Reward-CPS/batch_run.m at master · karkuspeter/Reward-CPS · 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
try
init
catch ME
    disp(ME.identifier);
end
%%
isdirect = true;
setting = struct('problem', ToyCannon2D0D3D, 'Algorithm',1, 'kappa', 1, 'Niter', 60);
direct_settings = setting;
%setting = struct('problem', ToyCannon1D0D2D, 'kappa', 2);
%direct_settings = [direct_settings; setting];

hyper_params = struct(...
                      'problem', '{ToyCannon0D2D3D, ToyCannon2D0D3D}' ...
                      );

common_params = struct('output_off', 1, ...
                       'RandomiseProblem', true, ...
                       'ReturnOptimal', 0, ... %computes optimal values and put in return struct
                       'InitialSamples', 9, ...
                       'Neval', [8 8 20 20 20 20 20 20 20 20 20 20 20], ...
                       'OptimisticMean', 0, ... %lowest possible value (will shift y values)
                       'EvalModulo', 10);
repeat_setting = 8;
fix_seeds = true;

% create a list of all permutations
hp_list = [common_params]; % start with params for all executions
if isdirect
    hp_list = repmat(hp_list, size(direct_settings, 1), 1);
    param_names = [fieldnames(hp_list); fieldnames(direct_settings)];
	hp_list = cell2struct([struct2cell(hp_list); struct2cell(direct_settings)], param_names, 1);
else
    param_names = fieldnames(hyper_params);
    for param_name = param_names(:)'
        hp_list2 = [];
        for i_tuner = 1:numel(hp_list)
            list_el = hp_list(i_tuner);
            param_values = eval(hyper_params.(param_name{:}));
            for j = 1:length(param_values)
                param_val = param_values(j);
                list_el.(param_name{:}) = param_val{:};
                hp_list2 = [hp_list2; list_el];
            end
        end
        hp_list = hp_list2;
    end
end

% execute each
if (true)
    h_stats = {};
    h_linstats = {};
    h_params = {};
    i_tuner_start = 1;
    % create seed list
    if fix_seeds
        fixed_seeds = cell(repeat_setting,1);
        for ind = 1:repeat_setting
            rng('shuffle');
            fixed_seeds{ind} = rng();
        end
    else
        fixed_seeds = {};
    end
else
    i_tuner_start = 3;
end

for i_tuner = i_tuner_start:numel(hp_list)  %dont use i, its being overwritten inside
    list_el = hp_list(i_tuner);

    repeats = repeat_setting;
    keep_prev = 0;

    run_struct = list_el

    %default
    spider_func = @MyEntropySearch;%@RBOCPS;
    show_func = @()(1);%@ShowRepeatResults;
    reward_name = 'R_mean';

    if isfield(run_struct, 'spider_func')
        %str = strcat(run_struct.spider_func,'(varagin)');
        spider_func = run_struct.spider_func;
        run_struct = rmfield(run_struct, 'spider_func');
    end
    % manually compute iterations to be always the same

    no_params = 1;
    repeat;

    % save results
    h_stats{i_tuner} = rep_stats;
    h_linstats{i_tuner} = linstat_vec;
    h_params{i_tuner} = params;

    save('results/hyperparam_temp.mat');
end

% make summary
res_list = hp_list;
% for i=1:numel(res_list)
%     res_list(i).total_samples = h_stats(i).mean_linstat.total_samples;
%     res_list(i).Rcumm_mean = h_stats(i).Rcumm_mean;
%     res_list(i).Rcumm_std = h_stats(i).Rcumm_std;
% end

%save
save(strcat('results/hyper-', datestr(now,'dd-mm-yyyy-HH-MM'), '.mat'));