CodeForge.com

Components_with_DF.m in mitmatlab.rar

			
function D = Components_with_DF(train_features, train_targets, Ncomponents, region)

% Classify points using component classifiers with discriminant functions
% Inputs:
% 	train_features	- Train features
%	train_targets	- Train targets
%	Ncomponents		- Number of component classifiers
%	region			- Decision region vector: [-x x -y y number_of_points]
%
% Outputs
%	D					- Decision sufrace
%
% This implementation works with logistic component classifiers and a softmax gating function
% The parameters of the components are learned using Newton descent, and the parameters
% of the gating system using gradient descent

[Ndim, M] 	    = size(train_features);
Ndim			= Ndim + 1;
x				= [train_features; ones(1,M)];
y				= train_targets;
theta			= zeros(Ndim, Ncomponents)+eps;
alpha			= randn(Ndim, Ncomponents);
alpha			= sqrtm(cov(x',1)+randn(Ndim))*alpha + mean(x')'*ones(1,Ncomponents);

old_err		    = 10;
err			    = 1;

while ((err > 1/M) & (err < old_err)),
    old_err = err;
    
    %Perform gradient descent on the component classifiers
    w			= exp(alpha'*x)./(ones(Ncomponents,1)*sum(exp(alpha'*x)));
    for i = 1:Ncomponents,
        p					= exp(			

			...
			...
			... to be continued.

  This is a preview. To get the complete source file, 
  please click here to download the whole source code package.