pnndemo.m

PNN又称为概率神经网络,它最初由数学家Specht于1990年提出,后经Master[1995]等不断发展和完善,已成功地应用于机器学习、人工智能、自动控制等众多领域.概率神经网络比多层前馈网络的数

% PNNDEMO -- Probabilistic Neural Network Demo m-file
% Ron Shaffer
% Naval Research Laboratory
% 3-21-96
%
fprintf ('Probabilistic Neural Network Demo \n')
fprintf ('Ron Shaffer \n')
fprintf ('Naval Research Laboratory \n \n');
junk = input ('Hit any key to continue');
fprintf ('Loading Mark Andersons simulated surface acoustic wave chemical sensor data set. \n');
load ma.mat
fprintf ('The goal of this classification problem is to predict the classification \n');
fprintf ('of the vapors in the test set.  The training set is a 300 by 3 matrix \n');
fprintf ('containing the responses for a three sensor array to 300 samples.  In \n');
fprintf ('this matrix the first 100 samples were contained chemical vapor 1, the \n');
fprintf ('second 100 samples were chemical vapor 2, and the last 100 \n');
fprintf ('samples were chemical vapor 3.  The test set is constructed in \n');
fprintf ('a similar manner but does not contain any samples found in the training set. \n');;
fprintf ('For more information about how this simulated data were constructed see \n');
fprintf ('Mark Anderson and Dave Venezsky, Investigation of Probabilistic Neural \n');
fprintf ('Networks for Sensor Array Pattern Recognition, NRL-Memorandum Report 7798 \n');
fprintf (', February 1996 or email:  [e-mail]. \n');
junk = input ('Hit any key to continue');
fprintf ('Probabilistic Neural Networks (PNN) require little training! \n');
fprintf ('Training involves simply placing the patterns in the training set into the hidden units \n');
fprintf ('of the neural network. Although, for optimal performance, adjustment of the smoothing \n');
fprintf ('factor must be performed.  Now we can predict the classification of the patterns \n');
fprintf ('in the prediction (test) set.\n');
smooth = input ('Input a smoothing factor for the PNN [I recommend 0.1]  ');
class_info = [100 100 100];
[raw,prob] = sprobnn(inp,class_info,test,smooth);
prob = prob .* 100;
fprintf ('Finished classification of the patterns in the test set \n');
junk = input ('Hit any key to see results');
fprintf('Probability of each pattern belonging to each class (chemical vapor): \n');
fprintf('Sample # Vapor1   Vapor2    Vapor3 \n'); 
for j = 1:300
	fprintf ('%d              %7.4f    %7.4f    %7.4f \n',j,prob(j,1),prob(j,2),prob(j,3));
end
%
% compute percentage correct
%
hold1 = prob';
[hold2,winout] = max(hold1);
winout = winout';
junk = input ('Hit any key to see the number correctly classified for each chemical vapor');
%
%
% now compute the percentage correct for each output class
%
count = 0;
missedct = 0;
for i = 1:3
	ncorrect(i) = 0;
	for j = 1:100
		count = count + 1;
		if winout(count) == i
			ncorrect(i) = ncorrect(i) + 1;
		else
			missedct = missedct + 1;
			mispat(missedct,1) = count;
			mispat(missedct,2:4) = raw(count,1:3);
			mispat(missedct,5:7) = test(count,1:3);
		end
	end
end
%

fprintf('The percentage correct is: \n');
for i = 1:3
	fprintf('Chemical Vapor %d \t %6.3f \n',i,ncorrect(i)');
end

fprintf ('End of PNN Demo \n');