8000 Rename _apply_transform to _transform_selected and make it a function · scikit-learn/scikit-learn@017afab · GitHub
[go: up one dir, main page]
Skip to content

Commit 017afab

Browse files
mblondelmblondel
committed
Rename _apply_transform to _transform_selected and make it a function
rather than a method.
1 parent 6e2aee2 commit 017afab

File tree

3 files changed

+84
-30
lines changed

3 files changed

+84
-30
lines changed

‎doc/modules/preprocessing.rst

Lines changed: 4 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -333,15 +333,16 @@ Continuing the example above::
333333

334334
>>> enc = preprocessing.OneHotEncoder()
335335
>>> enc.fit([[0, 0, 3], [1, 1, 0], [0, 2, 1], [1, 0, 2]])
336-
OneHotEncoder(dtype=<type 'float'>, n_values='auto')
336+
OneHotEncoder(categorical_features='all', dtype=<type 'float'>,
337+
n_values='auto')
337338
>>> enc.transform([[0, 1, 3]]).toarray()
338339
array([[ 1., 0., 0., 1., 0., 0., 0., 0., 1.]])
339340

340341
By default, how many values each feature can take is inferred automatically from the dataset.
341-
It is possible to specify this explicitly using the parameter ``n_values``.
342+
It is possible to specify this explicitly using the parameter ``n_values``.
342343
There are two genders, three possible continents and four web browsers in our
343344
dataset.
344-
Then we fit the estimator, and transform a data point.
345+
Then we fit the estimator, and transform a data point.
345346
In the result, the first two numbers encode the gender, the next set of three
346347
numbers the continent and the last four the web browser.
347348

‎sklearn/preprocessing.py

Lines changed: 52 additions & 25 deletions
Original file line numberDiff line numberDiff line change
@@ -15,6 +15,7 @@
1515
from .utils import check_arrays
1616
from .utils import array2d
1717
from .utils import atleast2d_or_csr
18+
from .utils import atleast2d_or_csc
1819
from .utils import safe_asarray
1920
from .utils import warn_if_not_float
2021
from .utils.fixes import unique
@@ -629,6 +630,53 @@ def transform(self, X, y=None, copy=None):
629630
return binarize(X, threshold=self.threshold, copy=copy)
630631

631632

633+
def _transform_selected(X, transform, selected):
634+
"""Apply a transform function to portion of selected features
635+
636+
Parameters
637+
----------
638+
X : array-like or sparse matrix, shape=(n_samples, n_features)
639+
Dense array or sparse matrix.
640+
641+
transform : callable
642+
A callable transform(X) -> X_transformed
643+
644+
selected: "all" or array of indices or mask
645+
Specify what features to apply the transform to.
646+
647+
Returns
648+
-------
649+
X : array or sparse matrix, shape=(n_samples, n_features_new)
650+
"""
651+
if len(selected) == 0:
652+
return X
653+
elif selected == "all":
654+
return transform(X)
655+
else:
656+
X = atleast2d_or_csc(X)
657+
n_features = X.shape[1]
658+
ind = np.arange(n_features)
659+
sel = np.zeros(n_features, dtype=bool)
660+
sel[np.array(selected)] = True
661+
not_sel = np.logical_not(sel)
662+
n_selected = np.sum(sel)
663+
664+
if n_selected == 0:
665+
# No features selected.
666+
return X
667+
elif n_selected == n_features:
668+
# All features selected.
669+
return transform(X)
670+
else:
671+
X_sel = transform(X[:, ind[sel]])
672+
X_not_sel = X[:, ind[not_sel]]
673+
674+
if sp.issparse(X_sel) or sp.issparse(X_not_sel):
675+
return sp.hstack((X_sel, X_not_sel))
676+
else:
677+
return np.hstack((X_sel, X_not_sel))
678+
679+
632680
class OneHotEncoder(BaseEstimator, TransformerMixin):
633681
"""Encode categorical integer features using a one-hot aka one-of-K scheme.
634682
@@ -722,29 +770,6 @@ def fit(self, X, y=None):
722770
self.fit_transform(X)
723771
return self
724772

725-
def _apply_transform(self, X, transform):
726-
if self.categorical_features == "all":
727-
return transform(X)
728-
else:
729-
X = check_arrays(X, sparse_format='dense')[0]
730-
n_features = X.shape[1]
731-
ind = np.arange(n_features)
732-
categorical = np.zeros(n_features, dtype=bool)
733-
categorical[np.array(self.categorical_features)] = True
734-
not_categorical = np.logical_not(categorical)
735-
n_categorical = np.sum(categorical)
736-
737-
if n_categorical == 0:
738-
# No categorical variables.
739-
return X
740-
elif n_categorical == n_features:
741-
# All categorical variables.
742-
return transform(X)
743-
else:
744-
X_cat = transform(X[:, categorical])
745-
X_not_cat = X[:, not_categorical]
746-
return sp.hstack((X_cat, X_not_cat))
747-
748773
def _fit_transform(self, X):
749774
"""Asssumes X contains only categorical features."""
750775
X = check_arrays(X, sparse_format='dense', dtype=np.int)[0]
@@ -793,7 +818,8 @@ def fit_transform(self, X, y=None):
793818
Equivalent to self.fit(X).transform(X), but more convenient and more
794819
efficient. See fit for the parameters, transform for the return value.
795820
"""
796-
return self._apply_transform(X, self._fit_transform)
821+
return _transform_selected(X, self._fit_transform,
822+
self.categorical_features)
797823

798824
def _transform(self, X):
799825
"""Asssumes X contains only categorical features."""
@@ -836,7 +862,8 @@ def transform(self, X):
836862
X_out : sparse matrix, dtype=int
837863
Transformed input.
838864
"""
839-
return self._apply_transform(X, self._transform)
865+
return _transform_selected(X, self._transform,
866+
self.categorical_features)
840867

841868

842869
class LabelEncoder(BaseEstimator, TransformerMixin):

‎sklearn/tests/test_preprocessing.py

Lines changed: 28 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -15,6 +15,7 @@
1515
from sklearn.preprocessing import Binarizer
1616
from sklearn.preprocessing import KernelCenterer
1717
from sklearn.preprocessing import LabelBinarizer
18+
from sklearn.preprocessing import _transform_selected
1819
from sklearn.preprocessing import OneHotEncoder
1920
from sklearn.preprocessing import Lab 10000 elEncoder
2021
from sklearn.preprocessing import Normalizer
@@ -599,6 +600,27 @@ def test_one_hot_encoder():
599600
assert_raises(ValueError, enc.transform, [[0], [-1]])
600601

601602

603+
def _check_transform_selected(X, Xexpected, sel):
604+
for M in (X, sp.csr_matrix(X)):
605+
Xtr = _transform_selected(M, Binarizer().transform, sel)
606+
assert_array_equal(toarray(Xtr), Xexpected)
607+
608+
609+
def test_transform_selected():
610+
X = [[3, 2, 1], [0, 1, 1]]
611+
612+
Xexpected = [[1, 2, 1], [0, 1, 1]]
613+
_check_transform_selected(X, Xexpected, [0])
614+
_check_transform_selected(X, Xexpected, [True, False, False])
615+
616+
Xexpected = [[1, 1, 1], [0, 1, 1]]
617+
_check_transform_selected(X, Xexpected, [0, 1, 2])
618+
_check_transform_selected(X, Xexpected, [True, True, True])
619+
620+
_check_transform_selected(X, X, [])
621+
_check_transform_selected(X, X, [False, False, False])
622+
623+
602624
def _run_one_hot(X, X2, cat):
603625
enc = OneHotEncoder(categorical_features=cat)
604626
Xtr = enc.fit_transform(X)
@@ -608,18 +630,22 @@ def _run_one_hot(X, X2, cat):
608630

609631
def _check_one_hot(X, X2, cat, n_features):
610632
ind = np.where(cat)[0]
633+
# With mask
611634
A, B = _run_one_hot(X, X2, cat)
635+
# With indices
612636
C, D = _run_one_hot(X, X2, ind)
637+
# Check shape
613638
assert_equal(A.shape, (2, n_features))
614639
assert_equal(B.shape, (1, n_features))
615640
assert_equal(C.shape, (2, n_features))
616641
assert_equal(D.shape, (1, n_features))
642+
# Check that mask and indices give the same results
617643
assert_array_equal(toarray(A), toarray(C))
618644
assert_array_equal(toarray(B), toarray(D))
619645

620646
def test_one_hot_encoder_categorical_features():
621-
X = [[3, 2, 1], [0, 1, 1]]
622-
X2 = [[1, 1, 1]]
647+
X = np.array([[3, 2, 1], [0, 1, 1]])
648+
X2 = np.array([[1, 1, 1]])
623649

624650
cat = [True, False, False]
625651
_check_one_hot(X, X2, cat, 4)

0 commit comments

Comments
 (0)
0