trainlssvm.m

支持向量机学习中用到的用Matlab编写的工具箱。

function [model,b,X,Y]  = trainlssvm(model,X,Y) 
% Train the support values and the bias term of an LS-SVM for classification or function approximation
% 
% >> [alpha, b] = trainlssvm({X,Y,type,gam,kernel_par,kernel,preprocess})
% >> model      = trainlssvm(model)
% 
% type can be 'classifier' or 'function estimation' (these strings
% can be abbreviated into 'c' or 'f', respectively). X and Y are
% matrices holding the training input and output data. The i-th
% data point is represented by the i-th row X(i,:) and Y(i,:). gam
% is the regularization parameter: for gam low minimizing of the
% complexity of the model is emphasized, for gam high, good fitting
% of the training data points is stressed. kernel_par is the
% parameter of the kernel; in the common case of an RBF kernel, a
% large sig2 indicates a stronger smoothing. The kernel_type
% indicates the function that is called to compute the kernel value
% (by default RBF_kernel). Other kernels can be used for example: 
% 
% >> [alpha, b] = trainlssvm({X,Y,type,gam,[d p],'poly_kernel'})
% >> [alpha, b] = trainlssvm({X,Y,type,gam,[]   ,'lin_kernel'})
% 
% The kernel parameter(s) are passed as a row vector, in the case
% no kernel parameter is needed, pass the empty vector! 
% 
% The training can either be proceeded by the preprocessing
% function ('preprocess') (by default) or not ('original'). The
% training calls the preprocessing (prelssvm, postlssvm) and the
% encoder (codelssvm) if appropiate. 
% 
% In the remainder of the text, the content of the cell determining
% the LS-SVM is given by {X,Y, type, gam, sig2}. However, the
% additional arguments in this cell can always be added in the
% calls. 
% 
% If one uses the object oriented interface (see also A.3.14), the training is done by
% 
% >> model = trainlssvm(model)
% >> model = trainlssvm(model, X, Y)
% 
% The status of the model checks whether a retraining is
% needed. The extra arguments X, Y allow to re-initialize the model
% with this new training data as long as its dimensions are the
% same as the old initiation. 
% 
% Three training implementations are included:
% 
%     * The C-implementation linked with CMEX: this implementation
%     is based on the iterative solver Conjugate Gradient algorithm
%     (CG) (lssvm.mex*). After this training call, a '-' is
%     displayed. This is recommended for use on larger data sets. 
% 
%     * The C-implementation called via a buffer file: this is
%     based on CG; check if the executable 'lssvmFILE.x' is in the
%     current directory; (lssvmFILE.x). After this training call, a
%     '-' is displayed. 
%  
%     * The Matlab implementation: a straightforward implementation
%     based on the matrix division '\' (lssvmMATLAB.m). After this
%     training call, a '~' is  displayed. This is recommended for a
%     number of training data points smaller than 500 (depending on
%     the computer memory).  
% 
% By default, the cmex implementation is called. If this one fails,
% the Matlab implementation is chosen instead. One can specify
% explicitly which implementation to use using the object oriented
% interface. 
% 
% This implementation allows to train a multidimensional output
% problem. If each output uses the same kernel type, kernel
% parameters and regularization parameter, this is
% straightforward. If not so, one can specify the different types
% and/or parameters as a row vector in the appropriate
% argument. Each dimension will be trained with the corresponding
% column in this vector. 
% 
% >> [alpha, b] = trainlssvm({X, [Y_1 ... Y_d],type,...
%                              [gam_1 ... gam_d], ...
%                             [sig2_1 ... sig2_d],...
%                           {kernel_1,...,kernel_d}})
% 
% Full syntax
% 
%     1. Using the functional interface:
% 
% >> [alpha, b] = trainlssvm({X,Y,type,gam,sig2})
% >> [alpha, b] = trainlssvm({X,Y,type,gam,sig2,kernel})
% >> [alpha, b] = trainlssvm({X,Y,type,gam,sig2,kernel,preprocess})
% 
%       Outputs    
%         alpha         : N x m matrix with support values of the LS-SVM
%         b             : 1 x m vector with bias term(s) of the LS-SVM
%       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')
%         gam           : Regularization parameter
%         sig2          : Kernel parameter (bandwidth in the case of the 'RBF_kernel')
%         kernel(*)     : Kernel type (by default 'RBF_kernel')
%         preprocess(*) : 'preprocess'(*) or 'original'
% 
%
%     * Using the object oriented interface:
% 
% >> model = trainlssvm(model)
% >> model = trainlssvm({X,Y,type,gam,sig2})
% >> model = trainlssvm({X,Y,type,gam,sig2,kernel})
% >> model = trainlssvm({X,Y,type,gam,sig2,kernel,preprocess})
% 
%       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
%         type(*)        : 'function estimation' ('f') or 'classifier' ('c')
%         gam(*)         : Regularization parameter
%         sig2(*)        : Kernel parameter (bandwidth in the case of the 'RBF_kernel')
%         kernel(*)      : Kernel type (by default 'RBF_kernel')
%         preprocess(*)  : 'preprocess'(*) or 'original'
% 
% See also:
%   simlssvm, initlssvm, changelssvm, plotlssvm, prelssvm, codelssvm


% Copyright (c) 2002,  KULeuven-ESAT-SCD, License & help @ http://www.esat.kuleuven.ac.be/sista/lssvmlab


%
% initialise the model 'model'
%
if (iscell(model)),
  model = initlssvm(model{:});
end




%
% given X and Y?
%
%model = codelssvm(model);
eval('model = changelssvm(model,''xtrain'',X);',';');
eval('model = changelssvm(model,''ytrain'',Y);',';');
eval('model = changelssvm(model,''selector'',1:size(X,1));',';');


%
% no training needed if status = 'trained'
%
if model.status(1) == 't',
  if (nargout>1),
    % [alpha,b]
    X = model.xtrain;
    Y = model.ytrain;
    b = model.b;
    model = model.alpha;
  end
  return
end
  

%
% control of the inputs
%
if ~((strcmp(model.kernel_type,'RBF_kernel') & length(model.kernel_pars)>=1) |...
     (strcmp(model.kernel_type,'lin_kernel') & length(model.kernel_pars)>=0) |...
     (strcmp(model.kernel_type,'MLP_kernel') & length(model.kernel_pars)>=2) |...
     (strcmp(model.kernel_type,'poly_kernel')& length(model.kernel_pars)>=1)),

  eval('feval(model.kernel_type,model.xtrain(1,:),model.xtrain(2,:),model.kernel_pars);model.implementation=''MATLAB'';',...
       'error(''The kernel type is not valid or to few arguments'');');
elseif (model.steps<=0),
  error('steps must be larger then 0');
elseif (model.gam<=0),
  error('gamma must be larger then 0');
% elseif (model.kernel_pars<=0),
%   error('sig2 must be larger then 0');
elseif or(model.x_dim<=0, model.y_dim<=0),
  error('dimension of datapoints must be larger than 0');
end


%
% coding if needed
%
if model.code(1) == 'c', % changed
  model = codelssvm(model);
end

%
% preprocess
%
eval('if model.prestatus(1)==''c'', changed=1; else changed=0;end;','changed=0;');
if model.preprocess(1) =='p' & changed,
  model = prelssvm(model);
elseif model.preprocess(1) =='o' & changed 
  model = postlssvm(model);
end

% clock
tic;

%
% set & control input variables and dimensions
% 
if (model.type(1) == 'f'), % function
  dyn_pars=[];
elseif (model.type(1) == 'c'), % class
  dyn_pars=[];  
end


% only MATLAB
if size(model.gam,1)>1, 
  model.implementation='MATLAB'; 
end


%
% output dimension > 1...recursive call on each dimension
%
if model.y_dim>1,
  if (length(model.kernel_pars)==model.y_dim | size(model.gam,2)==model.y_dim |prod(size(model.kernel_type,2))==model.y_dim)
    disp('multidimensional output...');
    model = trainmultidimoutput(model);
    %
    % wich output is wanted?
    %
    if (nargout>1),
      X = model.xtrain;
      Y = model.ytrain;
      b = model.b;
      model = model.alpha;
    else     
      model.duration = toc;
      model.status = 'trained';
    end
    return  
  end
end


%
% call lssvmMATLAB.m, lssvm.mex* or lssvmFILE.m
%
if strcmpi(model.implementation,'CMEX'),
  model.cga_startvalues = [];
  eval('model.cga_startvalues;','model.cga_startvalues = [];');

  
  eval(['[model.alpha, model.b,model.cga_startvalues] =' ...
	'lssvm(model.xtrain(model.selector, 1:model.x_dim)'',model.x_dim,'...
	'model.ytrain(model.selector, 1:model.y_dim),model.y_dim,'...
	'model.nb_data, model.type, model.gam,' ...
	'model.cga_eps, model.cga_fi_bound,model.cga_max_itr,' ...
	'model.cga_startvalues,'...
	'model.kernel_type,  model.kernel_pars,' ...
	'model.cga_show,dyn_pars);'],...
        'model.implementation=''CFILE''; disp(''converting now to CFILE implementation'');');% if error in CMEX ...

end

if strcmpi(model.implementation,'CFILE'),
  eval('model.cga_startvalues;','model.cga_startvalues = [];');
  eval('model = lssvmFILE(model,''buffer.mc'');',...
       ['model.implementation=''MATLAB'';'...
	'disp(''make sure lssvmFILE.x (lssvmFILE.exe) is in the' ...
	 ' current directory, change now to MATLAB implementation...'');']); 
  % if error in CFILE ...
end

if strcmpi(model.implementation(1),'m'),
  model = lssvmMATLAB(model);
end





%
% wich output is wanted?
%
if (nargout>1),
  X = model.xtrain;
  Y = model.ytrain;
  b = model.b;
  model = model.alpha;
else     
  model.duration = toc;
  model.status = 'trained';
end
  
  


%
%
function model = trainmultidimoutput(model)
%
% 
%
  model.alpha = zeros(model.nb_data, model.y_dim);
  model.b = zeros(1,model.y_dim);
  model.cga_startvalues = [];
  for d=1:model.y_dim,
    eval('gam = model.gam(:,d);','gam = model.gam(:);');
    eval('sig2 = model.kernel_pars(:,d);','sig2 = model.kernel_pars(:);');
    eval('kernel = model.kernel_type{d};','kernel=model.kernel_type;');
    [model.alpha(:,d),model.b(d)] = trainlssvm({model.xtrain,model.ytrain(:,d),model.type,gam,sig2,kernel,'original'});
  end
  
  %
  % wich output is wanted?
  %
  if (nargout>1),
    X = model.xtrain;
    Y = model.ytrain;
    b = model.b;
    model = model.alpha;
  else     
    model.duration = toc;
    model.status = 'trained';
  end