myperceptron.m

最新的模式识别分类工具箱,希望对朋友们有用!

function w_percept = MyPerceptron(train_features, train_targets, alg_param)

% Trains using the Perceptron algorithm
% Inputs:
% 	featvect    - feature vectors: column = feature vector, row = feature
%	labels	    - Labels of feature vectors: 0 and 1 (do not use 2, 3 etc.
%	alg_param   - Either: Number of iterations, weights vector or [weights, number of iterations]
%
% Outputs
%	weights	     - the weight vector
% x * weights > 0 -> class 1
% x * weights < 0 -> class 0

[c, r]		   = size(train_features);

%Weighted Perceptron or not?
switch length(alg_param),
case r + 1,
    %Ada boost form
    p           = alg_param(1:end-1);
    max_iter    = alg_param(end);
case {r,0},
    %No parameter given
    p           = ones(1,r);
    max_iter    = 5000;
otherwise
    %Number of iterations given
    max_iter    = alg_param;
    p           = ones(1,r);
end

train_features = [train_features ; ones(1,r)];
train_one      = find(train_targets == 1);
train_zero     = find(train_targets == 0);

%Preprocessing
processed_features = train_features;
processed_features(:,train_zero) = -processed_features(:,train_zero);

%Initial weights
w_percept	= rand(c+1,1);

correct_classified = 0;
n						 = length(train_targets);
iter					 = 0;

while (  (longest_run(w_percept, processed_features) < 0.9*n) ...
       & (iter < max_iter))
   iter = iter + 1;
   indice = 1 + floor(rand(1)*n);
   if (w_percept' *  processed_features(:,indice) <= 0)
      w_percept = w_percept + p(indice)* processed_features(:,indice);
   end
end
disp(w_percept);
if (iter == max_iter)&(length(alg_param)~= r + 1),
   disp(['Maximum iteration (' num2str(max_iter) ') reached']);
end

%Find decision region