svm_w_usage.py

一种基于局部密度比权重设置模型的加权支持向量回归模型来单步求解多分类问题：该方法先分别对类样本中每类样本利用局部密度比权重设置模型求出每个样本的权重隶属因子,然后运用加权lib支持向量回归算法对所有样

#!/usr/bin/env python

import string
from svm import *

f = open("../heart.10", "r")
labels = []
samples = []
weights = []
line = f.readline()
max_index = 0
while line:
	elems = string.split(line)
	sample = {}
	for e in elems[1:]:
		points = string.split(e, ":")
		sample[int(points[0])] = float(points[1])
                if max_index < int(points[0]):
                    max_index = int(points[0])
                    
	labels.append(float(elems[0]))
	samples.append(sample)
        weights.append(0.001)
	line = f.readline()
f.close()

print "%d samples loaded." % (len(samples))
param = svm_parameter(svm_type = C_SVC, kernel_type = RBF, gamma=1.0/max_index)

for i in range(10):
    print weights
    
    prob = svm_problem(labels, samples, weights)
    model=svm_model(prob, param)

    for i in range(len(samples)):
        if model.predict(samples[i]) != labels[i]:
	    print ("deemphasizing %d"%i)
            weights[i] = weights[i] / 2.0
        else:
            weights[i] = weights[i] * 2.0