scikit-learn
diff --git a/‎doc/modules/pipeline.rst
Lines changed: 1 addition & 1 deletion b/‎doc/modules/pipeline.rst
Lines changed: 1 addition & 1 deletion
diff --git a/‎sklearn/pipeline.py
Lines changed: 18 additions & 1 deletion b/‎sklearn/pipeline.py
Lines changed: 18 additions & 1 deletion
diff --git a/‎sklearn/tests/test_pipeline.py
Lines changed: 20 additions & 0 deletions b/‎sklearn/tests/test_pipeline.py
Lines changed: 20 additions & 0 deletions
@@ -152,7 +152,7 @@ and ``value`` is an estimator object::
     >>> estimators = [('linear_pca', PCA()), ('kernel_pca', KernelPCA())]
     >>> combined = FeatureUnion(estimators)   
     >>> combined # doctest: +NORMALIZE_WHITESPACE
-    FeatureUnion(n_jobs=1, transformer_list=[('linear_pca', PCA(copy=True,
+    FeatureUnion(fields=None, n_jobs=1, transformer_list=[('linear_pca', PCA(copy=True,
         n_components=None, whiten=False)), ('kernel_pca', KernelPCA(alpha=1.0,
         coef0=1, degree=3, eigen_solver='auto', fit_inverse_transform=False,
         gamma=None, kernel='linear', kernel_params=None, max_iter=None,
 
@@ -308,6 +308,23 @@ class FeatureUnion(BaseEstimator, TransformerMixin):
         Multiplicative weights for features per transformer.
         Keys are transformer names, values the weights.
 
+    fields : list, optional
+        If given needs to be of same length as transformer list.
+        The i-th transformer will receive X[fields[i]]. This allows to select a
+        subset of featues to be processed by each transformer. If None, all
+        features are passed on.
+
+    Examples
+    --------
+    >>> from sklearn.preprocessing import Normalizer
+    >>> union = FeatureUnion([("norm1", Normalizer(norm='l1')),  \
+                              ("norm2", Normalizer(norm='l1'))], \
+                             fields=['subset1', 'subset2'])
+    >>> X = {'subset1': [[0., 1.], [2., 2.]], 'subset2': [[1., 1.], [0., 1.]]}
+    >>> union.fit_transform(X)    # doctest: +NORMALIZE_WHITESPACE
+    array([[ 0. ,  1. ,  0.5,  0.5],
+           [ 0.5,  0.5,  0. ,  1. ]])
+
     """
     def __init__(self, transformer_list, n_jobs=1, transformer_weights=None,
                  fields=None):
@@ -446,7 +463,7 @@ def make_union(*transformers):
     --------
     >>> from sklearn.decomposition import PCA, TruncatedSVD
     >>> make_union(PCA(), TruncatedSVD())    # doctest: +NORMALIZE_WHITESPACE
-    FeatureUnion(n_jobs=1,
+    FeatureUnion(fields=None, n_jobs=1,
                  transformer_list=[('pca', PCA(copy=True, n_components=None,
                                                whiten=False)),
                                    ('truncatedsvd',
 
@@ -394,3 +394,23 @@ def test_classes_property():
     assert_raises(AttributeError, getattr, clf, "classes_")
     clf.fit(X, y)
     assert_array_equal(clf.classes_, np.unique(y))
+
+
+def test_fields():
+    # dictionary
+    X_dict = {'first': [[0], [1], [2]],
+              'second': [[2], [4], [6]]}
+    # recarray
+    X_recarray = np.recarray((3, 1),
+                             dtype=[('first', np.int), ('second', np.int)])
+    X_recarray['first'] = X_dict['first']
+    X_recarray['second'] = X_dict['second']
+
+    for X in [X_dict, X_recarray]:
+        first_feat = FeatureUnion([('trans', TransfT())], fields=['first'])
+        second_feat = FeatureUnion([('trans', TransfT())], fields=['second'])
+        both = FeatureUnion([('trans', TransfT()), ('trans', TransfT())],
+                            fields=['first', 'second'])
+        assert_array_equal(first_feat.fit_transform(X), X['first'])
+        assert_array_equal(second_feat.fit_transform(X), X['second'])
+        assert_array_equal(both.fit_transform(X), np.hstack([X['first'], X['second']]))