diff --git a/scikits/learn/linear_model/sparse/logistic.py b/scikits/learn/linear_model/sparse/logistic.py
index dbccc24ad1ecdbe3c7bfcb7872c5d6baede4e25f..73e76c76c3aabaa7be853cd8e661d3f3ee1a825d 100644
--- a/scikits/learn/linear_model/sparse/logistic.py
+++ b/scikits/learn/linear_model/sparse/logistic.py
@@ -6,7 +6,6 @@ designed to handle efficiently data in sparse matrix format.
"""
import numpy as np
-from scipy import sparse
from ...base import ClassifierMixin
from ...svm.sparse.base import SparseBaseLibLinear
@@ -71,17 +70,18 @@ class LogisticRegression(SparseBaseLibLinear, ClassifierMixin):
dual=dual, loss='lr', eps=eps, C=C,
fit_intercept=fit_intercept)
- def predict_proba(self, T):
+ def predict_proba(self, X):
"""
Probability estimates.
The returned estimates for all classes are ordered by the
label of classes.
"""
- T = sparse.csr_matrix(T)
- T.data = np.asanyarray(T.data, dtype=np.float64, order='C')
- probas = csr_predict_prob(T.shape[1], T.data, T.indices,
- T.indptr, self.raw_coef_,
+ import scipy.sparse
+ X = scipy.sparse.csr_matrix(X)
+ X.data = np.asanyarray(X.data, dtype=np.float64, order='C')
+ probas = csr_predict_prob(X.shape[1], X.data, X.indices,
+ X.indptr, self.raw_coef_,
self._get_solver_type(),
self.eps, self.C,
self.class_weight_label,
diff --git a/scikits/learn/svm/__init__.py b/scikits/learn/svm/__init__.py
index 9f121c3ad7230580e95f42003627559db5245a2d..a2ac5f3fefcd838dc23c13560630fa285f49cdae 100644
--- a/scikits/learn/svm/__init__.py
+++ b/scikits/learn/svm/__init__.py
@@ -1,8 +1,13 @@
"""
-Module that implements Support Vector Machnine related algorithms.
+Support Vector Machnine algorithms.
See http://scikit-learn.sourceforge.net/modules/svm.html for complete
documentation.
+
+Author: Fabian Pedregosa <fabian.pedregosa@inria.fr> with help from
+ the scikit-learn community. LibSVM and LibLinear are copyright
+ of their respective owners.
+License: New BSD, (C) INRIA 2010
"""
from .libsvm import SVC, NuSVC, SVR, NuSVR, OneClassSVM
diff --git a/scikits/learn/svm/base.py b/scikits/learn/svm/base.py
index fe168ec19c7a0044d6668eb67f3d60600c329ea5..be5bfa8af7033572052fafe1a52ee42e2908049c 100644
--- a/scikits/learn/svm/base.py
+++ b/scikits/learn/svm/base.py
@@ -5,7 +5,6 @@ from ._libsvm import libsvm_train, libsvm_predict, libsvm_predict_proba, \
from . import _liblinear
from ..base import BaseEstimator, RegressorMixin, ClassifierMixin
-
def _get_class_weight(class_weight, y):
"""
Estimate class weights for unbalanced datasets.
@@ -73,7 +72,7 @@ class BaseLibSVM(BaseEstimator):
return kernel_type, _X
- def fit(self, X, y, class_weight={}, sample_weight=[]):
+ def fit(self, X, y, class_weight={}, sample_weight=[], **params):
"""
Fit the SVM model according to the given training data and
parameters.
@@ -81,8 +80,8 @@ class BaseLibSVM(BaseEstimator):
Parameters
----------
X : array-like, shape = [n_samples, n_features]
- Training vector, where n_samples is the number of samples and
- n_features is the number of features.
+ Training vectors, where n_samples is the number of samples
+ and n_features is the number of features.
y : array-like, shape = [n_samples]
Target values (integers in classification, real numbers in
@@ -104,6 +103,8 @@ class BaseLibSVM(BaseEstimator):
self : object
Returns self.
"""
+ self._set_params(**params)
+
X = np.asanyarray(X, dtype=np.float64, order='C')
y = np.asanyarray(y, dtype=np.float64, order='C')
sample_weight = np.asanyarray(sample_weight, dtype=np.float64,
@@ -160,7 +161,7 @@ class BaseLibSVM(BaseEstimator):
Returns
-------
- C : array, shape = [nsample]
+ C : array, shape = [n_samples]
"""
T = np.atleast_2d(np.asanyarray(T, dtype=np.float64, order='C'))
kernel_type, T = self._get_kernel(T)
@@ -338,23 +339,15 @@ class BaseLibLinear(BaseEstimator):
def predict(self, X):
"""
- This function does classification or regression on an array of
- test vectors X.
-
- For a classification model, the predicted class for each
- sample in X is returned. For a regression model, the function
- value of X calculated is returned.
-
- For a one-class model, +1 or -1 is returned.
+ Predict target values of X according to the fitted model.
Parameters
----------
X : array-like, shape = [n_samples, n_features]
-
Returns
-------
- C : array, shape = [nsample]
+ C : array, shape = [n_samples]
"""
X = np.asanyarray(X, dtype=np.float64, order='C')
self._check_n_features(X)
diff --git a/scikits/learn/svm/liblinear.py b/scikits/learn/svm/liblinear.py
index 3beff32a4a85674d1d9ecacd2d2330537e34c7a6..e02be399d4f7ef7625ca358e16d6e1ba37c8c59d 100644
--- a/scikits/learn/svm/liblinear.py
+++ b/scikits/learn/svm/liblinear.py
@@ -4,7 +4,7 @@ from .base import BaseLibLinear
class LinearSVC(BaseLibLinear, ClassifierMixin):
- """Linear Support Vector Classification
+ """Linear Support Vector Classification.
Similar to SVC with parameter kernel='linear', but uses internally
liblinear rather than libsvm, so it has more flexibility in the
diff --git a/scikits/learn/svm/libsvm.py b/scikits/learn/svm/libsvm.py
index fe032492099ec9a75cb609b3f81f570ed0bd8aac..5a75fa88dcf1354ee2806ac6cb2d2d90021c41ec 100644
--- a/scikits/learn/svm/libsvm.py
+++ b/scikits/learn/svm/libsvm.py
@@ -4,9 +4,7 @@ from .base import BaseLibSVM
class SVC(BaseLibSVM, ClassifierMixin):
- """
- C-Support Vector Classification.
-
+ """C-Support Vector Classification.
Parameters
----------
@@ -94,9 +92,7 @@ class SVC(BaseLibSVM, ClassifierMixin):
class NuSVC(BaseLibSVM, ClassifierMixin):
- """
- Nu-Support Vector Classification.
-
+ """Nu-Support Vector Classification.
Parameters
----------
@@ -200,9 +196,7 @@ class NuSVC(BaseLibSVM, ClassifierMixin):
class SVR(BaseLibSVM, RegressorMixin):
- """
- Support Vector Regression.
-
+ """Support Vector Regression.
Parameters
----------
@@ -299,11 +293,11 @@ class SVR(BaseLibSVM, RegressorMixin):
class NuSVR(BaseLibSVM, RegressorMixin):
- """
- Nu Support Vector Regression. Similar to NuSVC, for regression,
- uses a paramter nu to control the number of support
- vectors. However, unlike NuSVC, where nu replaces with C, here nu
- replaces with the parameter p of SVR.
+ """Nu Support Vector Regression.
+
+ Similar to NuSVC, for regression, uses a paramter nu to control
+ the number of support vectors. However, unlike NuSVC, where nu
+ replaces with C, here nu replaces with the parameter p of SVR.
Parameters
----------
@@ -398,7 +392,7 @@ class NuSVR(BaseLibSVM, RegressorMixin):
class OneClassSVM(BaseLibSVM):
- """Unsupervised outliers detection
+ """Unsupervised Outliers Detection.
Estimate the support of a high-dimensional distribution.
diff --git a/scikits/learn/svm/sparse/__init__.py b/scikits/learn/svm/sparse/__init__.py
index 200bfeb0404b3782ce762e8aca3420f5fc133f6e..639fade57fda0b573e720e853a5195f8dd6b8744 100644
--- a/scikits/learn/svm/sparse/__init__.py
+++ b/scikits/learn/svm/sparse/__init__.py
@@ -1,5 +1,21 @@
"""
-Sparse version of algorithms found in the svm module
+Support Vector Machine algorithms for sparse matrices.
+
+This module should have the same API as scikits.learn.svm, except that
+matrices are expected to be in some sparse format supported by
+scipy.sparse.
+
+See http://scikit-learn.sourceforge.net/modules/svm.html
+
+Notes
+-----
+Some fields, like dual_coef_ are not sparse matrices strictly speaking.
+However, they are converted to a sparse matrix for consistency and
+efficiency when multiplying to other sparse matrices.
+
+Author: Fabian Pedregosa <fabian.pedregosa@inria.fr> with help from
+ the scikit-learn community.
+License: New BSD, (C) INRIA 2010
"""
from .libsvm import SVC, NuSVC, SVR, NuSVR, OneClassSVM
diff --git a/scikits/learn/svm/sparse/base.py b/scikits/learn/svm/sparse/base.py
index ba3a4019c2a3ffcc83584107432f4dde4cd6f7c4..7e5a1e20e5fd2a076946f1d7adbb9e3b668141a5 100644
--- a/scikits/learn/svm/sparse/base.py
+++ b/scikits/learn/svm/sparse/base.py
@@ -1,22 +1,4 @@
-"""
-Support Vector Machine algorithms for sparse matrices.
-
-Warning: this module is a work in progress. It is not tested and surely
-contains bugs.
-
-Notes
------
-
-Some fields, like dual_coef_ are not sparse matrices strictly speaking.
-However, they are converted to a sparse matrix for consistency and
-efficiency when multiplying to other sparse matrices.
-
-Author: Fabian Pedregosa <fabian.pedregosa@inria.fr>
-License: New BSD
-"""
-
import numpy as np
-from scipy import sparse
from ...base import ClassifierMixin
from ..base import BaseLibSVM, BaseLibLinear, _get_class_weight
@@ -67,13 +49,46 @@ class SparseBaseLibSVM(BaseLibSVM):
self.n_support = np.empty(0, dtype=np.int32, order='C')
- def fit(self, X, y, class_weight={}, sample_weight=[]):
+ def fit(self, X, y, class_weight={}, sample_weight=[], **params):
"""
- X is expected to be a sparse matrix. For maximum effiency, use a
- sparse matrix in csr format (scipy.sparse.csr_matrix)
+ Fit the SVM model according to the given training data and
+ parameters.
+
+ Parameters
+ ----------
+ X : sparse matrix, shape = [n_samples, n_features]
+ Training vectors, where n_samples is the number of samples and
+ n_features is the number of features.
+
+ y : array-like, shape = [n_samples]
+ Target values (integers in classification, real numbers in
+ regression)
+
+ class_weight : dict | 'auto', optional
+ Weights associated with classes in the form
+ {class_label : weight}. If not given, all classes are
+ supposed to have weight one.
+
+ The 'auto' mode uses the values of y to automatically adjust
+ weights inversely proportional to class frequencies.
+
+ sample_weight : array-like, shape = [n_samples], optional
+ Weights applied to individual samples (1. for unweighted).
+
+ Returns
+ -------
+ self : object
+ Returns an instance of self.
+
+ Notes
+ -----
+ For maximum effiency, use a sparse matrix in csr format
+ (scipy.sparse.csr_matrix)
"""
+ self._set_params(**params)
- X = sparse.csr_matrix(X)
+ import scipy.sparse
+ X = scipy.sparse.csr_matrix(X)
X.data = np.asanyarray(X.data, dtype=np.float64, order='C')
y = np.asanyarray(y, dtype=np.float64, order='C')
sample_weight = np.asanyarray(sample_weight, dtype=np.float64,
@@ -108,12 +123,12 @@ class SparseBaseLibSVM(BaseLibSVM):
# this will fail if n_SV is zero. This is a limitation
# in scipy.sparse, which does not permit empty matrices
- self.support_vectors_ = sparse.csr_matrix((self._support_data,
+ self.support_vectors_ = scipy.sparse.csr_matrix((self._support_data,
self._support_indices,
self._support_indptr),
(n_SV, X.shape[1]) )
- self.dual_coef_ = sparse.csr_matrix((self._dual_coef_data,
+ self.dual_coef_ = scipy.sparse.csr_matrix((self._dual_coef_data,
dual_coef_indices,
dual_coef_indptr),
(n_class, n_SV)
@@ -138,9 +153,10 @@ class SparseBaseLibSVM(BaseLibSVM):
Returns
-------
- C : array, shape = [nsample]
+ C : array, shape = [n_samples]
"""
- T = sparse.csr_matrix(T)
+ import scipy.sparse
+ T = scipy.sparse.csr_matrix(T)
T.data = np.asanyarray(T.data, dtype=np.float64, order='C')
kernel_type = self._kernel_types.index(self.kernel)
@@ -161,16 +177,25 @@ class SparseBaseLibLinear(BaseLibLinear):
def fit(self, X, y, class_weight={}, **params):
"""
+ Fit the model using X, y as training data.
+
Parameters
----------
- X : array-like, shape = [n_samples, n_features]
+ X : sparse matrix, shape = [n_samples, n_features]
Training vector, where n_samples in the number of samples and
n_features is the number of features.
y : array, shape = [n_samples]
Target vector relative to X
+
+ Returns
+ -------
+ self : object
+ Returns an instance of self.
"""
self._set_params(**params)
- X = sparse.csr_matrix(X)
+
+ import scipy.sparse
+ X = scipy.sparse.csr_matrix(X)
X.data = np.asanyarray(X.data, dtype=np.float64, order='C')
y = np.asanyarray(y, dtype=np.int32, order='C')
@@ -186,7 +211,19 @@ class SparseBaseLibLinear(BaseLibLinear):
return self
def predict(self, X):
- X = sparse.csr_matrix(X)
+ """
+ Predict target values of X according to the fitted model.
+
+ Parameters
+ ----------
+ X : sparse matrix, shape = [n_samples, n_features]
+
+ Returns
+ -------
+ C : array, shape = [n_samples]
+ """
+ import scipy.sparse
+ X = scipy.sparse.csr_matrix(X)
self._check_n_features(X)
X.data = np.asanyarray(X.data, dtype=np.float64, order='C')
diff --git a/scikits/learn/svm/sparse/libsvm.py b/scikits/learn/svm/sparse/libsvm.py
index fb6b05c54bcd3db64ff39c0beca1aaba9deb69f5..d338a85452a2ba27e0af99dfc384b6f93b58719a 100644
--- a/scikits/learn/svm/sparse/libsvm.py
+++ b/scikits/learn/svm/sparse/libsvm.py
@@ -4,8 +4,12 @@ from .base import SparseBaseLibSVM
class SVC(SparseBaseLibSVM, ClassifierMixin):
- """SVC for sparse matrices (csr)
+ """SVC for sparse matrices (csr).
+ See :class:`scikits.learn.svm.SVC` for a complete list of parameters
+
+ Notes
+ -----
For best results, this accepts a matrix in csr format
(scipy.sparse.csr), but should be able to convert from any array-like
object (including other sparse representations).
@@ -22,8 +26,12 @@ class SVC(SparseBaseLibSVM, ClassifierMixin):
class NuSVC (SparseBaseLibSVM, ClassifierMixin):
- """NuSVC for sparse matrices (csr)
+ """NuSVC for sparse matrices (csr).
+
+ See :class:`scikits.learn.svm.NuSVC` for a complete list of parameters
+ Notes
+ -----
For best results, this accepts a matrix in csr format
(scipy.sparse.csr), but should be able to convert from any array-like
object (including other sparse representations).
@@ -44,6 +52,10 @@ class NuSVC (SparseBaseLibSVM, ClassifierMixin):
class SVR (SparseBaseLibSVM, RegressorMixin):
"""SVR for sparse matrices (csr)
+ See :class:`scikits.learn.svm.SVR` for a complete list of parameters
+
+ Notes
+ -----
For best results, this accepts a matrix in csr format
(scipy.sparse.csr), but should be able to convert from any array-like
object (including other sparse representations).
@@ -65,6 +77,10 @@ class SVR (SparseBaseLibSVM, RegressorMixin):
class NuSVR (SparseBaseLibSVM, RegressorMixin):
"""NuSVR for sparse matrices (csr)
+ See :class:`scikits.learn.svm.NuSVC` for a complete list of parameters
+
+ Notes
+ -----
For best results, this accepts a matrix in csr format
(scipy.sparse.csr), but should be able to convert from any array-like
object (including other sparse representations).
@@ -83,6 +99,10 @@ class NuSVR (SparseBaseLibSVM, RegressorMixin):
class OneClassSVM (SparseBaseLibSVM):
"""NuSVR for sparse matrices (csr)
+ See :class:`scikits.learn.svm.NuSVC` for a complete list of parameters
+
+ Notes
+ -----
For best results, this accepts a matrix in csr format
(scipy.sparse.csr), but should be able to convert from any array-like
object (including other sparse representations).