Study-DFO/tests_sysbio/runMultiStarts.m at master · ICB-DCM/Study-DFO · GitHub

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
function parameters = runMultiStarts(objectiveFunction, maxFunEvals, nStarts, solver, nPar, parMin, parMax, useGradients)

if contains(solver,'hybrid')
    parameters = runMultiStartsHybrid(objectiveFunction, maxFunEvals, nStarts, solver, nPar, parMin, parMax, useGradients);
    return;
end

%     clearPersistentVariables();

%     tol = 1e-10;

% The maximum number of function evaluations is chosen proportional to the
% number of parameters to take into account that DFO algorithms usually
% require a number of function evaluations that increases linearly with the
% dimension of the search space, not taking into account that the objective
% function landscape may in general become more complex.
numevals = maxFunEvals*nPar;

options.obj_type = 'log-posterior';
options.comp_type = 'sequential';
options.proposal = 'latin hypercube';
options.n_starts = nStarts;
if useGradients
    options.objOutNumber = 2;
else
    options.objOutNumber = 1;
end

options.mode = 'text';
options.localOptimizer = solver;

% set maxFunEvals for the different optimizers
switch solver
    case 'fmincon'
        lOptions = optimoptions(@fmincon);
        lOptions.MaxFunctionEvaluations = numevals;
        lOptions.MaxIterations = numevals;
        lOptions.Display = 'off';
        if useGradients
            options.objOutNumber = 2;
            lOptions.GradObj = 'on';
        else
            lOptions.GradObj = 'off';
        end
    case 'dhc'
        lOptions.MaxFunEvals = numevals;
    case 'rcs'
        lOptions.MaxFunEvals = numevals;
        lOptions.MaxIter = numevals;
    case 'bobyqa'
        lOptions.MaxFunEvals = numevals;
        lOptions.Rhobeg = 0.1*max(abs(parMax-parMin));
    case 'mcs'
        lOptions.printLevel = 1;
        lOptions.MaxFunEvals = numevals;
    case 'direct'
        lOptions.maxevals = numevals;
        lOptions.maxits = numevals;
    case 'meigo-ess'
        lOptions.inter_save = false;
        lOptions.maxeval = numevals;
        lOptions.maxtime = Inf;
        lOptions.local.solver = 'fmincon';
        lOptions.local.finish = 'fmincon';
        if useGradients
            lOptions.local.solver_use_gradient = true;
            lOptions.local.finish_use_gradient = true;
        else
            lOptions.local.solver_use_gradient = false;
            lOptions.local.finish_use_gradient = false;
        end
    case 'meigo-ess-dhc'
        options.localOptimizer = 'meigo-ess';
        lOptions.inter_save = false;
        lOptions.maxeval = numevals;
        lOptions.maxtime = Inf;
        lOptions.local.solver = 'dhc';
        lOptions.local.finish = 'dhc';
    case 'meigo-ess-bobyqa'
        options.localOptimizer = 'meigo-ess';
        lOptions.inter_save = false;
        lOptions.maxeval = numevals;
        lOptions.maxtime = Inf;
        lOptions.local.solver = 'bobyqa';
        lOptions.local.finish = 'bobyqa';
    case 'cmaes'
        lOptions.MaxFunEvals = numevals;
        lOptions.MaxIter = numevals;
        lOptions.LBounds = parMin;
        lOptions.UBounds = parMax;
    case 'pswarm'
        lOptions.MaxIter = numevals;
        lOptions.MaxObj  = numevals;

    otherwise
        error('solver not recognized');
end
options.localOptimizerOptions = lOptions;

parameters.number = nPar;
parameters.min = parMin;
parameters.max = parMax;

parameters = getMultiStarts(parameters, objectiveFunction, options);

% do not want to store this
for j = 1:nStarts
    parameters.MS.hessian = [];
end

end % function

function parameters = runMultiStartsHybrid(objectiveFunction, maxFunEvals, nStarts, solver, nPar, parMin, parMax, useGradients)

numevals = maxFunEvals*nPar;

parameters = struct();
parameters.min = parMin;
parameters.max = parMax;
parameters.number = nPar;

optionsSS = PestoOptions();
optionsSS.obj_type = 'log-posterior';
optionsSS.n_starts = nStarts;
if contains(solver,'snobfit')
    optionsSS.ss_optimizer = 'snobfit';
elseif contains(solver,'mcs')
    optionsSS.ss_optimizer = 'mcs';
elseif contains(solver,'simple')
    optionsSS.ss_optimizer = 'simple';
else
    error('solver not recognized');
end
optionsSS.ss_maxFunEvals = min([numevals / 10, 50*nStarts]);

starttime = tic;

guess = getStartpointSuggestions(parameters, objectiveFunction, optionsSS);

time_ss = toc(starttime);

options = PestoOptions();
options.obj_type = 'log-posterior';
options.n_starts = nStarts;
options.proposal = 'user-supplied';
options.objOutNumber = 2;
options.mode = 'text';

options.localOptimizer = 'fmincon';
lOptions = optimoptions(@fmincon);
lOptions.MaxFunctionEvaluations = numevals-optionsSS.ss_maxFunEvals;
lOptions.MaxIterations = numevals-optionsSS.ss_maxFunEvals;
lOptions.Display = 'off';
lOptions.GradObj = 'on';
options.localOptimizerOptions = lOptions;

parameters.guess = guess;

parameters = getMultiStarts(parameters, objectiveFunction, options);

parameters.time_ss = time_ss;

end