Added verbose and a test case

sklearn/neural_network/extreme_learning_machines.py

@@ -65,9 +65,10 @@ class BaseELM(six.with_metaclass(ABCMeta, BaseEstimator)):
     }
 
     @abstractmethod
-    def __init__(self, n_hidden, activation, random_state):
+    def __init__(self, n_hidden, activation, verbose, random_state):
         self.n_hidden = n_hidden
         self.activation = activation
+        self.verbose = verbose
         self.random_state = random_state
 
     def _validate_params(self):
@@ -143,6 +144,12 @@ def fit(self, X, y):
         H = self._get_hidden_activations(X)
         self.coef_output_ = safe_sparse_dot(pinv2(H), y)
 
+        if self.verbose:
+            # compute training square error
+            cost = np.sum(
+                (y - self.decision_function(X)) ** 2) / (2 * n_samples)
+            print("Training square error = %f" % (cost))
+
         return self
 
     def decision_function(self, X):
@@ -188,6 +195,9 @@ class ELMClassifier(BaseELM, ClassifierMixin):
 
          - 'tanh' for the hyperbolic tangent.
 
+    verbose : bool, optional, default False
+        Whether to print training score to stdout.
+
     random_state : int or RandomState, optional, default None
         State of or seed for random number generator.
 
@@ -202,9 +212,11 @@ class ELMClassifier(BaseELM, ClassifierMixin):
 
     """
 
-    def __init__(self, n_hidden=100, activation='tanh', random_state=None):
+    def __init__(self, n_hidden=100, activation='tanh', verbose=False,
+                 random_state=None):
 
-        super(ELMClassifier, self).__init__(n_hidden, activation, random_state)
+        super(ELMClassifier, self).__init__(
+            n_hidden, activation, verbose, random_state)
 
         self._lbin = LabelBinarizer(-1, 1)
         self.classes_ = None
@@ -310,6 +322,9 @@ class ELMRegressor(BaseELM, RegressorMixin):
 
          - 'tanh' for the hyperbolic tangent.
 
+    verbose : bool, optional, default False
+        Whether to print training score to stdout.
+
     random_state : int or RandomState, optional, default None
         State of or seed for random number generator.
 
@@ -324,9 +339,11 @@ class ELMRegressor(BaseELM, RegressorMixin):
 
     """
 
-    def __init__(self, n_hidden=100, activation='tanh', random_state=None):
+    def __init__(self, n_hidden=100, activation='tanh', verbose=False,
+                 random_state=None):
 
-        super(ELMRegressor, self).__init__(n_hidden, activation, random_state)
+        super(ELMRegressor, self).__init__(
+            n_hidden, activation, verbose, random_state)
 
         self.classes_ = None
 

sklearn/neural_network/tests/test_elm.py

@@ -91,21 +91,6 @@ def test_regression():
         assert_greater(elm.score(X, y), 0.95)
 
 
-def test_multilabel_classification():
-    """
-    Tests whether multi-label classification works as expected
-    """
-    # test fit method
-    X, y = make_multilabel_classification(
-        n_samples=50, random_state=random_state)
-    for activation in ACTIVATION_TYPES:
-        elm = ELMClassifier(n_hidden=50,
-                            activation=activation,
-                            random_state=random_state)
-        elm.fit(X, y)
-        assert_greater(elm.score(X, y), 0.95)
-
-
 def test_multioutput_regression():
     """
     Tests whether multi-output regression works as expected
@@ -193,3 +178,28 @@ def test_sparse_matrices():
     pred1 = elm.predict(X)
     pred2 = elm.predict(X_sparse)
     assert_array_equal(pred1, pred2)
+
+
+def test_verbose():
+    """
+    Tests whether verbose displays the correct results
+    """
+    X = Xboston
+    y = yboston[:, np.newaxis]
+
+    elm = ELMRegressor(
+        verbose=True)
+    old_stdout = sys.stdout
+    sys.stdout = output = StringIO()
+
+    elm.fit(X, y)
+    sys.stdout = old_stdout
+
+    # manually compute cost
+    cost = np.sum(
+        (y - elm.decision_function(X)) ** 2) / (2 * X.shape[0])
+
+    # trim cost to 4 decimal places
+    cost = str(cost)[:6]
+
+    assert cost in output.getvalue()