SNF_all.m

function [W]=SNF_all(Wall,K,t,ALPHA)
%
if nargin < 2
    K = 20;
end
if nargin < 3
    K = 20;
end

if nargin < 4
    ALPHA = 1;
end
C = length(Wall);
[m,n]=size(Wall{1});


tracenorm = 1;
lambda=0.5;
for i = 1 : C
    Wall_d=diag(degrees(Wall{i})'); % degree matrix of subject i
    
    Wall_eig=diag(eigenCentrality(Wall{i})); % eigencentrality matrix of subject i
    
    Wall_cl=diag(closeness(Wall{i})); % closeness matrix of subject i
    
    Wall_topo=cat(3,Wall_d,(Wall_eig),(Wall_cl)); % Tensor stacking the topological measurements
    
    Wall_topolo = permute(Wall_topo,[3 2 1]);
    
    Average= mean(Wall_topolo) ; % Average topological matrix of subject i
    Average_all(i,:,:)=Average;
end
%% Supervised multiple kernel learning

[model] = easymkl_train(Average_all,t,lambda, tracenorm);
%% Normalisation of the global matrix
for i = 1 : C
    Wall_toponorm=model.weights(i)* squeeze(Average_all(i,:,:));
    
    Wall{i} = Wall{i}./repmat(diag(Wall_toponorm) ,1,n); % Normalized global topology matrix
    
    Wall{i} = (Wall{i} + Wall{i}')/2;
end
%% sparse matrix
for i = 1 : C
    newW{i} = FindDominateSet(Wall{i},round(K));
end

Wsum = zeros(m,n);
for i = 1 : C
    Wsum = Wsum + Wall{i};
end
for ITER=1:t
    for i = 1 : C
        %Wall0{i}=newW{i}*(0.95*(Wsum - Wall{i})/(C-1)+0.05*eye(length(Wsum)))*newW{i}';
        Wall0{i}=newW{i}'*(Wsum - Wall{i})/(C-1);
    end
    for i = 1 : C
        Wall{i} = BOnormalized(Wall0{i},ALPHA);
    end
    Wsum = zeros(m,n);
    for i = 1 : C
        Wsum = Wsum + Wall{i};
    end
    %
end

W = Wsum/C;
W = W./repmat(sum(W,2),1,n);
W = (W +W'+eye(n))/2;

    function W = BOnormalized(W,ALPHA)
        if nargin < 2
            ALPHA = 1;
        end
        W = W+ALPHA*eye(length(W));
        W = (W +W')/2;
    end

    function newW = FindDominateSet(W,K)
        [m,n]=size(W);
        [YW,IW1] = sort(W,2,'descend');
        clear YW;
        newW=zeros(m,n);
        temp=repmat((1:n)',1,K);
        I1=(IW1(:,1:K)-1)*m+temp;
        newW(I1(:))=W(I1(:));
        newW=newW./repmat(sum(newW,2),1,n);
        clear IW1;
        clear IW2;
        clear temp;
    end




end