1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
|
#! /usr/bin/env python
# Last Change: Mon Jul 09 08:00 PM 2007 J
__doc__ = """Example of doing classification with mixture of Gaussian. Note
that this is really a toy example: we do not use testing testset nor cross
validation.
We use the famous iris database used by Sir R.A. Fisher. You can try to change
the attributes used for classification, number of components used for the
mixtures, etc..."""
import numpy as N
import pylab as P
import matplotlib as MPL
from scipy.sandbox import svm
import utils
data = utils.iris.load()
def get_data(xattr, yattr, ln):
"""Given x and y attributes, returns label, samples, label, sample, where
the first couple (label, samples) is for training, the other one for
testing.
For each class, the first nl samples are used for training, the other for testing.
"""
lxdata = {}
lydata = {}
txdata = {}
tydata = {}
llabel = {}
tlabel = {}
data = utils.iris.load()
cnames = data.keys()
lnames = {}
for i in range(len(cnames)):
lnames[cnames[i]] = i
for i in cnames:
lxdata[i] = data[i][xattr][:ln]
txdata[i] = data[i][xattr][ln:]
lydata[i] = data[i][yattr][:ln]
tydata[i] = data[i][yattr][ln:]
lxdata = N.concatenate([lxdata[i] for i in cnames])
lydata = N.concatenate([lydata[i] for i in cnames])
txdata = N.concatenate([txdata[i] for i in cnames])
tydata = N.concatenate([tydata[i] for i in cnames])
llabel = N.concatenate([lnames[i] * N.ones(ln, N.int) for i in cnames])
tlabel = N.concatenate([lnames[i] * N.ones(ln, N.int) for i in cnames])
return llabel, N.vstack((lxdata, lydata)).T, tlabel, \
N.vstack((txdata, tydata)).T, cnames, lnames
#--------------------
# Data pre processing
#--------------------
# we use 25 samples of each class (eg half of the dataset), for
# learning, and the other half for testing. We use sepal width and petal width
# only as features.
ln = 25
llabel, ldata, tlabel, tdata, cnames, lnames = get_data('sepal width', 'petal width', ln)
training = svm.LibSvmClassificationDataSet(llabel, ldata)
testing = svm.LibSvmTestDataSet(tdata)
def train_svm(cost, gamma, fold = 5):
"""Train a SVM for given cost and gamma."""
kernel = svm.RBFKernel(gamma = gamma)
model = svm.LibSvmCClassificationModel(kernel, cost = cost)
cv = model.cross_validate(training, fold)
return cv
c_range = N.exp(N.log(2.) * N.arange(-5, 15))
g_range = N.exp(N.log(2.) * N.arange(-15, 3))
# Train the svm on a log distributed grid
gr = N.meshgrid(c_range, g_range)
c = gr[0].flatten()
g = gr[1].flatten()
cf = N.hstack((c, g))
cv = N.empty(c.size)
for i in range(cv.size):
cv[i] = train_svm(c[i], g[i])
v = P.contour(N.log2(gr[0]), N.log2(gr[1]), cv.reshape(g_range.size, c_range.size), 12)
P.clabel(v, inline = 1, fontsize = 10)
P.show()
P.legend()
|
