www.pudn.com > Classification_toolbox.part01.rar > Discrete_Bayes.m

function test_targets = Discrete_Bayes(train_patterns, train_targets, test_patterns, cost) 
 
% Classify discrete patterns using the Bayes decision theory 
% Inputs: 
% 	train_patterns	- Train patterns 
%	train_targets	- Train targets 
%   test_patterns   - Test  patterns 
%	cost			- Cost for class 0 (Optional, Unused yet) 
% 
% Outputs 
%	test_targets	- Predicted targets 
 
%First, find out if the patterns are indeed discrete  
%Of course, since this program is in Matlab, we define discrete as having  
%no more than one decimal place  
if (sum(sum(train_patterns*10~=floor(train_patterns*10))) ~= 0), 
   error('Patterns are not discrete (See the definition of discrete in the m-file)') 
end 
 
Uc  = unique(train_targets); 
 
%Find how the patterns are distributed 
N	            = 0; 
unique_patterns = []; 
counts			= []; 
for i = 1:size(train_patterns,2), 
    data    = train_patterns(:,i); 
    indices = find(sum((data*ones(1,size(train_patterns,2))-train_patterns).^2)==0); 
    if isempty(unique_patterns), 
        unique_patterns(:,1) = data; 
        for j = 1:length(Uc), 
            counts(j,1) = length(find(train_targets(indices) == Uc(j))); 
        end 
        N   = 2; 
    else 
        if isempty(find(sum((data*ones(1,size(unique_patterns,2))-unique_patterns).^2)==0)), 
            %Add this pattern to the bank 
            unique_patterns(:,N) = data; 
            for j = 1:length(Uc), 
                counts(j,N) = length(find(train_targets(indices) == Uc(j))); 
            end 
            N	= N + 1; 
        end 
    end 
end 
 
Px_given_w = (counts ./ (ones(length(Uc),1)* sum(counts))); 
p          = hist(train_targets, Uc)' / length(train_targets); 
 
Pw_given_x = Px_given_w .* (p*ones(1,N-1)); 
Pw_given_x = Pw_given_x ./ (ones(length(Uc),1)*sum(Pw_given_x)); 
 
[m, unique_targets] = max(Pw_given_x); 
unique_targets      = unique_targets - 1; 
 
test_targets = Nearest_Neighbor(unique_patterns,unique_targets,test_patterns,1);