gistfile1.py

# -*- coding:utf-8 -*-
# 单核多类分类程序
# 13-5-31 modified trainflag and  testflag

# 导入模块
import liyfun as liy                                            # load the data funcs
import numpy as np                                          # numpy tool
from shogun.Features import RealFeatures, MulticlassLabels
from shogun.Kernel import Chi2Kernel, CustomKernel , GaussianKernel
from shogun.Distance import ChiSquareDistance      
from shogun.Classifier import MulticlassLibSVM  

from shogun.Evaluation import CrossValidation, CrossValidationResult
from shogun.Evaluation import ContingencyTableEvaluation, ACCURACY
from shogun.Evaluation import StratifiedCrossValidationSplitting,  MulticlassAccuracy

from shogun.ModelSelection import ModelSelectionParameters, GridSearchModelSelection,  R_EXP, R_LINEAR
from shogun.ModelSelection import ParameterCombination



def multiclass (ind):                                 
    name='trainflag.mat'
    temp=liy.load(filedir, name)
    trainflag=np.squeeze( np.array( temp[ind-1] )[0] ) - 1
    
    name='testflag.mat'        # get test_flag
    temp=liy.load(filedir, name)
    testflag=np.squeeze( np.array( temp[ind-1] )[0] ) - 1
    
    fixlen=10; wordsnum=50
    name='bhist_testnum_%s'%ind+ '_wordsnum_%s'%wordsnum +'.mat'
    #    name='newhist_fixlen_%s'%fixlen + '_testnum_%s'%ind+ '_50.mat'    
    basicdata=np.float_( np.transpose( liy.load(filedir, name) ) )          
    

    basicdata=liy.unitnorm(basicdata, 1, 'col')             # each col is a data point

    train=basicdata[:, trainflag]
    features=RealFeatures(train)
    labels=np.arange(0, 600)/100 
    labels=np.float64(labels)
    trainlabel=labels[trainflag]    
    reallabel=labels[testflag]    
    trainlabel = MulticlassLabels(trainlabel)
    #分类器
    classifier=MulticlassLibSVM()
    #分开策略
    splitting_strategy=StratifiedCrossValidationSplitting(trainlabel, subsets)
    #评价
    evaluation_criterion=MulticlassAccuracy()
    # 交叉验证类，四个输入 
    cross=CrossValidation(classifier, features, trainlabel, splitting_strategy, evaluation_criterion)
    cross.set_num_runs(1)       #重复次数    


    # 创建参数树
    root=ModelSelectionParameters()     # 创建参数根节点
    c=ModelSelectionParameters("C")     # param 1 is C
    root.append_child(c)
    c.build_values(0.0, 5.0, R_EXP, 1.0, 10.0)             # c 从 1 10 100 1000 ~10^10
    chi2_kernel=Chi2Kernel()
    
    param_chi2_kernel=ModelSelectionParameters("kernel", chi2_kernel)       # 创建核函数的参数
    chi2_kernel_width=ModelSelectionParameters("width")
    chi2_kernel_width.build_values(0.05, 30, R_LINEAR, 0.5 )
    param_chi2_kernel.append_child(chi2_kernel_width)
    
    root.append_child(param_chi2_kernel)                            # 核函数参数添加到模型里
    
    grid_search=GridSearchModelSelection(root, cross)
    
    print_state=False
    best_combination=grid_search.select_model(print_state)
    
    best_combination.apply_to_machine(classifier)
    k=classifier.get_kernel()
    w=Chi2Kernel.obtain_from_generic(k).get_width()