demo.m

% Compare extracted subgraphs by greedy and adaptive greedy on Schizophrenia data
% Extracted subgraphs are further refined to bipartite by Pard
% Lambda in adaptive greedy is selected by itegrated likelihood

%% Print the raw data and do permutation
addpath('C:/Dropbox/dense');
load('edge_wise_res.mat');
figure;imagesc(W);colormap jet;colorbar;snapnow;


n=size(W,1);
perm_matrix = squareform(1:(n*(n-1)/2));
node_perm_idx = randperm(n);
perm_matrix = perm_matrix(node_perm_idx,node_perm_idx);
perm_vec = squareform(perm_matrix);
[result ID]=sort(perm_vec);
W_vec = squareform(W);
W1_vec = W_vec(perm_vec);
W1 = squareform(W1_vec);
figure;imagesc(W1);colormap jet;colorbar;snapnow

%% Use one step greedy_ad with parameter 1

[inlist outlist]=greedy_ad(W1,0.05,1);
W1_return = W1([inlist outlist],[inlist outlist]);
figure;imagesc(W1_return);colormap jet;colorbar;


%% Select lambda via integrated likelihood for binarized data with varied cutoff
% then use greedy_ad and kpartite
r=[0.01 0.005 0.001];
rpmf0 = [];
for i = 1:size(r,2)
   rpmf0(i) = 1/sum(W1(:)>-log(r(i)));
end
rpmf = rpmf0/sum(rpmf0);
p0=0.05;
lambda0=0.8:0.05:1.6;
[inlist2 outlist2 max_lambda] = greedy_ad_blik(W1,lambda0, r, rpmf, p0);
W1_return2 = W1([inlist2 outlist2],[inlist2 outlist2]);
figure;imagesc(W1_return2);colormap jet;colorbar;