-
Notifications
You must be signed in to change notification settings - Fork 11
/
main_originalBenchmarks.m
222 lines (210 loc) · 13.2 KB
/
main_originalBenchmarks.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
% This is the modified code for WCCI 2018 MTOE competition
% Dongrui Wu (drwu@hust.edu.cn), 3/20/2018
% For maximization problems, multiply objective function by -1.
%
% Settings of simulated binary crossover (SBX) in this code is Pc = 1,
% and probability of variable sawpping = 0.
%
clc; clearvars; close all; rng('default'); %warning off all;
popSize=200; % Population size
nGen=500; % Number of generations
selProcess = 'elitist'; % Choose either 'elitist' or 'roulette wheel'
pIL = 0; % Probability of individual learning (BFGA quasi-Newton Algorithm) --> Indiviudal Learning is an IMPORTANT component of the MFEA.
rmp=0.3; % Random mating probability
nRepeat = 20; % Number of repeats; should be 30 in final submission; must >1 to avoid no-display problem
pTransfer=0.4; % Portion of chromosomes to transfer from one task to another
eMin=0.01; % Threshold of accumulated survival rate of divergents
dqWorker = parallel.pool.DataQueue; afterEach(dqWorker, @(data) fprintf('%d-%s-%d ', data{1},data{2},data{3})); % print progress of parfor
dqClient=parallel.pool.DataQueue; afterEach(dqClient,@showResults);
parfor idxTask = 1:9 % parallel execution
tasks = benchmarkOriginal(idxTask); initPop=cell(2,nRepeat);
SOEA1=SOEA(tasks(1),popSize/length(tasks),nGen,selProcess,pIL,nRepeat,idxTask,dqWorker);
SOEA2=SOEA(tasks(2),popSize/length(tasks),nGen,selProcess,pIL,nRepeat,idxTask,dqWorker);
for r=1:nRepeat
initPop{1,r}=SOEA1.initPop{r}; initPop{2,r}=SOEA2.initPop{r}; % initial population
end
data2(idxTask)=MFEA(tasks,popSize,nGen,selProcess,rmp,pIL,nRepeat,idxTask,dqWorker,initPop); % The provided MFEA benchmark algorithm
data1(idxTask)=data2(idxTask);
data1(idxTask).wallClockTime=SOEA1.wallClockTime+SOEA2.wallClockTime;
data1(idxTask).bestFitness=cat(1,SOEA1.bestFitness,SOEA2.bestFitness);
data1(idxTask).bestIndData=[SOEA1.bestIndData; SOEA2.bestIndData];
data1(idxTask).totalEvals=SOEA1.totalEvals+SOEA2.totalEvals;
data3(idxTask)=MFEARR(tasks,popSize,nGen,selProcess,rmp,pIL,nRepeat,idxTask,dqWorker,eMin,initPop);
data4(idxTask)=LDAMFEA(tasks,popSize,nGen,selProcess,rmp,pIL,nRepeat,idxTask,dqWorker,initPop);
data5(idxTask)=MFEALBS(tasks,popSize,nGen,selProcess,rmp,pIL,nRepeat,idxTask,dqWorker,initPop);
data6(idxTask)=EBSGA(tasks,popSize/length(tasks),nGen,selProcess,rmp,pIL,nRepeat,idxTask,dqWorker,initPop);
data7(idxTask)=GMFEA(tasks,popSize,nGen,selProcess,rmp,pIL,nRepeat,idxTask,dqWorker,initPop);
data8(idxTask)=EMTEA(tasks,popSize/length(tasks),nGen,selProcess,pIL,nRepeat,idxTask,dqWorker,initPop);
data9(idxTask)=MTEAbest(tasks,popSize/length(tasks),nGen,selProcess,pIL,nRepeat,pTransfer,idxTask,dqWorker,initPop); % Our algorithm, v2
data=struct('wallClockTime1',data1(idxTask).wallClockTime,'bestFitness1',data1(idxTask).bestFitness,...
'bestIndData1',data1(idxTask).bestIndData,'totalEvals1',data1(idxTask).totalEvals,...
'wallClockTime2',data2(idxTask).wallClockTime,'bestFitness2',data2(idxTask).bestFitness,...
'bestIndData2',data2(idxTask).bestIndData,'totalEvals2',data2(idxTask).totalEvals,...
'wallClockTime3',data3(idxTask).wallClockTime,'bestFitness3',data3(idxTask).bestFitness,...
'bestIndData3',data3(idxTask).bestIndData,'totalEvals3',data3(idxTask).totalEvals,...
'wallClockTime4',data4(idxTask).wallClockTime,'bestFitness4',data4(idxTask).bestFitness,...
'bestIndData4',data4(idxTask).bestIndData,'totalEvals4',data4(idxTask).totalEvals,...
'wallClockTime5',data5(idxTask).wallClockTime,'bestFitness5',data5(idxTask).bestFitness,...
'bestIndData5',data5(idxTask).bestIndData,'totalEvals5',data5(idxTask).totalEvals,...
'wallClockTime6',data6(idxTask).wallClockTime,'bestFitness6',data6(idxTask).bestFitness,...
'bestIndData6',data6(idxTask).bestIndData,'totalEvals6',data6(idxTask).totalEvals,...
'wallClockTime7',data7(idxTask).wallClockTime,'bestFitness7',data7(idxTask).bestFitness,...
'bestIndData7',data7(idxTask).bestIndData,'totalEvals7',data7(idxTask).totalEvals,...
'wallClockTime8',data8(idxTask).wallClockTime,'bestFitness8',data8(idxTask).bestFitness,...
'bestIndData8',data8(idxTask).bestIndData,'totalEvals8',data8(idxTask).totalEvals,...
'wallClockTime9',data9(idxTask).wallClockTime,'bestFitness9',data9(idxTask).bestFitness,...
'bestIndData9',data9(idxTask).bestIndData,'totalEvals9',data9(idxTask).totalEvals,...
'idxTask',idxTask);
send(dqClient,data);
% parSave(['results_' num2str(idxTask) '.mat'],data1(idxTask),data2(idxTask),data3(idxTask),...
% data4(idxTask),data5(idxTask),data6(idxTask),data7(idxTask),data8(idxTask),data9(idxTask),nGen,nRepeat);
end
save('resultsOriginal9.mat','data1', 'data2', 'data3','data4', 'data5', 'data6','data7','data8','data9','nGen','nRepeat');
plotAllResults;
% function parSave(fname,data1,data2, data3,data4,data5, data6,data7,data8,data9,nGen,nRepeat)
% save(fname,'data1', 'data2', 'data3','data4', 'data5', 'data6','data7','data8','data9','nGen','nRepeat');
% end
% Display results in parfor
function showResults(data)
nRepeat=size(data.bestFitness1,1)/2;
fitness=[mean(data.bestFitness1(1:nRepeat,:))' mean(data.bestFitness1(nRepeat+1:end,:))' ...
mean(data.bestFitness2(1:2:end,:))' mean(data.bestFitness2(2:2:end,:))' ...
mean(data.bestFitness3(1:2:end,:))' mean(data.bestFitness3(2:2:end,:))' ...
mean(data.bestFitness4(1:2:end,:))' mean(data.bestFitness4(2:2:end,:))' ...
mean(data.bestFitness5(1:2:end,:))' mean(data.bestFitness5(2:2:end,:))' ...
squeeze(mean(data.bestFitness6,1)) ...
mean(data.bestFitness7(1:2:end,:))' mean(data.bestFitness7(2:2:end,:))' ...
squeeze(mean(data.bestFitness8,1)) squeeze(mean(data.bestFitness9,1))];
[fitness(end,1:2) data.wallClockTime1; ...
fitness(end,3:4) data.wallClockTime2; ...
fitness(end,5:6) data.wallClockTime3; ...
fitness(end,7:8) data.wallClockTime4; ...
fitness(end,9:10) data.wallClockTime5; ...
fitness(end,11:12) data.wallClockTime6; ...
fitness(end,13:14) data.wallClockTime7;...
fitness(end,15:16) data.wallClockTime8;...
fitness(end,17:18) data.wallClockTime9];
figure;
subplot(121);
semilogy(data.totalEvals1(1,:),fitness(:,1),'k-','linewidth',2); hold on;
semilogy(data.totalEvals2(1,:),fitness(:,3),'k--','linewidth',2);
semilogy(data.totalEvals3(1,:),fitness(:,5),'r--','linewidth',2);
semilogy(data.totalEvals4(1,:),fitness(:,7),'g--','linewidth',2);
semilogy(data.totalEvals5(1,:),fitness(:,9),'b--','linewidth',2);
semilogy(data.totalEvals6(1,:),fitness(:,11),'r-','linewidth',2);
semilogy(data.totalEvals7(1,:),fitness(:,13),'g-','linewidth',2);
semilogy(data.totalEvals8(1,:),fitness(:,15),'b-','linewidth',2);
semilogy(data.totalEvals9(1,:),fitness(:,17),'k:','linewidth',2);axis tight;
legend('SOEA','MFEA','MFEARR','LDAMFEA','MFEALBS','EBSGA','GMFEA','EMTEA','MTEA','location','northeast');
title(['Benchmark ' num2str(data.idxTask) ', Task 1']);
xlabel('# func. eval.'); ylabel('Objective');
subplot(122);
semilogy(data.totalEvals1(1,:),fitness(:,2),'k-','linewidth',2); hold on;
semilogy(data.totalEvals2(1,:),fitness(:,4),'k--','linewidth',2);
semilogy(data.totalEvals3(1,:),fitness(:,6),'r--','linewidth',2);
semilogy(data.totalEvals4(1,:),fitness(:,8),'g--','linewidth',2);
semilogy(data.totalEvals5(1,:),fitness(:,10),'b--','linewidth',2);
semilogy(data.totalEvals6(1,:),fitness(:,12),'r-','linewidth',2);
semilogy(data.totalEvals7(1,:),fitness(:,14),'g-','linewidth',2);
semilogy(data.totalEvals8(1,:),fitness(:,16),'b-','linewidth',2);
semilogy(data.totalEvals9(1,:),fitness(:,18),'k:','linewidth',2);axis tight;
legend('SOEA','MFEA','MFEARR','LDAMFEA','MFEALBS','EBS','GMFEA','EMTEA','MTEA','location','northeast');
title(['Benchmark ' num2str(data.idxTask) ', Task 2']);
xlabel('# func. eval.'); ylabel('Objective'); drawnow;
end
% Plot all results
function plotAllResults()
load resultsOriginal9; close all;
for idxTask = 1:9
fitness=[mean(data1(idxTask).bestFitness(1:nRepeat,:))' mean(data1(idxTask).bestFitness(nRepeat+1:end,:))' ...
mean(data2(idxTask).bestFitness(1:2:end,:))' mean(data2(idxTask).bestFitness(2:2:end,:))' ...
mean(data3(idxTask).bestFitness(1:2:end,:))' mean(data3(idxTask).bestFitness(2:2:end,:))' ...
mean(data4(idxTask).bestFitness(1:2:end,:))' mean(data4(idxTask).bestFitness(2:2:end,:))' ...
mean(data5(idxTask).bestFitness(1:2:end,:))' mean(data5(idxTask).bestFitness(2:2:end,:))' ...
squeeze(mean(data6(idxTask).bestFitness,1)) ...
mean(data7(idxTask).bestFitness(1:2:end,:))' mean(data7(idxTask).bestFitness(2:2:end,:))' ...
squeeze(mean(data8(idxTask).bestFitness,1)) squeeze(mean(data9(idxTask).bestFitness,1))];
[fitness(end,1:2) data1(idxTask).wallClockTime; ...
fitness(end,3:4) data2(idxTask).wallClockTime; ...
fitness(end,5:6) data3(idxTask).wallClockTime; ...
fitness(end,7:8) data4(idxTask).wallClockTime; ...
fitness(end,9:10) data5(idxTask).wallClockTime; ...
fitness(end,11:12) data6(idxTask).wallClockTime; ...
fitness(end,13:14) data7(idxTask).wallClockTime; ...
fitness(end,15:16) data8(idxTask).wallClockTime; ...
fitness(end,17:18) data9(idxTask).wallClockTime]
figure;
subplot(121);
semilogy(data1(idxTask).totalEvals(1,:),fitness(:,1),'k-','linewidth',2); hold on;
semilogy(data2(idxTask).totalEvals(1,:),fitness(:,3),'k--','linewidth',2);
semilogy(data3(idxTask).totalEvals(1,:),fitness(:,5),'r--','linewidth',2);
semilogy(data4(idxTask).totalEvals(1,:),fitness(:,7),'g--','linewidth',2);
semilogy(data5(idxTask).totalEvals(1,:),fitness(:,9),'b--','linewidth',2);
semilogy(data6(idxTask).totalEvals(1,:),fitness(:,11),'r-','linewidth',2);
semilogy(data7(idxTask).totalEvals(1,:),fitness(:,13),'g-','linewidth',2);
semilogy(data8(idxTask).totalEvals(1,:),fitness(:,15),'b-','linewidth',2);
semilogy(data9(idxTask).totalEvals(1,:),fitness(:,17),'k:','linewidth',2);axis tight;
legend('SOEA','MFEA','MFEARR','LDAMFEA','MFEALBS','EBSGA','GMFEA','EMTEA','MTEA','location','northeast');
title(['Benchmark ' num2str(idxTask) ', Task 1']);
xlabel('# func. eval.'); ylabel('Objective');
subplot(122);
semilogy(data1(idxTask).totalEvals(1,:),fitness(:,2),'k-','linewidth',2); hold on;
semilogy(data2(idxTask).totalEvals(1,:),fitness(:,4),'k--','linewidth',2);
semilogy(data3(idxTask).totalEvals(1,:),fitness(:,6),'r--','linewidth',2);
semilogy(data4(idxTask).totalEvals(1,:),fitness(:,8),'g--','linewidth',2);
semilogy(data5(idxTask).totalEvals(1,:),fitness(:,10),'b--','linewidth',2);
semilogy(data6(idxTask).totalEvals(1,:),fitness(:,12),'r-','linewidth',2);
semilogy(data7(idxTask).totalEvals(1,:),fitness(:,14),'g-','linewidth',2);
semilogy(data8(idxTask).totalEvals(1,:),fitness(:,16),'b-','linewidth',2);
semilogy(data9(idxTask).totalEvals(1,:),fitness(:,18),'k:','linewidth',2);axis tight;
legend('SOEA','MFEA','MFEARR','LDAMFEA','MFEALBS','EBSGA','GMFEA','EMTEA','MTEA','location','northeast');
title(['Benchmark ' num2str(idxTask) ', Task 2']);
xlabel('# func. eval.'); ylabel('Objective');
end
nAlgs=9;
nPoints=min(100,nGen);
s=nGen/nPoints;
score=nan(9,nAlgs,nPoints);
outX=s:s:nGen;
for idx = 1:9
[tasks,g1,g2] = benchmarkOriginal(idx);
% The globally optimal objective function value known in advance
BF1=tasks(1).fnc(g1);
BF2=tasks(2).fnc(g2);
BFEV1=[data1(idx).bestFitness(1:nRepeat,:); data2(idx).bestFitness(1:2:end,:); ...
data3(idx).bestFitness(1:2:end,:); data4(idx).bestFitness(1:2:end,:); ...
data5(idx).bestFitness(1:2:end,:); squeeze(data6(idx).bestFitness(:,:,1)); ...
data7(idx).bestFitness(1:2:end,:); squeeze(data8(idx).bestFitness(:,:,1)); ...
squeeze(data9(idx).bestFitness(:,:,1))] ...
- BF1*ones(nAlgs*nRepeat,nGen);
BFEV2=[data1(idx).bestFitness(nRepeat+1:end,:); data2(idx).bestFitness(2:2:end,:); ...
data3(idx).bestFitness(2:2:end,:); data4(idx).bestFitness(2:2:end,:); ...
data5(idx).bestFitness(2:2:end,:); squeeze(data6(idx).bestFitness(:,:,2)); ...
data7(idx).bestFitness(2:2:end,:); squeeze(data8(idx).bestFitness(:,:,2)); ...
squeeze(data9(idx).bestFitness(:,:,2))] ...
- BF2*ones(nAlgs*nRepeat,nGen);
u1=mean(BFEV1(:,outX));
u2=mean(BFEV2(:,outX));
st1=std(BFEV1(:,outX));
st2=std(BFEV2(:,outX));
st1(st1==0)=1;
st2(st2==0)=1;
ST1=repmat(st1,nRepeat,1);
ST2=repmat(st2,nRepeat,1);
for i=1:nAlgs
score(idx,i,:)=mean((BFEV1((i-1)*nRepeat+1 : i*nRepeat,outX)-u1)./ST1+(BFEV2((i-1)*nRepeat+1 : i*nRepeat,outX)-u2)./ST2);
end
end
Score=squeeze(sum(score));
figure;
linestyle={'k-','k--','r--','g--','b--','r-','g-','b-','k:'};
hold on;
title('Overall performance');
for i=1:nAlgs
plot(data1(1).totalEvals(1,outX),Score(i,:),linestyle{i},'linewidth',2);
end
axis tight;
legend('SOEA','MFEA','MFEARR','LDAMFEA','MFEALBS','EBSGA','GMFEA','EMTEA','MTEA','location','northeast');
xlabel('# func. eval.'); ylabel('score');
drawnow;
end