-
Notifications
You must be signed in to change notification settings - Fork 1
/
smc_for_flocking.m
182 lines (172 loc) · 6.76 KB
/
smc_for_flocking.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
%% importance splitting
function [px,py,pvx,pvy,mc_fit,reason,aheads,resA,resL,PSOInc, psoParticles] = smc_for_flocking()
global x y vx vy ahead Numb r_vax r_vay
rng('shuffle');
Numb = 7; % number of birds in a flock
steps = 1;
init_box = 3; % bounds for initial configuration
dmin = 1; % allowed minimum distance between the birds
[x,y,vx,vy] = flock(0,Numb,steps,init_box,dmin); %initialize the flock
% K = 1; % steps until the next level
stop = 0.0001; % stopping criterion
numPart = 20; % number of simulations
numLevels = 20; % total number of levels
maxAhead = 5; % number of maximum lookaheads before we resample if we couldnt find a new level
rng('shuffle');
fixedPSOParticles = false;
currentPSOParticles = 20;
startPSOParticles = 10;
endPSOParticles = 40;
incrementPSOParticles = 5;
PSOInc = 0;
reason = '';
fit_level = zeros(numPart,1); % fitness levels for each particle
level_dist = zeros(numPart,1); % distance between the levels
mc_fit = zeros(numPart,numLevels);
aheads = zeros(0,0);
psoParticles = zeros(0,0);
px = cell(numPart,1);
py = cell(numPart,1);
pvx = cell(numPart,1);
pvy = cell(numPart,1);
bestVAX = zeros(0,0);
bestVAY = zeros(0,0);
sorting_indices = zeros(0,0);
resA = 0;
resL = 0;
improved = zeros(numPart,1);
precision = .5;%1/numPart;%.5;
best_fit = Inf; % best fit among all the particles
for p=1:numPart
px{p} = x;
py{p} = y;
pvx{p} = vx;
pvy{p} = vy;
level_dist(p) = Inf;
fit_level(p) = best_fit;
mc_fit(p) = fit_level(p);
end
level = 1;
clock = 0;
ahead = 1;
if(~fixedPSOParticles)
currentPSOParticles = startPSOParticles;
end
tic
while best_fit>stop && level<numLevels && ahead<numLevels
for p=1:numPart
% ind = find(px{p}==0,1)-1
x = px{p}(end,:);
y = py{p}(end,:);
vx = pvx{p}(end,:);
vy = pvy{p}(end,:);
[fit_level(p),level_dist(p),improved(p)] = fly_flock(fit_level(p),level_dist(p)...
,currentPSOParticles,level,numLevels);
if level==1 || improved(p)
px{p} = [px{p}; x];
py{p} = [py{p}; y];
pvx{p} = [pvx{p}; vx];
pvy{p} = [pvy{p}; vy];
if fit_level(p) < best_fit
bestVAX(level,:) = r_vax;
bestVAY(level,:) = r_vay;
end
end
end
if min(fit_level)<best_fit
aheads(level) = ahead;
psoParticles(level) = currentPSOParticles;
if(~fixedPSOParticles)
currentPSOParticles = startPSOParticles; %we could learn which ones are beneficial and stick to those! or at least skip low level non-beneficial ones...
end
if ahead>1
ahead = 1; %ahead = ahead - 1;
end
best_fit = min(fit_level);
mc_fit(:,level) = fit_level;
clock(level) = clock(end)+toc;
% waitforbuttonpress;
level = level+1;
% resample bad particles from top positions
[sort_fit,sort_ind]= sort(fit_level,'ascend');
sorting_indices = [sorting_indices sort_ind];
L=numPart*precision;
top_pos = sort_ind(1:L);
bad_pos = sort_ind(L+1:numPart);
resL = resL + 1;
for r=1:numPart-L
% sample from top positionsm
pos = randi(length(top_pos));
% assign a random top position to a bad one
px{bad_pos(r)} = [px{bad_pos(r)}; px{top_pos(pos)}(end,:)];
py{bad_pos(r)} = [py{bad_pos(r)}; py{top_pos(pos)}(end,:)];
pvx{bad_pos(r)} = [pvx{bad_pos(r)}; pvx{top_pos(pos)}(end,:)];
pvy{bad_pos(r)} = [pvy{bad_pos(r)}; pvy{top_pos(pos)}(end,:)];
level_dist(bad_pos(r)) = level_dist(top_pos(pos));
fit_level(bad_pos(r)) = fit_level(top_pos(pos));
end
% tic
else
if ahead < maxAhead
ahead = ahead + 1
else %we reached max aheed
% if (sum(improved) >= numPart*.2) % some configs have improved and we resample
% 'resampling'
% resA = resA + 1;
% ahead = 1;
% % resample bad particles from top positions
% [sort_fit,sort_ind]= sort(fit_level,'ascend');
% L=numPart*precision; % number of configurations we keep = configurations that improved
% top_pos = sort_ind(1:L);
% bad_pos = sort_ind(L+1:numPart);
%
% for r=1:numPart-L
% % sample from top positionsm
% pos = randi(length(top_pos));
% % assign a random top position to a bad one
% px{bad_pos(r)} = [px{bad_pos(r)}; px{top_pos(pos)}(end,:)];
% py{bad_pos(r)} = [py{bad_pos(r)}; py{top_pos(pos)}(end,:)];
% pvx{bad_pos(r)} = [pvx{bad_pos(r)}; pvx{top_pos(pos)}(end,:)];
% pvy{bad_pos(r)} = [pvy{bad_pos(r)}; pvy{top_pos(pos)}(end,:)];
% level_dist(bad_pos(r)) = level_dist(top_pos(pos));
% fit_level(bad_pos(r)) = fit_level(top_pos(pos));
% end
% else %not enough have improved. what now?
disp('no improvement');
if(fixedPSOParticles)
break;
else
disp('pso increase and startover');
if(currentPSOParticles < endPSOParticles)
ahead = 1;
currentPSOParticles = currentPSOParticles + incrementPSOParticles;
PSOInc = PSOInc + 1;
else
disp('ALL PSO EXHAUSTED improvement');
end
%load last good level!
end
% end
end
end
end
if(best_fit<stop)
reason = 'best';
else
if(level>=numLevels)
reason = 'level';
else
if(ahead>=numLevels)
reason = 'ahead';
else
if(currentPSOParticles >= endPSOParticles)
reason = 'pso exhausted'
else
reason = 'no improv';
end
end
end
end
toc
best_fit
end