train.m

clc;
clear;
addpath('./libsvm-3.11/matlab/');
addpath('./svminfor/')
addpath('./code_coregspectral')
load('./data/mgae_cora.mat');

num_views=2;
numClust=length(unique(gnd));
noise = 0.4;    % corruption level, 0.4 is good for cora
layers = 3;     % number of layers to stack, 3 for cora 
projev = 1.5;

fprintf('======================================\n');
disp('create A_n');
%create A_bar
n = size(fea,1);
A_bar = W + speye(n);
d = sum(A_bar);
d_sqrt = 1.0./sqrt(d);
d_sqrt(d_sqrt == Inf) = 0;
DH = diag(d_sqrt);
DH = sparse(DH);
A_n = DH * sparse(A_bar) * DH;   %d*d
disp('compute A_n finished');

gcn = A_n * fea;   %d*d * d*n
[n,m] = size(gcn);

[allhx] = mSDA(fea', noise, layers, A_n);
Z0 = allhx(end-m+1:end,:);
Z1 = Z0' * Z0;
Z2 = (abs(Z1) + abs(Z1'))/2;

[V Eval F P R nmi_v avgent AR C,ACCb] = baseline_spectral_onRW(Z2,numClust,gnd,projev);
fprintf('Clustering Results on Z2: ACC=%f, nmi score=%f, F=%f, P=%f, R=%f, avgent=%f,  AR=%f, \n',ACCb(1), nmi_v(1),F(1),P(1),R(1),avgent(1),AR(1));
fprintf('======================================\n');