Automation of the finding size and the number of variables which depends on that size

for i=1:numDP
    data.T{i}=eye(DIM);
    nearest=knnsearch(data.DataTree,P(i,:),'K',K+1);
    R{i}=zeros(DIM);
    for j=1:K
       v=P(i,:)-P(nearest(j+1),:);
       V11{i}=P(i,:);
       V22{i}=P(nearest(j+1),:);
       v=v';
       s=norm(v);
       B=exp(-s^2/sigma^2)*(eye(DIM)-v*v'/norm(v'*v)); 
       R{i}=R{i}+B;
       [data.T{i},data.D{i}]=eig(R{i});
       n1{i}=data.T{i}(:,1);
       N2{i}=data.T{i}(:,2);
       n3{i}=data.T{i}(:,3);
    
       lambda1=max(diag(data.D{i}));
       lambda3=min(diag(data.D{i}));
       lambda2=data.D{i}(2,2);

'My question is for example the number of 'lambda' and 'n' depends on 'DIM'. But as you can see this code works just for DIM=3; I need to generalize that code for every dimensions and my Matlab knowledge is not enough to do as a beginner. I put example to make it clear

[Walter Roberson]

Replace

       n1{i}=data.T{i}(:,1);
       N2{i}=data.T{i}(:,2);
       n3{i}=data.T{i}(:,3);

with

       c = size(data.T{i},2);
       for K = 1 : c
           N{c}{i} = data.T{i}(:,K);
       end

... But it doesn't look like a useful memory organization to me. Why not just store all of data.T{i} ?

[Zeynep Naz Kocabas]

Why is not useful ? I use these variables on another code so I coded them inside data.T and just put another functions, data as a input. If it is your question?

Answer 1

[Mahesh Taparia]

Hi

The function ‘knnsearch’ can work for N dimensional data. By looking at your code, it seems the value of DIM will be the number of elements in P(i,:), if you want to increase the DIM, change the data accordingly.

[Zeynep Naz Kocabas]

Yes but the thing is I need to programme without using 'knnsearch', while using matlabcoder it doesn't work

[Mahesh Taparia]

Hi,

The function 'knnsearch' is optimized one. In order to not use this function, you can create a custom function which can replicate the same output.

[Zeynep Naz Kocabas]

Sorry I did not understand what you mean, I do not know Matlab very well.

[Mahesh Taparia]

Hi

The function 'knnsearch' returns the indices of the nearest neighbors of the test points in your dataset. Since you doesn't want to use this function, you can create a new function which will return the same indices. For example, finding euclidean distance of test points from all the datapoints points, sort them and store the indices.

You can refer to this documentation for creating a custom function.

[Walter Roberson]

knnsearch is supported by MATLAB Coder

https://www.mathworks.com/help/stats/knnsearch.html

[Zeynep Naz Kocabas]

Thanks for your help I did that by checking 'knnsearch'.