libsvm是台湾大学林智仁教授等研究人员开发的一个用于支持向量机分类的c++开源库,支持多种语言接口,并且可进行多分类与回归问题的建模
在数据预处理阶段,需要将数据集转化为libsvm要求的格式
通常的原始数据集格式为
value1 value2 value3...... label
而libsvm要求格式为
label index1:value1 index2:value2 ........
实现上述转换,简单的方法是利用 FormatDataLibsvm.xls ,缺点是网上一般没有免费的资源可下载
而用python实现也不失为一种易行的方法
"""
original data form:
value1 value2 value3 value4 ......label
target data form:
label index1:value1 index2:value2 ......
"""
import numpy as np
def readFromTxt(filename):
o_data = np.loadtxt(filename)
lines = o_data.shape[0]
columns = o_data.shape[1]
new_data = np.zeros([lines, 2*columns-1], dtype=list)
i = 0
while i < lines:
j = 1
new_data[i][0] = o_data[i][columns-1]
while j < columns:
new_data[i][2*j-1] = j
new_data[i][2*j] = o_data[i][j-1]
j = j+1
i = i+1
return new_data,lines,2*columns-1
def main():
# python 2.x needs raw_input()
filename = input("Enter the file name:")
new_data,newlines,newcolumns = readFromTxt(filename)
newfilename = input("Enter the new file name:")
f = open(newfilename,"w+")
i = 0
while i < newlines:
# if classification --> label shall be int
f.write(str(int(new_data[i][0])))
# if regression --> label shall be real
# f.write(str(new_data[i][0]))
f.write(" ")
j = 1
while j < newcolumns:
f.write(str(new_data[i][j]))
if j % 2 == 1:
f.write(":")
else:
f.write(" ")
j = j+1
f.write("\n")
i = i+1
f.close()
main()
FormatDataLibsvm.xls