I am new to matlab. I can't figure out the problem with this code. It won't execute. It's for naive bayes to classify setosa (class = 1) against combined versicolor and virginca (class = 0) without using gaussian pdf

    load iris.dat
    %discretize data

    numBins = 5;
    dd = zeros(150,5);
    for q = 1:size(iris,2)-1                                                   
       x = iris(:,q);                                                          
       binEdges = linspace(min(x), max(x)+1 , numBins+1);              
       [bincount ,whichBin] = histc(x, binEdges);
       dd(:,q) = whichBin;
    end

     dd(:,5)= iris(:,5);

    %change class values 2 and 3 to 0 in the discretized data 

    for w=1:size(dd,1)
       if  dd(w,5)~= 1
          dd(w,5) = 0;                                
      end
    end

    %divide into train and test

    [train_indices,validation_set_indices,test_indices] = dividerand(150,0.5,0,0.5);
    traind = dd(train_indices,:);
    testd = dd(test_indices,:);

    % calculate P(class)
    s1 = 0;
    s2 = 0;
    for row = 1:size(traind,1)
     if traind(row,5) == 1
      s1 = s1 + 1;
     if traind(row,5) == 0
      s2 = s2 + 1;
    p1 = s1/size(traind,1); % P(class = 1)
    p0 = s2/size(traind,1); % P(class = 0)

    %the following cell array will have the vector of unique values for each column/feature

    v = cell(1,1);
    for c = 1:size(traind,2)-1
       v{c}=unique(traind(:,c)); %cell array stores vectors of unique values for each feature
       % v{1} = vector of unique values in column 1 and so on.
    end

    %the following matrix helps compute P(feature and class)
    %for every unique value of the feature, it stores a count for when the class is 1 

    f1 = zeros(4,5); %all sums need to be initialized to zero

    %the following matrix helps compute P(feature and not class)
    %for every unique value of the feature, it stores a count for when the class is 0 

    f2 = zeros(4,5); %all sums need to be initialized to zero

     col = 1;
      u = 1;
    while col<=size(traind,2)-1 
         for row = 1:size(traind,1)
            if traind(row,col) ==v{col}( u)
              if traind(row,5) == 1
                 f1(col,u) = f1(col,u) + 1;
             else if traind(row,5) == 0
                 f2(col,u) = f2(col,u) + 1;
            end %end if
        end %end if
    end %end for for all rows
    u = u + 1;
    if u>length(v{col})
    u=1;
    col = col + 1;
    end  %end of if   
    end %end of while
    %Here, we begin classification for each data point(row)
    pfgc = []; %empty vector
    %p(feature given class), pfgc(1) = p(feature and class)/p(class)
    %it does this for each row because each whole row represents a data point
    pfgnc = []; %empty vector
    %p(feature given not class), pfgnc(1) = p(feature and not class)/p(not class)
    %it does this for each row because each whole row represents a data point
    for row = 1:size(traind,1)
        for c = 1:size(traind,2)-1
            for u = 1: length(v{c})
          if traind(row,c) == v{c}(u)
           pfgc(c) = f1(c,u)/p1;
           pfgnc(c) = f2(c,u)/p0;
         end
         break 
    end 
    end 

%compute the final p(features given class) as a multiplication of all p(feature given class) values

    fpfgc = cumprod(pfgc);

%compute the final p(features given not class) as a multiplication of all p(feature given not class) values

    fpfgnc = cumprod(pfgnc);

    %use bayes theorem to compute p(class given feature). 

    pcgf = fpfgc * p1;
    pncgf = fpfgnc * p0;

%The right hand side has not been normalized (no denominator) as for both p(class given feature) and p(notclass given feature), the denominators are the same. For comparision, we need not compute them. It is sufficient to compare the numerators.

%Finally, determine class of datapoint

    if pcgf>pncgf
       disp(['The class of datapoint ' num2str(row) 'is' num2str(1)]);
    else
       disp(['The class of datapoint ' num2str(row) 'is' num2str(0)]);
    end 
    end %end of loop running through all rows
    %to test run the same program with testd instead of traind

Answer 1

[John D'Errico]

The crystal ball is so foggy. Don't just tell someone that it won't execute. Tell us what DOES happen when you try. If there is an error message, then show the ENTIRE message.

Would you go to your doctor and say only that something hurts, then refuse to say what hurts?

I'll guess that the very first line of your code fails, since you have given us absolutely no information otherwise. Is it possible that you don't have the file iris.dat?

[Kunal Roy]

I do have the file. Sorry about that. I figured out what the problem was. There was a missing 'end'. Will be sure to post the error message next time.