pdist and squareform

[Peter Hinow]

Hello all,

the matlab functions pdist and squareform (from the statistics toolbox) are missing in scilab. So I wrote them myself and just want to know if the community has any use for it. The codes are pasted below. pdist is written with only the Euclidean distance in mind, I guess one can generalize it with an optional argument that specifies other metrics.

Best regards,
      Peter


// this function computes the pairwise distances in the vector x 

function [r] = pdist(x)
[rows,columns] =  size(x);
if rows == 1
    x = x'; swap = rows; rows = columns; columns = swap;
end
r = zeros(rows*(rows-1)/2,1);
start = 1;
for k = 1:rows-1
    mat1 = kron(x(k,:),ones(rows-k,1)); mat2 = x(k+1:rows,:);
    r(start:start+rows-k-1) = sqrt(sum((mat1-mat2).^2,2));
    start = start+rows-k;
end
endfunction



// this function rearranges a vector of length n*(n-1)/2 in squareform (with zero diagonal)

function [r] = squareform(x)
[rows,columns] =  size(x);
if rows == 1
    x = x'; swap = rows; rows = columns; columns = swap;
end
m = (1+sqrt(1+8*rows))/2;
if abs(int(m)-m)>0.0001
    disp('error: wrong dimension')
    r=[-1];
    return;
end
r=zeros(m,m);
start = 1;
for k = 1:m-1
    r(k+1:m,k) = x(start:start+m-k-1); r(k,k+1:m) = x(start:start+m-k-1)';
    start = start+m-k;
end
endfunction