From b806b12f41220ebbe1a1097a0276a899d4120628 Mon Sep 17 00:00:00 2001
From: Fabian Pedregosa <fabian.pedregosa@inria.fr>
Date: Mon, 15 Mar 2010 11:01:12 +0000
Subject: [PATCH] SVM refactoring
From: Fabian Pedregosa <fabian.pedregosa@inria.fr>
git-svn-id: https://scikit-learn.svn.sourceforge.net/svnroot/scikit-learn/trunk@534 22fbfee3-77ab-4535-9bad-27d1bd3bc7d8
---
scikits/learn/svm.py | 43 ++++++++++++++-----------------------------
1 file changed, 14 insertions(+), 29 deletions(-)
diff --git a/scikits/learn/svm.py b/scikits/learn/svm.py
index c470219307..c3b19b12f1 100644
--- a/scikits/learn/svm.py
+++ b/scikits/learn/svm.py
@@ -21,8 +21,7 @@ class BaseSVM(object):
"""
def __init__(self, svm, kernel, degree, gamma, coef0, cache_size,
- eps, C, nr_weight, nu, p, shrinking, probability,
- scale):
+ eps, C, nr_weight, nu, p, shrinking, probability):
self.svm = _svm_types.index(svm)
self.kernel = _kernel_types.index(kernel)
self.degree = degree
@@ -36,7 +35,6 @@ class BaseSVM(object):
self.p = p
self.shrinking = shrinking
self.probability = probability
- self.scale = scale
def fit(self, X, y):
"""
@@ -45,11 +43,6 @@ class BaseSVM(object):
X = np.asanyarray(X, dtype=np.float, order='C')
y = np.asanyarray(y, dtype=np.float, order='C')
- if self.scale:
- self.mean = X.mean(0)
- self.std = X.std(0)
- X = (X - self.mean) / self.std
-
# check dimensions
if X.shape[0] != y.shape[0]: raise ValueError("Incompatible shapes")
@@ -63,7 +56,6 @@ class BaseSVM(object):
def predict(self, T):
T = np.asanyarray(T, dtype=np.float, order='C')
- if self.scale: T = (T - self.mean) / self.std
return libsvm.predict_from_model_wrap(T, self.support_,
self.coef_, self.rho_, self.svm,
self.kernel, self.degree, self.gamma,
@@ -129,32 +121,24 @@ class SVC(BaseSVM):
Y : array, shape = [nsamples]
Target vector relative to X
- Optional Parameters
- -------------------
- impl : string
+ impl : string, optional
SVM implementation to choose from. This refers to different
formulations of the SVM optimization problem.
Can be one of 'c_svc', 'nu_svc'. By default 'c_svc' will be chosen.
- scale : boolean
- Scale data before fitting the model. Caution: if activated,
- estimated parameters will also be scaled.
- Default is True.
-
- nu: float
+ nu: float, optional
An upper bound on the fraction of training errors and a lower
bound of the fraction of support vectors. Should be in the
interval (0, 1].
By default 0.5 will be taken.
Only available is impl is set to 'nu_svc'
-
- kernel : string
+ kernel : string, optional
Specifies the kernel type to be used in the algorithm.
one of 'linear', 'poly', 'rbf', 'sigmoid', 'precomputed'.
If none is given 'rbf' will be used.
- degree : int
+ degree : int, optional
degree of kernel function
is significant only in POLY, RBF, SIGMOID
@@ -182,19 +166,20 @@ class SVC(BaseSVM):
>>> print clf.predict([[-0.8, -1]])
[ 1.]
- References
- ----------
+ See also
+ --------
http://scikit-learn.sourceforge.net/doc/modules/svm.html
+
http://www.csie.ntu.edu.tw/~cjlin/papers/libsvm.pdf
"""
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=1,
- probability=0, scale=True):
+ probability=0):
BaseSVM.__init__(self, impl, kernel, degree, gamma, coef0,
cache_size, eps, C, nr_weight, nu, p,
- shrinking, probability, scale)
+ shrinking, probability)
class SVR(BaseSVM):
@@ -212,10 +197,10 @@ class SVR(BaseSVM):
def __init__(self, svm='epsilon_svr', 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=1,
- probability=0, scale=True):
+ probability=0):
BaseSVM.__init__(self, svm, kernel, degree, gamma, coef0,
cache_size, eps, C, nr_weight, nu, p,
- shrinking, probability, scale)
+ shrinking, probability)
class OneClassSVM(BaseSVM):
"""
@@ -224,8 +209,8 @@ class OneClassSVM(BaseSVM):
def __init__(self, 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=1,
- probability=0, scale=True):
+ probability=0):
svm = 'one_class'
BaseSVM.__init__(self, svm, kernel, degree, gamma, coef0,
cache_size, eps, C, nr_weight, nu, p,
- shrinking, probability, scale)
+ shrinking, probability)
--
GitLab