-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathplot4EXP.m
239 lines (229 loc) · 9.21 KB
/
plot4EXP.m
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
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
function plot4EXP(idx,EXP)
%% plot from EXP.mat
% if EXP is given, as called in the main_exp function, use it,
% else as called alone, load EXP from the 'result' folder
close all;
% dir to save the plots
saveDir = ['./img/EXP_idx_' num2str(idx) '/'];
if ~exist(saveDir,'dir')
mkdir(saveDir);
end
if nargin ==1
dir = ['./result/EXP_idx_' num2str(idx) '/EXP.mat'];
load(dir);
end
%% plot
% see if plot the matrix
if EXP.isplot == 1
figure(1),clf;
h1 = imagesc(EXP.A);
colorbar; hold all;
title('Matrix A');
set(gca,'fontsize',15);
saveas(h1,[EXP.output_dir 'figure1.png']);
figure(2),clf;
h2 = imagesc(EXP.B);
colorbar; hold all;
title('Matrix B');
set(gca,'fontsize',15);
saveas(h2,[EXP.output_dir 'figure2.png']);
figure(3),clf;
h3 = imagesc(EXP.C);
colorbar; hold all;
title('Matrix C');
set(gca,'fontsize',15);
saveas(h3,[EXP.output_dir 'figure3.png']);
end
%% if plot the convergence
% first choose the index of experiment with
% %
% mean of iters %
% %
% number of epochs for each experiment with mat A and CBCD1
% is read from EXP.epoch1
% sort the number of epochs and choose the middle one
[~,I] = sort(EXP.epoch1(1:EXP.n_loop));
index = I(ceil(EXP.n_loop/2));
%%
% then plot the cooresponding curve of convergence
figure(4),clf;
% in figure 4 we show the convergence of the KKT condition
% which is based on the normal cone
% plot matrix A
semilogy(0:size(EXP.KKT_A{index},1)-1,...
EXP.KKT_A{index},'r','LineWidth',2.5);hold on;
semilogy(0:size(EXP.KKT_A{index+EXP.n_loop},1)-1,...
EXP.KKT_A{index+EXP.n_loop},'g','LineWidth',2.5);hold on;
semilogy(0:size(EXP.KKT_A{index+EXP.n_loop*2},1)-1,...
EXP.KKT_A{index+EXP.n_loop*2},'b','LineWidth',2.5);hold on;
% plot matrix A of RBCD
semilogy(0:size(EXP.KKTr_A{index},1)-1,...
EXP.KKTr_A{index},'r--','LineWidth',2.5);hold on;
semilogy(0:size(EXP.KKTr_A{index+EXP.n_loop},1)-1,...
EXP.KKTr_A{index+EXP.n_loop},'g--','LineWidth',2);hold on;
semilogy(0:size(EXP.KKTr_A{index+EXP.n_loop*2},1)-1,...
EXP.KKTr_A{index+EXP.n_loop*2},'b--','LineWidth',1.5);hold on;
% legend of A: CBCD & RBCD
% first add runtime into legend
l1=sprintf('CBCD1, #%d',...%.4f s, #%d',mean(EXP.T_c(1,:)),...
size(EXP.KKT_A{index},1)-1);
l2=sprintf('CBCD2, #%d',...%.4f s, #%d',mean(EXP.T_c(2,:)),...
size(EXP.KKT_A{index+EXP.n_loop},1)-1);
l3=sprintf('CBCD3, #%d',...%.4f s, #%d',mean(EXP.T_c(3,:)),...
size(EXP.KKT_A{index+EXP.n_loop*2},1)-1);
l1r=sprintf('RBCD1, #%d',...%.4f s, #%d',mean(EXP.T_r(1,:)),...
size(EXP.KKTr_A{index},1)-1);
l2r=sprintf('RBCD2, #%d',...%.4f s, #%d',mean(EXP.T_r(2,:)),...
size(EXP.KKTr_A{index+EXP.n_loop},1)-1);
l3r=sprintf('RBCD3, #%d',...%.4f s, #%d',mean(EXP.T_r(3,:)),...
size(EXP.KKTr_A{index+EXP.n_loop*2},1)-1);
legend(l1,l2,l3,l1r,l2r,l3r);
xlabel('#epoch');ylabel('KKT Condition');
title(['Convergence Speed A; #EXP ' num2str(idx) '; #' num2str(index) ]);
set(gca,'fontsize',15);
saveas(gca,[saveDir 'figure4.png']);
%%
figure(5),clf;
% in figure 5 we show the convergence of the KKT condition
% which is based on the normal cone
% plot matrix B
semilogy(0:size(EXP.KKT_B{index},1)-1,...
EXP.KKT_B{index},'r','LineWidth',2.5);hold on;
semilogy(0:size(EXP.KKT_B{index+EXP.n_loop},1)-1,...
EXP.KKT_B{index+EXP.n_loop},'g','LineWidth',2.5);hold on;
semilogy(0:size(EXP.KKT_B{index+EXP.n_loop*2},1)-1,...
EXP.KKT_B{index+EXP.n_loop*2},'b','LineWidth',2.5);hold on;
% plot matrix B of RBCD
semilogy(0:size(EXP.KKTr_B{index},1)-1,...
EXP.KKTr_B{index},'r--','LineWidth',2.5);hold on;
semilogy(0:size(EXP.KKTr_B{index+EXP.n_loop},1)-1,...
EXP.KKTr_B{index+EXP.n_loop},'g--','LineWidth',2);hold on;
semilogy(0:size(EXP.KKTr_B{index+EXP.n_loop*2},1)-1,...
EXP.KKTr_B{index+EXP.n_loop*2},'b--','LineWidth',1.5);hold on;
% legend of B; CBCD & RBCD
% first add runtime into legend
l4 =sprintf('CBCD1p, #%d',...%.4f s, #%d',mean(EXP.T_c(4,:)),...
size(EXP.KKT_B{index},1)-1);
l5 =sprintf('CBCD2p, #%d',...%.4f s, #%d',mean(EXP.T_c(5,:)),...
size(EXP.KKT_B{index+EXP.n_loop},1)-1);
l6 =sprintf('CBCD3p, #%d',...%.4f s, #%d',mean(EXP.T_c(6,:)),...
size(EXP.KKT_B{index+EXP.n_loop*2},1)-1);
l4r=sprintf('RBCD1p, #%d',...%.4f s, #%d',mean(EXP.T_r(4,:)),...
size(EXP.KKTr_B{index},1)-1);
l5r=sprintf('RBCD2p, #%d',...%.4f s, #%d',mean(EXP.T_r(5,:)),...
size(EXP.KKTr_B{index+EXP.n_loop},1)-1);
l6r=sprintf('RBCD3p, #%d',...%.4f s, #%d',mean(EXP.T_r(6,:)),...
size(EXP.KKTr_B{index+EXP.n_loop*2},1)-1);
legend(l4,l5,l6,l4r,l5r,l6r);
xlabel('#epoch');ylabel('KKT Condition');
title(['Convergence Speed B; #EXP ' num2str(idx) '; #' num2str(index) ]);
set(gca,'fontsize',15);
saveas(gca,[saveDir 'figure5.png']);
%%
figure(6),clf;
% in figure 6 we show the convergence of the KKT condition
% which is based on the normal cone
% plot matrix C, first cyclic and then random
semilogy(0:size(EXP.KKT_C{index},1)-1,...
EXP.KKT_C{index},'r','LineWidth',2.5);hold on;
semilogy(0:size(EXP.KKT_C{index+EXP.n_loop},1)-1,...
EXP.KKT_C{index+EXP.n_loop},'g','LineWidth',2.5);hold on;
semilogy(0:size(EXP.KKT_C{index+EXP.n_loop*2},1)-1,...
EXP.KKT_C{index+EXP.n_loop*2},'b','LineWidth',2.5);hold on;
semilogy(0:size(EXP.KKTr_C{index},1)-1,...
EXP.KKTr_C{index},'r--','LineWidth',2.5);hold on;
semilogy(0:size(EXP.KKTr_C{index+EXP.n_loop},1)-1,...
EXP.KKTr_C{index+EXP.n_loop},'g--','LineWidth',2.5);hold on;
semilogy(0:size(EXP.KKTr_C{index+EXP.n_loop*2},1)-1,...
EXP.KKTr_C{index+EXP.n_loop*2},'b--','LineWidth',2.5);hold on;
% legend of C
% first add runtime into legend
l7 =sprintf('CBCD1r, #%d',...%.4f s, #%d',mean(EXP.T_c(7,:)),...
size(EXP.KKT_C{index},1)-1);
l8 =sprintf('CBCD2r, #%d',...%.4f s, #%d',mean(EXP.T_c(8,:)),...
size(EXP.KKT_C{index+EXP.n_loop},1)-1);
l9 =sprintf('CBCD3r, #%d',...%.4f s, #%d',mean(EXP.T_c(9,:)),...
size(EXP.KKT_C{index+EXP.n_loop*2},1)-1);
l7r=sprintf('RBCD1r, #%d',...%.4f s, #%d',mean(EXP.T_r(7,:)),...
size(EXP.KKTr_C{index},1)-1);
l8r=sprintf('RBCD2r, #%d',...%.4f s, #%d',mean(EXP.T_r(8,:)),...
size(EXP.KKTr_C{index+EXP.n_loop},1)-1);
l9r=sprintf('RBCD3r, #%d',...%.4f s, #%d',mean(EXP.T_r(9,:)),...
size(EXP.KKTr_C{index+EXP.n_loop*2},1)-1);
legend(l7,l8,l9,l7r,l8r,l9r);
xlabel('#epoch');ylabel('KKT Condition');
RCMstr = sprintf(';Reorder=%.4fs',EXP.tRCM);
title(['Convergence Speed C; #EXP ' num2str(idx) '; #' num2str(index) ]);%RCMstr]);
set(gca,'fontsize',15);
saveas(gca,[saveDir 'figure6.png']);
%%
% first get all the function values according to the index
% also get the min of f(x) from these values
fx1a = EXP.OBJ_A{index};%fx1a = fx1a(1:nnz(fx1a)+1);
fx2a = EXP.OBJ_A{index+EXP.n_loop};%fx2a = fx2a(1:nnz(fx2a)+1);
fx3a = EXP.OBJ_A{index+EXP.n_loop*2};%fx3a = fx3a(1:nnz(fx3a)+1);
fx1b = EXP.OBJ_B{index};%fx1b = fx1b(1:nnz(fx1b)+1);
fx2b = EXP.OBJ_B{index+EXP.n_loop};%fx2b = fx2b(1:nnz(fx2b)+1);
fx3b = EXP.OBJ_B{index+EXP.n_loop*2};%fx3b = fx3b(1:nnz(fx3b)+1);
fx1c = EXP.OBJ_C{index};%fx1c = fx1c(1:nnz(fx1c)+1);
fx2c = EXP.OBJ_C{index+EXP.n_loop};%fx2c = fx2c(1:nnz(fx2c)+1);
fx3c = EXP.OBJ_C{index+EXP.n_loop*2};%fx3c = fx3c(1:nnz(fx3c)+1);
fmin = min([fx1a;fx2a;fx3a;fx1b;fx2b;fx3b;fx1c;fx2c;fx3c]);
%% then plot function value for matrix A
figure(7),clf;
% in figure 7 we show the convergence of the Objective
% which is the function value f(x^k)
% plot matrix A
semilogy(0:size(fx1a,1)-1,fx1a-fmin,...
'r','LineWidth',2.5);hold on;
semilogy(0:size(fx2a,1)-1,fx2a-fmin,...
'g','LineWidth',2.5);hold on;
semilogy(0:size(fx3a,1)-1,fx3a-fmin,...
'b','LineWidth',2.5);hold on;
% legend of A & B
l1=sprintf('CBCD1, #%d',size(fx1a,1)-1);
l2=sprintf('CBCD2, #%d',size(fx2a,1)-1);
l3=sprintf('CBCD3, #%d',size(fx3a,1)-1);
legend(l1,l2,l3);
xlabel('#epoch');ylabel('\phi(x^t)-p^*');
title(['Convergence Speed A; #EXP ' num2str(idx) '; #' num2str(index) ]);
set(gca,'fontsize',15);
saveas(gca,[saveDir 'figure7.png']);
%% then plot function value for matrix B
figure(8),clf;
% in figure 8 we show the convergence of the Objective
% which is the function value f(x^k)
% plot matrix B
semilogy(0:size(fx1b,1)-1,fx1b-fmin,...
'r','LineWidth',2.5);hold on;
semilogy(0:size(fx2b,1)-1,fx2b-fmin,...
'g','LineWidth',2.5);hold on;
semilogy(0:size(fx3b,1)-1,fx3b-fmin,...
'b','LineWidth',2.5);hold on;
% legend of A & B
l4=sprintf('CBCD1p, #%d',size(fx1b,1)-1);
l5=sprintf('CBCD2p, #%d',size(fx2b,1)-1);
l6=sprintf('CBCD3p, #%d',size(fx3b,1)-1);
legend(l4,l5,l6);
xlabel('#epoch');ylabel('\phi(x^t)-p^*');
title(['Convergence Speed B; #EXP ' num2str(idx) '; #' num2str(index) ]);
set(gca,'fontsize',15);
saveas(gca,[saveDir 'figure8.png']);
%%
figure(9),clf;
% in figure 9 we show the convergence of the Objective
% which is the function value f(x^k)
% plot matrix C
semilogy(0:size(fx1c,1)-1,fx1c-fmin,'r','LineWidth',2.5);hold on;
semilogy(0:size(fx2c,1)-1,fx2c-fmin,'g','LineWidth',2.5);hold on;
semilogy(0:size(fx3c,1)-1,fx3c-fmin,'b','LineWidth',2.5);hold on;
% legend of A & B
l7=sprintf('CBCD1r, #%d',size(fx1c,1)-1);
l8=sprintf('CBCD2r, #%d',size(fx2c,1)-1);
l9=sprintf('CBCD3r, #%d',size(fx3c,1)-1);
legend(l7,l8,l9);
xlabel('#epoch');ylabel('\phi(x^t)-p^*');
title(['Convergence Speed C; #EXP ' num2str(idx) '; #' num2str(index) ]);
set(gca,'fontsize',15);
saveas(gca,[saveDir 'figure9.png']);
end