SAMI/MissingNodes_SrcCode2014/CalculateAffinityByAdamicAdar_S3o.m

function [ affinity ] = CalculateAffinityByAdamicAdar_S3o( data, actual_graph_size, num_missing_nodes, num_attr_nodes, attWeight, connectPHsToNeighbors, addMissingAtt)
%UNTITLED2 adamic/adar(i,j) = sum( 1 / log(num_neighbors(k) | k is a
%neighbor of both i and j
%   Detailed explanation goes here

fprintf('CalculateAffinityByAdamicAdar_Sparse: NNZ(data) = %d\n', nnz(data));
%n = size(data,1);
%diag_indices = 1:n+1:n*n;
%data(diag_indices) = 1; %we consider each node as a neighbor of itself to obtain higher connectivity in the affinity matrix - i.e. each node will have a positive affinity to all its neighbors
if connectPHsToNeighbors > 0
      data = ConnectMissingNodesToNeighborsNeighbors( data, actual_graph_size, num_missing_nodes, num_attr_nodes, addMissingAtt );
end
affinity = sparse(size(data,1), size(data,2));

% Sigal 5.3.13: normalizeWeight
% make sure that ratio att/link is same w/1-w
if attWeight < 1 && attWeight >= 0
    normalizeWeight = attWeight/(1-attWeight);
else
    normalizeWeight = 1;
end

numNodes = size(data,1);

for k = 1 :numNodes
    %find all the neighbors of k
    neighbors_k_vec = data(k,:);
    neighbors_k_vec(k) = 1;%we consider each node as a neighbor of itself to obtain higher connectivity in the affinity matrix - i.e. each node will have a positive affinity to all its neighbors
    neighbors_k = find(neighbors_k_vec);
    num_neighbors_k = size(neighbors_k, 2);
    
    %sigal - 17.12.12 - find # social neighbors of k
    %sigal - update 23.3.13
    if num_attr_nodes>0
        neighbors_k_vec_social = data(k,:);
        neighbors_k_att = find(neighbors_k_vec_social(1:num_attr_nodes));
        neighbors_k_vec_social(1:num_attr_nodes)=0;
        neighbors_k_vec_social(k) = 1;
        %neighbors_k_social2 = find(neighbors_k_vec_social); %% temp !!!
        neighbors_k_social = num_attr_nodes+find(neighbors_k_vec_social(1+num_attr_nodes:end));
        num_neighbors_k_social = size(neighbors_k_social, 2);
    else
        num_neighbors_k_social = num_neighbors_k;
    end
    
    
    if num_neighbors_k_social > 1 && num_neighbors_k > 1
        if k>num_attr_nodes
            w = 1;
        else
            w = normalizeWeight;
        end
        inv_log_num_neighbors_k = w / log(num_neighbors_k);
        inv_log_num_neighbors_k_social = w / log(num_neighbors_k_social);
        if num_attr_nodes==0
            affinity(neighbors_k,neighbors_k) = affinity(neighbors_k,neighbors_k)+ inv_log_num_neighbors_k_social;
        else
            affinity(neighbors_k_social,neighbors_k_social) = affinity(neighbors_k_social,neighbors_k_social)+ inv_log_num_neighbors_k_social;
            if k>num_attr_nodes
                affinity(neighbors_k_att,neighbors_k_social) = affinity(neighbors_k_att,neighbors_k_social) + inv_log_num_neighbors_k;
                affinity(neighbors_k_social,neighbors_k_att) = affinity(neighbors_k_social,neighbors_k_att) + inv_log_num_neighbors_k;
                affinity(neighbors_k_att,neighbors_k_att) = affinity(neighbors_k_att,neighbors_k_att) + inv_log_num_neighbors_k;
            end
        end
    end
end

%normalize the matrix
Inf_ind = find(affinity==Inf);
affinity(Inf_ind) = 1;
affinity = affinity ./ max(max(affinity));
affinity(Inf_ind) = 1;

%filter low values
% thIn = 0.15;
% thFactor = 100; %% 50*normalizeWeight
% if num_attr_nodes>0
%     threshold = thIn/thFactor;
%     ind = affinity<threshold;
%     fprintf('CalculateAffinityByAdamicAdar_Sparse: NNZ(affinity) = %d (before threshold %.4f)\n', nnz(affinity), threshold);
%     affinity(ind) = 0;
% end

fprintf('CalculateAffinityByAdamicAdar_Sparse: NNZ(affinity) = %d\n', nnz(affinity));

end