MATLAB som

% 1 4 5 7 -- sample 1
% 2 5 6 6 -- sample 2

% in cndx = you will get  the position of each training sample in cluster no
% as SOM takes each sample as a column we need to transpose feature before applying on SOM

%  -------  train -----------

trainFeature = [ 1 2  ; 4 5    ; 5 6 ; 7 6 ]; % Two training data with 4 feature each . So each column is a sample
net = newsom(  trainFeature'  ,[2 2] ); % 4 cluster
net.trainParam.epochs = 100;
[net,tr,Y,E,Pf,Af] = train(net, trainFeature'  );
distances = dist( trainFeature  ,net.IW{1}' );
[d,cndx_train ] = min(distances,[],2); % cndx gives the cluster index, d gives the distance
cndx_train % the position of each training sample in cluster no


bookKeep=zeros(SOMd1 * SOMd2 ,2);%one column for +ve,one column for -ve
    for c=1: totTrain
        index = cndx_train(c);
        if(c<=totPosTrain)
            bookKeep( index,1) = bookKeep( index,1) + 1; %+ve sample
        else
            bookKeep( index,2) = bookKeep( index,2) + 1; %-ve sample
        end
    end
    dlmwrite('clusterInfo',bookKeep,'\t');

% ---------  test  -------------

out = sim(net, featureInTest' );
% format: clusterNo*testSample : 0 0 0 1 0 0 0 0 % that means 4th sample in this custer;
% outFormat (clusterNo, serial )

     out = out'; % #testSample * clusterNo 
    dlmwrite('out',out,'\t');
    for c=1:totTest
      
        clusterNo(c) = find( out(c ,:) );     
        if c<=totPosTest
            origLabel(c) = 1;
        else
            origLabel(c) = -1;
        end
      
    end
    dlmwrite('predicted',[origLabel' clusterNo'],'\t');