├── README.md
└── SpectralClustering.m


/README.md:
--------------------------------------------------------------------------------
1 | SpectralClustering
2 | ==================
3 | 
4 | Spectral Clustering based on Random Walk (Normalized Laplacian Matrix)
5 | 


--------------------------------------------------------------------------------
/SpectralClustering.m:
--------------------------------------------------------------------------------
 1 | %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 2 | %Spectral Clustering based on Random Walk (Normalized laplacian Matrix)
 3 | %Input : 
 4 | %affinity : n*n matrix define the similarities between each instance in the
 5 | %           data
 6 | %K (optional) : K Expected clusters. (If it's not provided, clustering will
 7 | %               be based on Eigenvalues
 8 | %
 9 | %Output : 
10 | %KEigenVectors : The K eigenvectors that can be used to cluster all the
11 | %data
12 | %For example you can use it this way:
13 | %   KEigenVectors = SpectralClustering(affinity,K)
14 | %   [IDX,C] = kmeans(KEigenVectors,K)
15 | %   %Where IDX contains the cluster indices of each instance. 
16 | 
17 | function [KEigenVectors]= SpectralClustering(affinity,K)
18 | 
19 | sz = size(affinity,1);
20 | D = zeros(sz,sz);
21 | for i=1:sz
22 |     D(i,i) = sum(affinity(i,:));
23 | end
24 | 
25 | L = D - affinity;
26 | [eigVec, eigVal] = eig( inv(D) * L);
27 | 
28 | cEigVal =  circshift(diag(eigVal),1);
29 | [~, idx] = max(cEigVal - diag(eigVal));
30 | if ~exist('K','var')
31 |     KEigenVectors = eigVec(:,idx:size(eigVec,1)-1);
32 | else
33 |     KEigenVectors = eigVec(:,size(eigVec,1)-K:size(eigVec,1)-1);
34 | end
35 | end
36 | 


--------------------------------------------------------------------------------