pmsvm.m

来自「很好用的统计模式识别工具箱」· M 代码 · 共 82 行

function pmsvm(Xtrn,Itrn,malpha,mbias,ker,arg,background)
% MPSVM vizualizes decision rule of multiclass SVM problem.
%   pmsvm(X,I,malpha,mbias,ker,arg )
%
% PMSVM vzualizes multiclass SVM problem decomposed in 2D.
%
% Input:
%  Xtrn [2xM] set of training patterns.
%  Itrn [1xM] labels (1,2,...K).
%  malpha [KxM] weights of traning patterns for each dichotomi. 
%    Supprot Vectors are Xtrn(:,find(alpha>0)).
%  mbias [Kx1] bias of used discrimination function.
%  ker [string] identifies kernel (see help KERNEL).
%  arg [...] arguments of given kernel (see help KERNEL).
%
% See also MSVMCLASS, SVM.
%

% Statistical Pattern Recognition Toolbox, Vojtech Franc, Vaclav Hlavac
% (c) Czech Technical University Prague, http://cmp.felk.cvut.cz
% Written Vojtech Franc (diploma thesis) 23.12.1999, 5.4.2000
% Modifications
% 19-sep-2001, V. Franc, comments changed.
% 20-may-2001, V. Franc, new approach
% 16-april-2001, V. Franc, created

if nargin < 7,
  background = 0;
end

% points size
POINTSSIZE=10;  

% grid for x-axis and y-axis
GRIDX=150;
GRIDY=150;

epsilon=1e-5;

if nargin < 6,
   error('Not enough input arguments.');
   return;
end

ppatterns(Xtrn,Itrn,POINTSSIZE);
hold on;

V = axis;
dx = (V(2)-V(1))/GRIDX;
dy = (V(4)-V(3))/GRIDY;

[X,Y] = meshgrid(V(1):dx:V(2),V(3):dy:V(4));

% make testing points
Xtst=[reshape(X',1,prod(size(X)));reshape(Y',1,prod(size(Y)))];


% classify points
nclass=max(Itrn);
D = zeros(nclass,size(Xtst,2) );

for i=1:nclass,
  [Ipred,fpred] = svmclass2(Xtst,Xtrn,multi2dicho(Itrn,i),...
      malpha(i,:),mbias(i),ker,arg);
  
  D(i,:) = fpred; 
end

pdiscrim( D, V(1):dx:V(2), V(3):dy:V(4),background );

if background,
  ppatterns(Xtrn,'kx',POINTSSIZE);
else
  ppatterns(Xtrn,Itrn,POINTSSIZE);
end

axis(V);
hold off;

return;