Subscript indices must either be real positive integers or logicals.

Hi everyone, I am trying to implement the K-Nearest Neighboors (K-NN) function and I have trouble understanding the error. I have looked at the forum of Mathworks already but didn't find answer of my problem. I tried to cast "idx(:)" to int16 but it still didn't work. Can anyone help me fix this problem and help me understand the reason why it doesn't work? Thanks!

   function [ y_est ] =  my_knn(X_train,  y_train, X_test, k, type)
  %MY_KNN Implementation of the k-nearest neighbor algorithm
  %   for classification.
  %
  %   input -----------------------------------------------------------------
  %   
  %       o X_train  : (N x M_train), a data set with M_test samples each being of dimension N.
  %                           each column corresponds to a datapoint
  %       o y_train  : (1 x M_train), a vector with labels y \in {0,1} corresponding to X_train.
  %       o X_test   : (N x M_test), a data set with M_test samples each being of dimension N.
  %                           each column corresponds to a datapoint
  %       o k        : number of 'k' nearest neighbors
  %       o type   : (string), type of distance {'L1','L2','LInf'}
  %
  %   output ----------------------------------------------------------------
  %
  %       o y_est   : (1 x M_test), a vector with estimated labels y \in {0,1} 
  %                   corresponding to X_test.
  %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

[N,M_train]=size(X_train);
[N,M_test]=size(X_test);
d=zeros(M_test,M_train);
y_est=zeros(1,M_test);
B=zeros(1,k);

for test=1:M_test
    for train=1:M_train
        d(test,train)=my_distance(X_train(:,train),X_test(:,test), 'L2');
    end
    %find the k minimum value for each line
    val=zeros(k);
    idx=zeros(k);
    for i=1:k
        [val(i),idx(i)]=min(d(test,:));
        d(test,idx(i))=max(d(test,:));
    end
    y_est(1,test)=mode(y_train(idx(:))); 
end

end

Here is the error that still appear:

_Subscript indices must either be real positive integers or logicals.

Error in my_knn (line 38) y_est(1,test)=mode(y_train(idx(:)));_

Answer 1

[Image Analyst]

Use the debugger to step through and find out what the values of idx are. They can't be 0, negative, or fractional numbers. See the FAQ: http://matlab.wikia.com/wiki/FAQ#Why_is_0.3_-_0.2_-_0.1_.28or_similar.29_not_equal_to_zero.3F

[Cyril SCHMITT]

Hey Image Analyst, Your comment actually helped me solve the problem! It appears that my "idx" matrix was initialized to zeros and was a square matrix but I only needed a vector. So I just selected the vector I wa sinterested into and it fixed the problem! All the best! Cyril