giniplot.m

the matlab code of ginisvm - an svm implementation good at probability regression.

function plotMsv(X,Y,alpha,bias,ratedist,aspect,mag,xaxis,yaxis,input)
%Gini-Support Vector Machine Plotting routine
% Adapted from Steve Gunn's SVC plotting function
%
%  Usage: giniplot(X,Y,ker,alpha,bias,zoom,xaxis,yaxis,input)
%
%  Parameters: X      - Training inputs
%              alpha  - Lagrange Multipliers
%              bias   - Bias term 
%              aspect - Aspect Ratio (default: 0 (fixed), 1 (variable))
%              mag    - display magnification 
%              xaxis  - xaxis input (default: 1) 
%              yaxis  - yaxis input (default: 2)
%              input  - vector of input values (default: zeros(no_of_inputs))
%

  nargin
  if (nargin <5 | nargin > 10) % check correct number of arguments
    help plotMsvbw
  else

    epsilon = 1e-5;  
    if (nargin < 10) input = zeros(1,size(X,2));, end
    if (nargin < 9) yaxis = 2;, end
    if (nargin < 8) xaxis = 1;, end
    if (nargin < 7) mag = 0.1;, end
    if (nargin < 6) aspect = 0;, end
   
    % Number of classes 
    M = length(alpha(1,:));
 
    % Scale the axes
    xmin = min(X(:,xaxis));, xmax = max(X(:,xaxis)); 
    ymin = min(X(:,yaxis));, ymax = max(X(:,yaxis)); 
    xa = (xmax - xmin);, ya = (ymax - ymin);
    if (~aspect)
       if (0.75*abs(xa) < abs(ya)) 
          offadd = 0.5*(ya*4/3 - xa);, 
          xmin = xmin - offadd - mag*0.5*ya;, xmax = xmax + offadd + mag*0.5*ya;
          ymin = ymin - mag*0.5*ya;, ymax = ymax + mag*0.5*ya;
       else
          offadd = 0.5*(xa*3/4 - ya);, 
          xmin = xmin - mag*0.5*xa;, xmax = xmax + mag*0.5*xa;
          ymin = ymin - offadd - mag*0.5*xa;, ymax = ymax + offadd + mag*0.5*xa;
       end
    else
       xmin = xmin - mag*0.5*xa;, xmax = xmax + mag*0.5*xa;
       ymin = ymin - mag*0.5*ya;, ymax = ymax + mag*0.5*ya;
    end
    
    set(gca,'XLim',[xmin xmax],'YLim',[ymin ymax]);  

    % Plot function value

    [x,y] = meshgrid(xmin:(xmax-xmin)/25:xmax,ymin:(ymax-ymin)/25:ymax); 
    for x1 = 1 : size(x,1)
      for y1 = 1 : size(x,2)
        input(xaxis) = x(x1,y1);, input(yaxis) = y(x1,y1);
        %for i = 1 : length(alpha(:,1))
        %  if (abs(alpha(i)) > epsilon)
            %prob = clusterMrun(input,X,Y,alpha,bias);
            %prob = logsmoMrun(input,X,alpha,bias);
            mvalue = kernel(input,X)*alpha+bias';
            prob = ginitest(mvalue,ratedist');
            for m = 1:M,
               z(x1,y1,m) = prob(m);
            end;
            zmax(x1,y1) = 1-max(prob);
        %  end
        %end
      end
      %waitbar((x1)/size(x,1))
    end
    % Plot Training points
    sp = pcolor(x,y,zmax);
    shading interp;

    hold on
    for i = 1:size(alpha(:,1))
      ind = find(Y(i,:)==1);
     
      % 3 class plot 
      if ind == 1,
        plot(X(i,xaxis),X(i,yaxis),'wo','LineWidth',1) % Class A
      else
        if ind == 2,
           plot(X(i,xaxis),X(i,yaxis),'w+','LineWidth',1) % Class B
        else
           if ind == 3,
              plot(X(i,xaxis),X(i,yaxis),'w*','LineWidth',1) % Class B
           else
              if ind == 4,
                 plot(X(i,xaxis),X(i,yaxis),'ws','LineWidth',1) % Class B
              else
                 plot(X(i,xaxis),X(i,yaxis),'w^','LineWidth',1) % Class B
              end;
           end;
        end;
      end
      %if (abs(alpha(i)) > epsilon)
      %  plot(X(i,xaxis),X(i,yaxis),'kh') % Support Vector
      %end
    end 

    % Plot Boundary contour

    hold on
    for m = 1:M,
       %contour(x,y,z(:,:,m),[0 0],'k')
       contour(x,y,z(:,:,m),[0.5 0.5],'w--')
       %contour(x,y,z(:,:,m),[0.7 0.7],'k.')
       %contour(x,y,z(:,:,m),[0.3 0.3],'k.-')
    end;
    xlabel('x1');
    ylabel('x2');
    hold off
    colormap(gray);

  end