-
Notifications
You must be signed in to change notification settings - Fork 1
/
Kriging_fit_mixed.m
257 lines (207 loc) · 6.98 KB
/
Kriging_fit_mixed.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
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
function model = Kriging_fit_mixed(inputpar,hyperpar)
% Training a Kriging model with mixed kernel
% All inputs share the same hyper-parameter
%% Preparation
theta = hyperpar.theta ;
lb = hyperpar.lb;
ub = hyperpar.ub;
num = hyperpar.multistarts;
corr_fun = hyperpar.corr_fun;
opt_algorithm = hyperpar.opt_algorithm;
x = inputpar.x;
y = inputpar.y;
model.orig_output = y;
[m n] = size(x); % number of design sites and their dimension
dim = n;
lb_input = min(x);
ub_input = max(x);
mean_output = mean(y);
std_output = std(y);
% u = (x-repmat(lb_input,m,1))./(repmat(ub_input,m,1)-repmat(lb_input,m,1)); % Normalization of input data
% y = (y-repmat(mean_output,m,1))./repmat(std_output,m,1); % Normalization of output data
u = x;
f = ones(m,1);
input_bound = [lb_input; ub_input]; % Input bound
output_moment = [mean_output; std_output]; % Output moment
model.tran_input = u;
model.output = y;
model.orig_input = x;
model.input_dim = n;
model.sample_size = m;
model.dim = dim;
model.order = 1:n;
model.corr_fun = corr_fun;
model.input_bound = input_bound;
model.output_moment = output_moment;
%% Hyper-parameter optimization
%log_ub = log10(ub); log_lb = log10(lb);
for kk = 1: num % Multistart for hype-parameter optimization
theta = rand(1,2).*(ub-lb) + lb;
%theta = 10.^(rand(1,2).*(log_ub-log_lb)+log_lb);
switch opt_algorithm
case 'Hooke-Jeeves'
[t_opt(kk,:), fmin(kk), perf] = boxmin(theta, lb, ub, model); % Hooke & Jeeves Algorithm
case 'Fmincon'
options = optimoptions('fmincon','Display','off','Algorithm','sqp'); % fmincon
[t_opt(kk,:),fmin(kk)] = fmincon(@(x)Kriging_likelihood(x),theta,[],[],[],[],lb,ub,[],options);
case 'GA'
gaoptions = optimoptions('ga','UseParallel', true, 'UseVectorized', false, 'Display','off',...
'FunctionTolerance',1e-3, 'PopulationSize', 800, 'MaxGenerations', 2000);
[t_opt(kk,:),fmin(kk)] = ga(@(x)Kriging_likelihood(x),size(theta,2),[],[],[],[],lb,ub,[],gaoptions);% Genetic algorithm
case 'CMAES'
opts.LBounds = lb; opts.UBounds = ub;
[t_opt(kk,:),fmin(kk)] = Cmaes(@(x)Kriging_likelihood(x),theta,[],opts);
end
end
[value, ind] = min(fmin);
model.theta = t_opt(ind,1);
model.weight = t_opt(ind,end);
model.likelihood = value;
corrmat = feval(corr_fun,u,model.theta,dim,'off');
corrmat = (1-model.weight)*corrmat + model.weight*triu(u*u') + 10^(-12)*eye(m); % mixed kernel
[upper_mat rd] = chol(corrmat);
model.upper_mat = upper_mat;
model.corrmat = corrmat;
sigma2 = sum((upper_mat'\y).^2)/m;
model.sigma2 = sigma2;
%% Likelihood function
function Likelihood = Kriging_likelihood(theta)
corrmat = feval(corr_fun,u,theta(1),dim,'off');
corrmat = (1-theta(end))*corrmat + theta(end)*triu(u*u') + 10^(-12)*eye(m); % mixed kernel
[upper_mat rd] = chol(corrmat);
sigma2 = sum((upper_mat'\y).^2)/m;
Likelihood = m*log(sigma2)+2*sum(log(diag(upper_mat)));
end
%% Hooke & Jeeves Algorithm
function [t, f, perf] = boxmin(t0, lo, up,par)
% Hooke & Jeeves Algorithm for hyper-parameters optimization (This part comes from Dace toolbox)
% Initialize
[t, f, itpar] = start(t0, lo, up, par);
if ~isinf(f)
% Iterate
p = length(t);
if p < 2, kmax = 5; else, kmax = 10; end
for k = 1 : kmax
th = t;
[t, f, itpar] = explore(t, f, itpar, par);
[t, f, itpar] = move(th, t, f, itpar, par);
end
end
perf = struct('nv',itpar.nv, 'perf',itpar.perf(:,1:itpar.nv));
end
function [t, f, itpar] = start(t0, lo, up, par)
% Get starting point and iteration parameters
% Initialize
t = t0(:); lo = lo(:); up = up(:); p = length(t);
D = 2 .^ ([1:p]'/(p+2));
ee = find(up == lo); % Equality constraints
if ~isempty(ee)
D(ee) = ones(length(ee),1); t(ee) = up(ee);
end
ng = find(t < lo | up < t); % Free starting values
if ~isempty(ng)
t(ng) = (lo(ng) .* up(ng).^7).^(1/8); % Starting point
end
ne = find(D ~= 1);
% Check starting point and initialize performance info
[f] = Kriging_likelihood(t); nv = 1;
itpar = struct('D',D, 'ne',ne, 'lo',lo, 'up',up, ...
'perf',zeros(p+2,200*p), 'nv',1);
itpar.perf(:,1) = [t; f; 1];
if isinf(f) % Bad parameter region
return
end
if length(ng) > 1 % Try to improve starting guess
d0 = 16; d1 = 2; q = length(ng);
th = t; fh = f; jdom = ng(1);
for k = 1 : q
j = ng(k); fk = fh; tk = th;
DD = ones(p,1); DD(ng) = repmat(1/d1,q,1); DD(j) = 1/d0;
alpha = min(log(lo(ng) ./ th(ng)) ./ log(DD(ng))) / 5;
v = DD .^ alpha; tk = th;
for rept = 1 : 4
tt = tk .* v;
[ff ] = Kriging_likelihood(tt); nv = nv+1;
itpar.perf(:,nv) = [tt; ff; 1];
if ff <= fk
tk = tt; fk = ff;
if ff <= f
t = tt; f = ff; jdom = j;
end
else
itpar.perf(end,nv) = -1; break
end
end
end % improve
% Update Delta
if jdom > 1
D([1 jdom]) = D([jdom 1]);
itpar.D = D;
end
end % free variables
itpar.nv = nv;
end
% --------------------------------------------------------
function [t, f, itpar] = explore(t, f, itpar, par)
% Explore step
nv = itpar.nv; ne = itpar.ne;
for k = 1 : length(ne)
j = ne(k); tt = t; DD = itpar.D(j);
if t(j) == itpar.up(j)
atbd = 1; tt(j) = t(j) / sqrt(DD);
elseif t(j) == itpar.lo(j)
atbd = 1; tt(j) = t(j) * sqrt(DD);
else
atbd = 0; tt(j) = min(itpar.up(j), t(j)*DD);
end
% [ff fitt] = objfunc(tt,par); nv = nv+1;
[ff] = Kriging_likelihood(tt); nv = nv+1;
itpar.perf(:,nv) = [tt; ff; 2];
if ff < f
t = tt; f = ff;
else
itpar.perf(end,nv) = -2;
if ~atbd % try decrease
tt(j) = max(itpar.lo(j), t(j)/DD);
% [ff fitt] = objfunc(tt,par); nv = nv+1;
[ff ] = Kriging_likelihood(tt); nv = nv+1;
itpar.perf(:,nv) = [tt; ff; 2];
if ff < f
t = tt; f = ff;
else
itpar.perf(end,nv) = -2;
end
end
end
end % k
itpar.nv = nv;
end
% --------------------------------------------------------
function [t, f, itpar] = move(th, t, f, itpar, par)
% Pattern move
nv = itpar.nv; ne = itpar.ne; p = length(t);
v = t ./ th;
if all(v == 1)
itpar.D = itpar.D([2:p 1]).^.2;
return
end
% Proper move
rept = 1;
while rept
tt = min(itpar.up, max(itpar.lo, t .* v));
% [ff fitt] = objfunc(tt,par); nv = nv+1;
[ff ] = Kriging_likelihood(tt); nv = nv+1;
itpar.perf(:,nv) = [tt; ff; 3];
if ff < f
t = tt; f = ff;
v = v .^ 2;
else
itpar.perf(end,nv) = -3;
rept = 0;
end
if any(tt == itpar.lo | tt == itpar.up), rept = 0; end
end
itpar.nv = nv;
itpar.D = itpar.D([2:p 1]).^.25;
end
end