lithuak
diff --git a/‎doc/whats_new/v0.21.rst
Lines changed: 11 additions & 0 deletions b/‎doc/whats_new/v0.21.rst
Lines changed: 11 additions & 0 deletions
diff --git a/‎sklearn/preprocessing/data.py
Lines changed: 12 additions & 3 deletions b/‎sklearn/preprocessing/data.py
Lines changed: 12 additions & 3 deletions
diff --git a/‎sklearn/preprocessing/tests/test_data.py
Lines changed: 11 additions & 0 deletions b/‎sklearn/preprocessing/tests/test_data.py
Lines changed: 11 additions & 0 deletions
@@ -48,10 +48,21 @@ Support for Python 3.4 and below has been officially dropped.
   to set and that scales better, by :user:`Shane <espg>` and
   :user:`Adrin Jalali <adrinjalali>`.
 
+:mod:`sklearn.preprocessing`
+............................
+
+- |Efficiency| |API| Speed improvement in :class:`preprocessing.PolynomialFeatures`,
+  in the dense case. Also added a new parameter ``order`` which controls output
+  order for further speed performances. :issue:`12251` by `Tom Dupre la Tour`_.
+
+:mod:`sklearn.datasets`
+............................
+
 - |Fix| Fixed integer overflow in :func:`datasets.make_classification`
   for values of ``n_informative`` parameter larger than 64.
   :issue:10811 by :user:`Roman Feldbauer <VarIr>`.
 
+
 Multiple modules
 ................
 
 
@@ -1323,6 +1323,12 @@ class PolynomialFeatures(BaseEstimator, TransformerMixin):
         all polynomial powers are zero (i.e. a column of ones - acts as an
         intercept term in a linear model).
 
+    order : str in {'C', 'F'}, default 'C'
+        Order of output array in the dense case. 'F' order is faster to
+        compute, but may slow down subsequent estimators.
+
+        .. versionadded:: 0.21
+
     Examples
     --------
     >>> X = np.arange(6).reshape(3, 2)
@@ -1363,10 +1369,12 @@ class PolynomialFeatures(BaseEstimator, TransformerMixin):
     See :ref:`examples/linear_model/plot_polynomial_interpolation.py
     <sphx_glr_auto_examples_linear_model_plot_polynomial_interpolation.py>`
     """
-    def __init__(self, degree=2, interaction_only=False, include_bias=True):
+    def __init__(self, degree=2, interaction_only=False, include_bias=True,
+                 order='C'):
         self.degree = degree
         self.interaction_only = interaction_only
         self.include_bias = include_bias
+        self.order = order
 
     @staticmethod
     def _combinations(n_features, degree, interaction_only, include_bias):
@@ -1454,7 +1462,7 @@ def transform(self, X):
         """
         check_is_fitted(self, ['n_input_features_', 'n_output_features_'])
 
-        X = check_array(X, dtype=FLOAT_DTYPES, accept_sparse='csc')
+        X = check_array(X, order='F', dtype=FLOAT_DTYPES, accept_sparse='csc')
         n_samples, n_features = X.shape
 
         if n_features != self.n_input_features_:
@@ -1475,7 +1483,8 @@ def transform(self, X):
                     columns.append(sparse.csc_matrix(np.ones((X.shape[0], 1))))
             XP = sparse.hstack(columns, dtype=X.dtype).tocsc()
         else:
-            XP = np.empty((n_samples, self.n_output_features_), dtype=X.dtype)
+            XP = np.empty((n_samples, self.n_output_features_), dtype=X.dtype,
+                          order=self.order)
             for i, comb in enumerate(combinations):
                 XP[:, i] = X[:, comb].prod(1)
 
 
@@ -157,6 +157,17 @@ def test_polynomial_feature_names():
                        feature_names)
 
 
+def test_polynomial_feature_array_order():
+    X = np.arange(10).reshape(5, 2)
+
+    def is_c_contiguous(a):
+        return np.isfortran(a.T)
+
+    assert is_c_contiguous(PolynomialFeatures().fit_transform(X))
+    assert is_c_contiguous(PolynomialFeatures(order='C').fit_transform(X))
+    assert np.isfortran(PolynomialFeatures(order='F').fit_transform(X))
+
+
 @pytest.mark.parametrize(['deg', 'include_bias', 'interaction_only', 'dtype'],
                          [(1, True, False, int),
                           (2, True, False, int),