1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
|
% function [v,s] = quickNcutHardBiais(W,U,nbEigenValues,dataGraphCut)
%ligne 35 : changement tim
%[v,s] = ncut(W,nbEigenValues,[],offset);
%devient :
%[v,s] = tim_ncut_rapide(W,nbEigenValues,[],offset);
%et eigs devient tim_eigs
% Input:
% W = affinity matrix
% U = hard constraint matrix, could be a cell of partial grouping
% nbEigenValues = number of eigenvectors
% offset = regularization factor, default = 0
% Output:
% v = eigenvector
% s = eigenvalue of (W,d), s.t. U' * x = 0.
% call eigs using my own * operation
% Stella X. Yu, Jan 2002.
function [v,s] = quickNcutHardBiais2(W,U,nbEigenValues,dataGraphCut)%voir : rajouter sigma
n = size(W,1);
nbEigenValues = min(nbEigenValues,n);
offset = dataGraphCut.offset;
%offset = 2;
% degrees and regularization
d = sum(abs(W),2);
dr = 0.5 * (d - sum(W,2));
d = d + offset * 2;
dr = dr + offset;
W = W + spdiags(dr,0,n,n);
clear dr
% normalize
Dinvsqrt = 1./sqrt(d+eps);
P = spmtimesd(W,Dinvsqrt,Dinvsqrt);
clear W;
% if max(max(P-P')) < 1e-10 %ou eps
% %S = sparse(1:n,1:n,0.5);
% P =max(P,P');
% % P=S*(P+P');
% %P=0.5*(P+P');
% options.issym = 1;
% end
P = sparsifyc(P,dataGraphCut.valeurMin);
options.issym = 1;
Ubar = spmtimesd(U,Dinvsqrt,[]);
%Ubar = sparsifyc(Ubar,dataGraphCut.valeurMin); %voir
options.disp = dataGraphCut.verbose;
options.maxit = dataGraphCut.maxiterations;
options.tol = dataGraphCut.eigsErrorTolerance;
options.v0 = ones(size(P,1),1);%voir
options.p = max(35,2*nbEigenValues); %voir
options.p = min(options.p , n);
% nouvelle idee : factorisation de Cholesky
C=Ubar'*Ubar;
%permutation = symamd(C);
%R = cholinc(C(permutation,permutation));
t_chol_Ubar = cputime;
[R,ooo] = cholinc(C,'0');
t_chol_Ubar = cputime - t_chol_Ubar;
%if error occurs, check if Ubar = sparsifyc(Ubar,dataGraphCut.valeurMin);
%sparsifies too much
% compute H = (Ubar'*Ubar)^(-1)
% t_inv_H = cputime;
% H = inv(sparsifyc(Ubar' * Ubar,dataGraphCut.valeurMin)); %changer
% t_inv_H = cputime - t_inv_H;
% H = sparsifyc(H,dataGraphCut.valeurMin);
% tEigs = cputime;
% if options.issym & max(max(H-H')) < 1e-10
% [vbar,s,convergence] = tim_eigs(@mex_projection_inv_symmetric,n,nbEigenValues,'lm',options,triu(P),Ubar,triu(H));
% else
% [vbar,s,convergence] = tim_eigs(@mex_projection_inv,n,nbEigenValues,'lm',options,P,Ubar,H);
% end
% tEigs = cputime - tEigs;
%
R = sparsifyc(R,dataGraphCut.valeurMin);
tEigs = cputime;
if options.issym
[vbar,s,convergence] = tim_eigs(@mex_projection_QR_symmetric,n,nbEigenValues,'lm',options,tril(P),Ubar,R);
else
[vbar,s,convergence] = tim_eigs(@mex_projection_QR,n,nbEigenValues,'lm',options,P,Ubar,R);
end
tEigs = cputime - tEigs;
%afficheTexte(sprintf('\n\nTemps ecoule pendant eigs : %g',tEigs),dataGraphCut.verbose,2);
%afficheTexte(sprintf('\nTemps ecoule pendant chol(Ubar''*Ubar) : %g',t_chol_Ubar),dataGraphCut.verbose);
if convergence~=0
afficheTexte(sprintf(' (Non-convergence)'),dataGraphCut.verbose);
end
%disp(sprintf('nnz(P) : %f\n',nnz(P)));
%disp(sprintf('nnz(Ubar) : %f\n',nnz(Ubar)));
%disp(sprintf('nnz(R) : %f\n',nnz(R)));
%disp(sprintf('nnz(global) : %f\n',nnz(P) + 4 * nnz(Ubar) + 4*nnz(R)));
s = real(diag(s));
[x,y] = sort(-s);
s = -x;
vbar = vbar(:,y);
v = spdiags(Dinvsqrt,0,n,n) * vbar;
for i=1:size(v,2)
%v(:,i) = v(:,i) / max(abs(v(:,i)));
v(:,i) = (v(:,i) / norm(v(:,i)) )*norm(ones(n,1));
if v(1,i)~=0
v(:,i) = - v(:,i) * sign(v(1,i));
end
end
% % nouvelle idee : factorisation de Cholesky
% t_chol_Ubar = cputime;
% R = chol(Ubar' * Ubar);
% t_chol_Ubar = cputime - t_chol_Ubar;
% R = sparsifyc(R,dataGraphCut.valeurMin);
% tEigs = cputime;
% if options.issym
% [vbar,s,convergence] = tim_eigs(@mex_projection_QR_symmetric,n,nbEigenValues,'lm',options,triu(P),Ubar,R);
% else
% [vbar,s,convergence] = tim_eigs(@mex_projection_QR,n,nbEigenValues,'lm',options,P,Ubar,R);
% end
% tEigs = cputime - tEigs;
% % compute H = (Ubar'*Ubar)^(-1)
% t_inv_H = cputime;
% H = inv(sparsifyc(Ubar' * Ubar,dataGraphCut.valeurMin)); %changer
% t_inv_H = cputime - t_inv_H;
% H = sparsifyc(H,dataGraphCut.valeurMin);
% tEigs = cputime;
% if options.issym & max(max(H-H')) < 1e-10
% [vbar,s,convergence] = tim_eigs(@mex_projection_inv_symmetric,n,nbEigenValues,'lm',options,triu(P),Ubar,triu(H));
% else
% [vbar,s,convergence] = tim_eigs(@mex_projection_inv,n,nbEigenValues,'lm',options,P,Ubar,H);
% end
% tEigs = cputime - tEigs;
% % idee de mon rapport... semble pas marcher car R = qr(Ubar,0) est plus
% % lent que H = inv(sparsifyc(Ubar' * Ubar,dataGraphCut.valeurMin));
% t_qr_Ubar = cputime;
% R = qr(Ubar,0);
% t_qr_Ubar = cputime - t_qr_Ubar;
% R = sparsifyc(R,dataGraphCut.valeurMin);
% tEigs2 = cputime;
% if options.issym
% [vbar2,s2,convergence] = tim_eigs(@mex_projection_QR_symmetric,n,nbEigenValues,'lm',options,triu(P),Ubar,R);
% else
% [vbar2,s2,convergence] = tim_eigs(@mex_projection_QR,n,nbEigenValues,'lm',options,P,Ubar,R);
% end
% tEigs2 = cputime - tEigs2;
% idee de Jianbo... semble pas marcher car on a besoin de prendre k maximal
% dans [A,S,B] = svds(Ubar,k);
%
% [A,S,B] = svds(Ubar,300);
% A = sparsifyc(A,dataGraphCut.valeurMin);
% tEigs = cputime;
% [vbar,s,convergence] = tim_eigs(@mex_projection_svd,n,nbEigenValues,'lm',options,P,A);
% afficheTexte(sprintf('\ninv(H) : %g',t_inv_H),dataGraphCut.verbose);
% afficheTexte(sprintf('\n\nTemps ecoule pendant eigs : %g',tEigs2),dataGraphCut.verbose,2);
% afficheTexte(sprintf('\nqr(Ubar) : %g',t_qr_Ubar),dataGraphCut.verbose);
% disp(sprintf('nnz(H) : %f\n',nnz(H)));
% disp(sprintf('nnz(global) : %f\n',nnz(P) + 4 * nnz(Ubar) + 2*nnz(H)));
|
