svmdemo.m

支持向量机工具箱

   
   [Alpha,bias] = svm_train( data, trn, ker, arg, C );
   
   % focus on the axes
   axes( data.axes );
   
   % Clear axes
   clrchild( data.axes);
   
   % get options
   options.background = get( data.xb_background, 'Value'); 
   
   % plot decision function
   psvm( trn.X,trn.I,Alpha,bias,ker,arg, options );
   

%== Handler for Algorithm popup menu ==========================
case 'algo_handler'
   data=get(hfigure,'UserData');
   
   % which algorithm ?
   switch get(data.pu_algo, 'Value' )
     
    case 1   % SMO
        set(data.tx_cconst,'Enable','on');
        set(data.ed_cconst,'Enable','on');
        set(data.tx_param,'Enable','on');
        set(data.tx_param,'String',SMO_PARAM);
        set(data.ed_param,'Enable','on');
        set(data.ed_param,'String',DEF_SMO_PARAM);
        
    case 2    % Matlab toolbox
        set(data.tx_cconst,'Enable','on');
        set(data.ed_cconst,'Enable','on');
        set(data.tx_param,'Enable','off');
        set(data.ed_param,'Enable','off');
     
    case 3   % Kernel-SK
        set(data.tx_cconst,'Enable','on');
        set(data.ed_cconst,'Enable','on');
        set(data.tx_param,'Enable','on');
        set(data.ed_param,'Enable','on');
        set(data.tx_param,'String',KERNELSK_PARAM);
        set(data.ed_param,'String',DEF_KERNELSK_PARAM);
    case 4    % Matlab toolbox
        set(data.tx_cconst,'Enable','on');
        set(data.ed_cconst,'Enable','on');
        set(data.tx_param,'Enable','off');
        set(data.ed_param,'Enable','off');
    case 5    % Kernel Perceptron
        set(data.tx_cconst,'Enable','off');
        set(data.ed_cconst,'Enable','off');
        set(data.tx_param,'Enable','on');
        set(data.ed_param,'Enable','on');
        set(data.tx_param,'String',KPERCEPTR_PARAM);
        set(data.ed_param,'String',DEF_KPERCEPTR_PARAM);
   end
   
   
%== Handler for C-const edit line ===========================
case 'cconst_handler'
   data=get(hfigure,'UserData');

   C = str2num(get(data.ed_cconst,'String'));
   
   if C <= 0,
      C = 100;
   end
   set( data.ed_cconst,'String',num2str(C));
   
%== Handle for arg edit line ===========================
case 'arg_handler'
   data=get(hfigure,'UserData');

   arg = str2num(get(data.ed_arg,'String'));
   
   if arg < 0,
      arg = 1;
   end
   
   set( data.ed_arg, 'String',num2str( arg));

   
%== Handle for kernel pop up menu ===========================
case 'kernel_handler'
   data=get(hfigure,'UserData');

   ker_inx = get( data.pu_kernel,'Value');
   if ker_inx >= 2,
      set( data.ed_arg,'Enable','on');
      set( data.tx_arg,'Enable','on');
   else
      set( data.ed_arg,'Enable','off');
      set( data.tx_arg,'Enable','off');
   end

%== Calls data creator ==========================================
case 'creatdata'
   creatset('finite',2,'svmdemo','created',hfigure);


% == Loads recently created data ================================
case 'created'

   % get handler and make this figure active
   figure(hfigure);
   data = get(hfigure,'UserData');

   % get file name
   path=varargin{1};
   name=varargin{2};
   pathname=strcat(path,name);

   if checkdat(pathname,DATA_IDENT,2,[0 0])==1,
      file.pathname=pathname;
      file.path=path;
      file.name=name;
      set( data.bt_load,'UserData',file);
      svmdemo('loadsets',hfigure);
      svmdemo('reset',hfigure);
   else
      errordlg('This file does not contain required data.','Bad file','modal');
   end

% == Calls standard open file dialog ==========================
case 'getfile'
   
   data=get(hfigure,'UserData');

   [name,path]=uigetfile('*.mat','Open file');
   if name~=0,
      file.pathname=strcat(path,name);
      file.path=path;
      file.name=name;
      if checkdat(file.pathname,DATA_IDENT,2,[0 0])==1,
         set(data.bt_load,'UserData',file);
         svmdemo('loadsets',hfigure);
         svmdemo('reset',hfigure);
      else
         errordlg('This file does not contain required data.','Bad file','modal');
      end
   end

case 'loadsets'
   % == Load data from file ==========================================

   data = get( hfigure,'UserData' );

   % Clear axes
   clrchild( data.axes);
   
   % set x and y axes labels
   xlabel('feature x');
   ylabel('feature y');

   % Get file name with sets
   file=get( data.bt_load,'UserData');

   % Load sets
   trn = load(file.pathname );

   % store loaded sets
   set( data.axes,'UserData', trn);

   % focus on axes
   axes( data.axes );

   % plots points
   ppatterns( trn.X, trn.I, 5);

   drawnow   
  

% == Reset ==========================================================
case 'reset'

   data = get(hfigure,'UserData');                 
   
   % Clear axes
   clrchild( data.axes);

   % get data set
   trn = get( data.axes, 'UserData');

   % get file
   file = get( data.bt_load,'UserData');
   
   % create comment
   if isempty( trn ) == 0,
     consoletext=sprintf('Data loaded.\nSelect algorithm and press Train SVM button.\n');
     titletext=sprintf('File: %s, # of points K = %d', file.name , size(trn.X,2));
     
     set( data.axes, 'XLimMode','auto', 'YLimMode','auto');
     ppatterns( trn.X, trn.I, 5 );
     
   else
     consoletext=sprintf('No data loaded.\nPress Load data button.\n');
     titletext='';

     pos=get( data.axes,'Position');
     fsize=min(pos(3),pos(4))/10;
     setaxis( data.axes,[-1 1 -1 1]);
%     axis([-1 1 -1 1]);
      
     builtin('text',0,0,'Press ''Load data'' button.',...
        'HorizontalAlignment','center',...
        'FontUnits','normalized',...
        'Clipping','on',...
        'FontSize',fsize);
   end

   % show comment
   set( data.console,'String',consoletext );

   % print title
   pos=get( data.axes,'Position');
   fsize=(1-pos(2)-pos(4))*1;
   title(titletext,...
      'Parent', data.axes,...
      'VerticalAlignment','bottom',...
      'HorizontalAlignment','left',...
      'FontUnits','normalized',...
      'Units','normalized',...
      'Position',[0 1 0],...
      'FontSize',fsize);


% == Calls Matlab`s info box ==================================
case 'info'
   helpwin(mfilename);
end

return;

%===============================================================
function [Alpha,bias] = svm_train( data, trn, ker, arg, C )

if strcmpi( ker, 'linear'),
  strarg = '-';
else
  strarg = num2str( arg);
end

param = str2num( get( data.ed_param,'String'));

switch get( data.pu_algo, 'Value' ),
  case 1   % Sequential Minimal Optimizer

%   if length(param) ~= length( str2num(DEF_SMO_PARAM)),
%     param = str2num(DEF_SMO_PARAM);
%   end
   
   [Alpha,bias,nsv,kercnt,trn_err,margin]=...
         smo(trn.X,trn.I,ker,arg,C, param(1),param(2) );
          
     text = sprintf(...
        ['SVM (L1) by Sequential Minimal Optimizer\n',...
        'Kernel: %s (%s), C: %.4f\n',...
        'Kernel evaluations: %d\n',...
        'Number of Support Vectors: %d\n',...
        'Margin: %.4f\n',...
        'Training error: %.2f%%'],...
        ker, strarg, C, kercnt, nsv, margin, 100*trn_err );
      
  case 2   % Matlab Optimization toolbox (L1)
     [Alpha,bias,nsv,exitflag,flps,margin,trn_err]=...
         svmmot(trn.X,trn.I,ker,arg,C);    
     
     text = sprintf(...
        ['SVM (L1) by QP from the Matlab Optimization Toolbox\n',...
        'Kernel: %s (%s), C: %.4f\n',...
        'Optimization: %s\n',...
        'Number of Support Vectors: %d\n',...
        'Floating point operations: %d\n',...
        'Margin: %.4f\n',...
        'Training error: %.2f%%'],...
        ker, strarg, C, exitflag, nsv, flps, margin,trn_err);
   
  case 3   % Kernel Schlesinger-Kozinec's algorithm (L2)

%     if length(param) ~= length( str2num(DEF_KERNELSK_PARAM)),
%       param = str2num(DEF_KERNELSK_PARAM);
%     end

%     [Alpha,bias,sol,t,flps,margin,up,lo,trn_err]=...
%         kernelsk(trn.X,trn.I,param(1),ker,arg,param(2),C);
%     [Alpha,bias,sol,t,flps,margin,up,lo]=...
%         kerskfmat(trn.X,trn.I,param(1),ker,arg,param(2),C);

%     inx1=find(trn.I==1);
%     inx2=find(trn.I==2);
     
     [Alpha,bias,sol,t,kercnt,margin,trn_err]= ...
         kernelskf(trn.X, trn.I,[1 param(1)],ker,arg,param(2),C);    
%     Alpha = zeros(1,length(Alpha1)+length(Alpha2));
%     Alpha(inx1) = Alpha1;
%     Alpha(inx2) = Alpha2;
     
     nsv = length( find( Alpha > 1e-9 ));
     
     if( sol == 1), sol = 'found'; else sol = 'not found'; end
     
     text = sprintf(...
        ['SVM (L2) by Kernel Schlesinger-Kozinec`s algorithm\n',...
        'Kernel: %s (%s), C: %.4f\n',...
        'Solution: %s\n', ...
        'Number of Support Vectors: %d\n',...
        'Kernel evaluations: %d\n',...
        'Margin: %.4f\n',...
        'Training error: %.2f%%'],...
        ker, strarg, C, sol, nsv, kercnt, margin, trn_err*100 );

  case 4   % Matlab Optimization toolbox (L2)
     [Alpha,bias,nsv,exitflag,flps,margin,trn_err]=...
         svm2mot(trn.X,trn.I,ker,arg,C);

     text = sprintf(...
        ['SVM (L2) by QP from the Matlab Optimization Toolbox\n',...
        'Kernel: %s (%s), C: %.4f\n',...
        'Optimization: %s\n',...
        'Number of Support Vectors: %d\n',...
        'Floating point operations: %d\n',...
        'Margin: %.4f\n',...
        'Training error: %.2f%%'], ...
        ker, strarg, C, exitflag, nsv, flps, margin, trn_err*100);
  
  case 5   % Kernel Perceptron
     [Alpha,bias,sol,t,nsv]=kperceptr(trn.X,trn.I,ker,arg,param(1));

     if( sol == 1), sol = 'found'; else sol = 'not found'; end

     text = sprintf(...
        ['Kernel Perceptron\n',...
        'Kernel: %s (%s)\n',...
        'Solution: %s\n',...
        'Number of iterations: %d\n',...
        'Number of Support Vectors: %d'], ...
        ker, strarg, sol, t, nsv);
end

% show comment
set( data.console,'String', text );

return;