[10] | 1 | %EXPDISTM Exponential-type of Distance Matrix |
---|
| 2 | % |
---|
| 3 | % D = EXPDISTM (A,B,R) |
---|
| 4 | % or |
---|
| 5 | % D = EXPDISTM (A,B) |
---|
| 6 | % or |
---|
| 7 | % D = EXPDISTM (A,R) |
---|
| 8 | % or |
---|
| 9 | % D = EXPDISTM (A) |
---|
| 10 | % |
---|
| 11 | % INPUT |
---|
| 12 | % A NxK Matrix or dataset |
---|
| 13 | % B MxK Matrix or dataset (optional; default: B=A) |
---|
| 14 | % R Parameter to scale the Gaussian function |
---|
| 15 | % (optional; default: sqrt(N)*(max(max(A))- min(min(A)))) |
---|
| 16 | % |
---|
| 17 | % OUTPUT |
---|
| 18 | % D NxM Dissimilarity matrix or dataset |
---|
| 19 | % |
---|
| 20 | % DESCRIPTION |
---|
| 21 | % Computes the distance matrix D between two sets of vectors, A and B. |
---|
| 22 | % Given the vectors X and Y, distances are computed as: |
---|
| 23 | % D(X,Y) = 1 - exp (-(X-Y)'(X-Y)/R^2) |
---|
| 24 | % |
---|
| 25 | % If A and B are datasets, then D is a dataset as well with the labels defined |
---|
| 26 | % by the labels of A and the feature labels defined by the labels of B. If A is |
---|
| 27 | % not a dataset, but a matrix of doubles, then D is also a matrix of doubles. |
---|
| 28 | % |
---|
| 29 | % SEE ALSO |
---|
| 30 | % SIMDISTM, JACSIMDISTM, COSDISTM, CORRDISTM, LPDISTM, EUDISTM |
---|
| 31 | |
---|
| 32 | % Copyright: Elzbieta Pekalska, ela.pekalska@googlemail.com |
---|
| 33 | % Faculty EWI, Delft University of Technology and |
---|
| 34 | % School of Computer Science, University of Manchester |
---|
| 35 | |
---|
| 36 | |
---|
| 37 | function [D,r] = expdistm(A,B,r) |
---|
| 38 | [ra,ca] = size(A); |
---|
| 39 | |
---|
| 40 | bisa = 0; |
---|
| 41 | if nargin < 2, |
---|
| 42 | bisa = 1; |
---|
| 43 | r = (max(max(+A)) - min(min(+A)))*sqrt(ra); |
---|
| 44 | B = A; |
---|
| 45 | [rb,cb] = size(B); |
---|
| 46 | else |
---|
| 47 | [rb,cb] = size(B); |
---|
| 48 | |
---|
| 49 | if nargin < 3, |
---|
| 50 | if max (rb,cb) == 1, |
---|
| 51 | bisa = 1; |
---|
| 52 | r = B; |
---|
| 53 | B = A; |
---|
| 54 | [rb,cb] = size(B); |
---|
| 55 | else |
---|
| 56 | r = (max(max(+A)) - min(min(+A)))*sqrt(ra); |
---|
| 57 | end |
---|
| 58 | end |
---|
| 59 | end |
---|
| 60 | |
---|
| 61 | |
---|
| 62 | if ~bisa & ca ~= cb, |
---|
| 63 | error ('The matrices should have the same number of columns.'); |
---|
| 64 | end |
---|
| 65 | |
---|
| 66 | isda = isa(A,'dataset'); |
---|
| 67 | isdb = isa(B,'dataset'); |
---|
| 68 | a = +A; |
---|
| 69 | b = +B; |
---|
| 70 | |
---|
| 71 | |
---|
| 72 | %d = zeros(ra,rb); |
---|
| 73 | %d = sum ((abs (repmat (permute(a,[1 3 2]), [1 rb 1]) - ... |
---|
| 74 | % repmat (permute(b,[3 1 2]), [ra 1 1]))).^2,3); |
---|
| 75 | |
---|
| 76 | D = 1 - exp (-distm(a,b)/r^2); |
---|
| 77 | D(find(D < eps)) = 0; |
---|
| 78 | |
---|
| 79 | |
---|
| 80 | % Set object labels and feature labels |
---|
| 81 | if xor(isda, isdb), |
---|
| 82 | prwarning(1,'One matrix is a dataset and the other not. ') |
---|
| 83 | end |
---|
| 84 | if isda, |
---|
| 85 | if isdb, |
---|
| 86 | D = setdata(A,D,getlab(B)); |
---|
| 87 | else |
---|
| 88 | D = setdata(A,D); |
---|
| 89 | end |
---|
| 90 | D.name = 'Distance matrix'; |
---|
| 91 | if ~isempty(A.name) |
---|
| 92 | D.name = [D.name ' for ' A.name]; |
---|
| 93 | end |
---|
| 94 | end |
---|
| 95 | return |
---|