.mu and .su

What do lines 41-42 mean?

% ICS5110 - Applied Machine Learning
% University of Malta
% Lecturer: Dr. George Azzopardi
% Date: 27 October, 2015

function accuracy = NaiveBayesIris(L2norm)

load('irisData.mat');
load('irisLabels.mat');

% Create a random permutation
if exist('randpermlist.mat')
    load('randpermlist.mat');
else
    randpermlist = randperm(numel(irisLabels));
    save randpermlist randpermlist;
end

if L2norm
    irisData = normr(irisData);
end

% Split data set into 50% training and 50% testing
ntraining = floor(0.5*numel(irisLabels));
trainingData = irisData(randpermlist(1:ntraining),:);
trainingLabels = irisLabels(randpermlist(1:ntraining));
testingData = irisData(randpermlist(ntraining+1:end),:);
testingLabels = irisLabels(randpermlist(ntraining+1:end));

% Prior class probabilities
uniqueClasses = unique(trainingLabels);
[classidx,classlbl] = grp2idx(trainingLabels);
h = hist(classidx,numel(uniqueClasses));
prior = h./sum(h);

% Likelihood
likelihood.mu = zeros(numel(uniqueClasses),size(trainingData,2));  _/% explanation required_
likelihood.su = zeros(numel(uniqueClasses),size(trthainingData,2)); /% explanation required
for i = 1:numel(uniqueClasses)
    idx = find(classidx == i);
    likelihood.mu(i,:) = mean(trainingData(idx,:));
    likelihood.su(i,:) = std(trainingData(idx,:));    
end

% Classification
for i = 1:size(testingData,1)
   for j = 1:numel(uniqueClasses) 
       % Guassian Function Kernel
       squaredDifference = (testingData(i,:) - likelihood.mu(j,:)).^2;
       normFactor = 1./(sqrt(2*pi)*likelihood.su(j,:));
       likelihood.prob = normFactor .* exp(-squaredDifference/(2.*(likelihood.su(j,:).^2)));
       %posterior(j) = prod(likelihood.prob) * prior(j);
       posterior(j) = sum(log(likelihood.prob)) + log(prior(j));
   end
   [mx,mxind] = max(posterior);
   predictedLabel(i) = classlbl(mxind);
end

accuracy = sum(strcmp(predictedLabel',testingLabels))/numel(testingLabels);

Answer 1

[Walter Roberson]

http://www.mathworks.com/help/matlab/structures.html

[Jon Camilleri]

I did not quite find the answer to my question as yet but thanks.

[Walter Roberson]

The mu are means of each class and the su are standard deviations of each class.