ada_boost.m

最新的模式识别分类工具箱,希望对朋友们有用!

function D = ada_boost(train_features, train_targets, params, region);

% Classify using the AdaBoost algorithm
% Inputs:
% 	features	- Train features
%	targets	- Train targets
%	Params:
%			1. NumberOfIterations
%			2. Weak Learner Type (Perceptron, Pocket, LS, ML, RDA )
%			3. Learner's parameters ( 	Perceptron - Number Of  Iterations,
%							  	Pocket - Number Of  Iterations,
%								LS - None,
%								ML - None,
%								Friedman - Lambda   )
%	region	- Decision region vector: [-x x -y y number_of_points]
%
% Outputs
%	D			- Decision sufrace
disp(params);
comma_loc = findstr(params,',');
Iter      = str2num(params(2:comma_loc(1)-1));
alg_iter  = str2num(params(comma_loc(2)+1:length(params)-1));
method    = params(comma_loc(1)+2:comma_loc(2)-2);

indexs 			= round((train_features - (region([1,3])' * ones(1,size(train_features,2)))) * (region(5)-1) ./...
   					((region([2,4]) - region([1,3]))' * ones(1,size(train_features,2))))...
                  +  ones(size(train_features));

N = size(train_targets);
ind = (indexs(1,:)-1)*region(5) + indexs(2,:);
beta = [];
D_final = zeros(region(5));
distr_vector = ones(N) / N(2);

flg = 0;

for k = 1:Iter,
   p            = distr_vector / sum(distr_vector);
   D            = feval(method, train_features, train_targets, [p,alg_iter], region);    
   D1           = D(:)';
   new_targets  = D1(ind);
   delta        = xor(train_targets,new_targets);
   e            = sum(p .* delta);
   
   if e > 0.5
       D            = ~D;
       D1           = D(:)';
       new_targets  = D1(ind);
       delta        = xor(train_targets,new_targets);
       e            = sum(p .* delta);
   end
   
   entropy = - sum(p.*log(p))/log(N(2));
   %disp(['Error = ',num2str(e),' Entropy = ',num2str(entropy)]);
   
   if (abs(e-0.5)<1e-8),
      break
   end
   
   beta(k)      = e / (1 - e);
   distr_vector = distr_vector .* (beta(k) .^ ~delta);
   if beta(k) ~= 0
       D_final = D_final + log(1/beta(k)) * D;
   else
       flg = 1;
       break
   end
end

if k > 1
        D = (D_final >= 0.5 * sum(log(1./beta)));
end