_array(y, 'csr', force_all_finite=True, ensure_2d=False, dtype=None)
+        y = check_array(y, 'csr', force_all_finite=True, ensure_2d=False,
+                        dtype=None)
     else:
         y = column_or_1d(y, warn=True)
         _assert_all_finite(y)