ls_svm2.asv

支持向量机学习中用到的用Matlab编写的工具箱。

% 支持向量机最小二乘算法(LS-SVM)--可用于分类和回归？？？
% function [ output_args ] = SvmLs( input_args )

clc
clear
close all

m = 4;         % 样本维数
TrainSample = [1+rand(m,5),2+rand(m,5),3+rand(m,5),4+rand(m,5)];
TrainTarget = [ones(1,5),2*ones(1,5),3*ones(1,5),4*ones(1,5)];

TestSample = [1+rand(m,5),2+rand(m,5),3+rand(m,5),4+rand(m,5)];
TestTarget = [ones(1,5),2*ones(1,5),3*ones(1,5),4*ones(1,5)];

% 这里 X 每一行是一个数据，Y 是列向量
X = TrainSample';
Y = TrainTarget';
Xt = TestSample';
Yt = TestTarget';

%===========================================================================
% 参数初始化

type = 'c';
gam = 1;
sig2 = 1;
kernel = 'RBF_kernel';
preprocess = 'preprocess';
codefct = 'code_MOC';      % 将“多类”转换成“两类”的编码方案

%===========================================================================

[Yc,codebook,old_codebook] = code(Y, codefct)

% Encode and decode a multi-class classification task into multiple binary classifiers
%
%     1. For encoding:
%
% >> [Yc, codebook, old_codebook] = code(Y, codefct)
% >> [Yc, codebook, old_codebook] = code(Y, codefct, codefct_args)
% >> Yc = code(Y, given_codebook)
% 
%       Outputs    
%         Yc               : N x nbits encoded output classifier
%         codebook(*)      : nbits*nc matrix representing the used encoding
%         old_codebook(*)  : d*nc matrix representing the original encoding
%       Inputs    
%         Y                : N x d matrix representing the original classifier
%         codefct(*)       : Function to generate a new codebook (e.g. code_MOC)
%         codefct_args(*)  : Extra arguments for codefct
%         given_codebook(*): nbits*nc matrix representing the encoding to use
%
% Different encoding schemes are available:
% 
%     1. Minimum Output Coding (code_MOC) 
%     2. Error Correcting Output Code (code_ECOC)
%       This coding scheme uses redundant bits. 
%     3. One versus All Coding (code_OneVsAll)
%     4. One Versus One Coding (code_OneVsOne) 

%===========================================================================
% 寻找最优参数
%[gam, sig2, cost] = tunelssvm({X,Y,type,gam,sig2,kernel,preprocess})
%[model, cost] = tunelssvm(model)

% Tune the hyperparameters of the model with respect to the given performance measure
%
% >> [gam, sig2, cost] = tunelssvm({X,Y,type,igam,isig2,kernel,preprocess})
% >> [gam, sig2, cost] = tunelssvm({X,Y,type,igam,isig2,kernel,preprocess}, StartingValues)
% >> [gam, sig2, cost] = tunelssvm({X,Y,type,igam,isig2,kernel,preprocess},...
%                                          StartingValues, optfun, optargs)
% >> [gam, sig2, cost] = tunelssvm({X,Y,type,igam,isig2,kernel,preprocess},...
%                                          StartingValues, optfun, optargs, costfun, costargs)
%
%      Outputs    
%        gam     : Optimal regularization parameter
%        sig2    : Optimal kernel parameter(s)
%        cost(*) : Estimated cost of the optimal hyperparameters
%      Inputs    
%        X       : N x d matrix with the inputs of the training data
%        Y       : N x 1 vector with the outputs of the training data
%        type    : 'function estimation' ('f') or 'classifier' ('c')
%        igam    : Starting value of the regularization parameter
%        isig2   : Starting value of the kernel parameter(s) (bandwidth in the case of the 'RBF_kernel')
%        kernel(*) : Kernel type (by default 'RBF_kernel')
%        preprocess(*) : 'preprocess'(*) or 'original'
%        StartingValues(*) : Starting values of the optimization routine (or '[]')
%        optfun(*) : Optimization function (by default 'gridsearch')
%        optargs(*) : Cell with extra optimization function arguments
%        costfun(*) : Function estimating the cost-criterion (by default 'crossvalidate')
%        costargs(*) : Cell with extra cost function arguments

%===========================================================================
% 训练

[alpha, b] = trainlssvm({X,Yc,type,gam,sig2,kernel,preprocess})

% Train the support values and the bias term of an LS-SVM for classification or function approximation
% 
% >> model = trainlssvm(model)
% >> model = trainlssvm(model, X, Y)
%
%       Outputs    
%         model          : Trained object oriented representation of the LS-SVM model
%       Inputs    
%         model          : Object oriented representation of the LS-SVM model
%         X(*)           : N x d matrix with the inputs of the training data
%         Y(*)           : N x 1 vector with the outputs of the training data
%

%===========================================================================
% 识别

Yd0 = simlssvm({X,Yc,type,gam,sig2,kernel}, {alpha,b}, Xt)

% Evaluate the LS-SVM at the given points
%
% >> [Yt, Zt, model] = simlssvm(model, Xt)
% 
%       Outputs    
%         Yt       : Nt x m matrix with predicted output of test data
%         Zt(*)    : Nt x m matrix with predicted latent variables of a classifier
%         model(*) : Object oriented representation of the LS-SVM model
%       Inputs    
%         model    : Object oriented representation of the LS-SVM model
%         Xt       : Nt x d matrix with the inputs of the test data

%===========================================================================
% 将训练输出解码成原始格式

Yd = code(Yd0,old_codebook,[],codebook)

%     2. For decoding:
% 
% >> Yd = code(Yc, old_codebook,[], codebook)
% >> Yd = code(Yc, old_codebook,[], codebook, codedist_fct)
% >> Yd = code(Yc, old_codebook,[], codebook, codedist_fct, codedist_args)
% 
%       Outputs    
%         Yd               : N x nc decoded output classifier
%       Inputs    
%         Y                : N x d matrix representing the original classifier
%         codebook         : d*nc matrix representing the original encoding
%         old_codebook     : bits*nc matrix representing the encoding of the given classifier
%         codedist_fct     : Function to calculate the distance between to encoded classifiers (e.g. codedist_hamming)
%         codedist_args(*) : Extra arguments of codedist_fct

Percent = 

%===========================================================================
%===========================================================================

% 注意：以这两种写法等价
% ============== 1 ==============
% [Yc,codebook,old_codebook] = code(Y, codefct)
% [alpha, b] = trainlssvm({X,Yc,type,gam,sig2,kernel,preprocess})
% Yd0 = simlssvm({X,Yc,type,gam,sig2,kernel}, {alpha,b}, Xt)
% Yd = code(Yd0,old_codebook,[],codebook)

% ============== 2 ==============
% model = initlssvm(X,Y,type,gam,sig2,kernel,preprocess)
% model = changelssvm(model,'codetype',codefct)
% model = trainlssvm(model)
% Yd = simlssvm(model, Xt)