diff --git a/scikits/learn/src/libsvm_helper.c b/scikits/learn/src/libsvm_helper.c
index f8edee2a1f5d8d880bc197b5ca72d03ebcabb1f2..f131eaed9a7fa76f6e9be92fed6e49c614590912 100644
--- a/scikits/learn/src/libsvm_helper.c
+++ b/scikits/learn/src/libsvm_helper.c
@@ -338,7 +338,8 @@ int copy_prob_predict(char *predict, struct svm_model *model, npy_intp *predict_
if (predict_nodes == NULL)
return -1;
for(i=0; i<n; ++i) {
- svm_predict_probability(model, predict_nodes[i], &t[i*m]);
+ svm_predict_probability(model, predict_nodes[i],
+ dec_values + i*m*sizeof(double));
free(predict_nodes[i]);
}
return 0;
diff --git a/scikits/learn/svm.py b/scikits/learn/svm.py
index 76db50020a83ce97f84ec74927f0c0f8589ef201..1e117fc7a18dfb15c3288608897a1bd5c02db94a 100644
--- a/scikits/learn/svm.py
+++ b/scikits/learn/svm.py
@@ -5,7 +5,7 @@ _kernel_types = ['linear', 'poly', 'rbf', 'sigmoid', 'precomputed']
_svm_types = ['c_svc', 'nu_svc', 'one_class', 'epsilon_svr', 'nu_svr']
-class BaseSVM(object):
+class BaseLibsvm(object):
"""
Base class for classifiers that use support vector machine.
@@ -62,8 +62,8 @@ class BaseSVM(object):
if X.shape[0] != y.shape[0]: raise ValueError("Incompatible shapes")
if (self.gamma == 0): self.gamma = 1.0/X.shape[0]
- self.coef_, self.rho_, self.support_, self.nclass_, self.nSV_, self.label_, \
- self.probA_, self.probB_ = libsvm.train_wrap(X, y,
+ self.coef_, self.rho_, self.support_, self.nclass_, self.nSV_, \
+ self.label_, self.probA_, self.probB_ = libsvm.train_wrap(X, y,
self.svm, self.kernel, self.degree, self.gamma,
self.coef0, self.eps, self.C, self.nr_weight,
self.weight_label, self.weight, self.nu, self.cache_size, self.p,
@@ -100,7 +100,7 @@ class BaseSVM(object):
# Public API
# No processing should go into these classes
-class SVC(BaseSVM):
+class SVC(BaseLibsvm):
"""
Support Vector Classification
@@ -174,46 +174,16 @@ class SVC(BaseSVM):
SVR
"""
- _penalties = {'l2': 0, 'l1' : 6}
-
-
- def __init__(self, impl='c_svc', kernel='rbf', degree=3, penalty='l2',
+ def __init__(self, impl='c_svc', kernel='rbf', degree=3,
gamma=0.0, coef0=0.0, cache_size=100.0, eps=1e-3,
C=1.0, nr_weight=0, nu=0.5, p=0.1, shrinking=True,
probability=False):
- BaseSVM.__init__(self, impl, kernel, degree, gamma, coef0,
+ BaseLibsvm.__init__(self, impl, kernel, degree, gamma, coef0,
cache_size, eps, C, nr_weight, nu, p,
shrinking, probability)
- self.penalty = self._penalties[penalty]
- if self.kernel == 0:
- # this must be called after BaseSVM.__init__
- # because liblinear expects this to be ints
- self.weight_label = np.empty(0, dtype=np.int32)
-
- def fit(self, X, Y):
- if self.kernel > 0:
- return BaseSVM.fit(self, X, Y)
- X = np.asanyarray(X, dtype=np.float64, order='C')
- Y = np.asanyarray(Y, dtype=np.int32, order='C')
- self.coef_, self.label_, self.bias_ = liblinear.train_wrap(X,
- Y, self.penalty, self.eps, 1.0,
- self.C, 0,
- self.weight_label,
- self.weight)
-
- return self
- def predict(self, T):
- if self.kernel > 0:
- return BaseSVM.predict(self, T)
- T = np.asanyarray(T, dtype=np.float64, order='C')
- return liblinear.predict_wrap(T, self.coef_, self.penalty,
- self.eps, self.C,
- self.weight_label,
- self.weight, self.label_,
- 1.0)
-class SVR(BaseSVM):
+class SVR(BaseLibsvm):
"""
Support Vector Regression.
@@ -244,11 +214,11 @@ class SVR(BaseSVM):
gamma=0.0, coef0=0.0, cache_size=100.0, eps=1e-3,
C=1.0, nr_weight=0, nu=0.5, p=0.1, shrinking=True,
probability=False):
- BaseSVM.__init__(self, impl, kernel, degree, gamma, coef0,
+ BaseLibsvm.__init__(self, impl, kernel, degree, gamma, coef0,
cache_size, eps, C, nr_weight, nu, p,
shrinking, probability)
-class OneClassSVM(BaseSVM):
+class OneClassSVM(BaseLibsvm):
"""
Outlayer detection
@@ -265,6 +235,51 @@ class OneClassSVM(BaseSVM):
C=1.0, nr_weight=0, nu=0.5, p=0.1, shrinking=True,
probability=False):
impl = 'one_class'
- BaseSVM.__init__(self, impl, kernel, degree, gamma, coef0,
+ BaseLibsvm.__init__(self, impl, kernel, degree, gamma, coef0,
cache_size, eps, C, nr_weight, nu, p,
shrinking, probability)
+
+
+class LinearSVC(object):
+ """
+ Linear Support Vector Classification.
+
+
+ Parameters
+ ----------
+
+ Also accepts parameter penalty, that can have values 'l1' or 'l2'
+
+ Similar to SVC with parameter kernel='linear', but uses internally
+ liblinear rather than libsvm, so it has more flexibility in the
+ choice of penalties and loss functions and should be faster for
+ huge datasets.
+ """
+ _penalties = {'': 0}
+
+ def __init__(self, loss='l2', penalty='l2', dual=False, eps=1e-4, C=1.0):
+ self.penalty = self._penalties[penalty]
+ self.eps = eps
+ self.C = C
+
+ _penalties = {'l2': 0, 'l1' : 6}
+ _weight_label = np.empty(0, dtype=np.int)
+ _weight = np.empty(0, dtype=np.float64)
+
+ def fit(self, X, Y):
+ X = np.asanyarray(X, dtype=np.float64, order='C')
+ Y = np.asanyarray(Y, dtype=np.int, order='C')
+ self.coef_, self.label_, self.bias_ = liblinear.train_wrap(X,
+ Y, self.penalty, self.eps, 1.0,
+ self.C, 0,
+ self._weight_label,
+ self._weight)
+
+ def predict(self, T):
+ T = np.asanyarray(T, dtype=np.float64, order='C')
+ return liblinear.predict_wrap(T, self.coef_, self.penalty,
+ self.eps, self.C,
+ self._weight_label,
+ self._weight, self.label_,
+ self.bias_)
+