LightNeuNet-MATLAB/GAModel.m at main · RapDoodle/LightNeuNet-MATLAB · 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
classdef GAModel < matlab.mixin.Heterogeneous & handle
    %GAMODEL Summary of this class goes here
    %   Detailed explanation goes here

    properties
        forest = {}
        % Blueprint for the layers
        layersbp = {}
    end

    methods
        function gamodel = GAModel()
        end

        function add(gamodel, layer)
            if ~isa(layer, 'GALayer')
                throw(MException('GAModel:notALayer', ...
                    'Not a GALayer.'));
            end
            gamodel.layersbp{end+1} = layer;
        end

        function gaseqmodel = newindividual(gamodel)
            gaseqmodel = GASequentialModel();

            for i = 1:length(gamodel.layersbp)
                gaseqmodel.add(gamodel.layersbp{i}.copy());
            end

            gaseqmodel.compile();
            gamodel.forest{end+1} = gaseqmodel;
        end

        function populate(gamodel, n)
            % n is the size the of population
            for i = 1:n
                gamodel.newindividual();
            end
        end

        function [minfitnesses, maxfitnesses] = run(gamodel, func, X, y, options, verbose)
            % verbose can be ignored. By defulat: 1
            if nargin < 6
                verbose = 1;
            end

            nummodels = length(gamodel.forest);
            minfitnesses = zeros(options.generations, 1);
            maxfitnesses = zeros(options.generations, 1);

            for generation = 1:options.generations
                for i = 1:nummodels
                    if verbose == 2
                        if mod(i, 10) == 0 || i == 1
                            fprintf('[Generation %d / %d] Simulating: %d / %d\n', ...
                            generation, options.generations, i, nummodels);
                        end
                    end
                    func(gamodel.forest{i}, X, y);
                end

                gamodel.sortforest();
                minfitnesses(generation) = gamodel.forest{nummodels}.fitness;
                maxfitnesses(generation) = gamodel.forest{1}.fitness;

                if verbose > 0 || ...
                        (verbose == -1 && generation == 1) || ...
                        (verbose == -1 && generation ~= 1 && maxfitnesses(generation) > maxfitnesses(generation - 1))
                    fprintf('[Generation [%d / %d] Fitness: %3.2f / %3.2f]\n', ...
                        generation, options.generations, ...
                        minfitnesses(generation), maxfitnesses(generation));
                end

                % Natural selection
                keepidx = int32(nummodels * options.keeprate);
                gamodel.forest = {gamodel.forest{:, 1:keepidx}};

                fitnesses = cellfun(@(x)x.fitness, gamodel.forest);
                chances = fitnesses / sum(fitnesses);
                cp = [0, cumsum(chances)];

                % Reproduction
                for i = keepidx+1:nummodels
                    r = rand();
                    sela = find(r>cp, 1, 'last');

                    selb = sela;
                    while selb == sela
                        r = rand();
                        selb = find(r>cp, 1, 'last');
                    end

                    if sela == selb
                        disp("a=" + sela + " b=" + selb);
                    end
                    modela = gamodel.forest{sela};
                    modelb = gamodel.forest{selb};

                    newmodel = gamodel.newindividual();

                    % Set the parameters
                    for layeridx = 2:length(newmodel.layers)
                        newmodel.layers{layeridx}.W = modela.layers{layeridx}.W;
                        newmodel.layers{layeridx}.b = modela.layers{layeridx}.b;
                        newmodel.layers{layeridx}.crossover(modelb.layers{layeridx}.W, ...
                            modelb.layers{layeridx}.b);
                        newmodel.layers{layeridx}.mutate(options.mutationrate);
                    end
                end

                assert(nummodels == length(gamodel.forest));
            end
        end

        function sortforest(gamodel)
            fitnesses = cellfun(@(x)x.fitness, gamodel.forest);
            [~,sortidx] = sort(fitnesses, "descend");
            gamodel.forest = gamodel.forest(sortidx);
        end

        function mutateall(gamodel, options)
            nummodels = length(gamodel.forest);
            for i = 1:nummodels
                numlayers = length(gamodel.forest{i}.layers);
                for j = 1:numlayers
                    gamodel.forest{i}.layers{j}.mutate(options.mutationrate)
                end
            end
        end

        function replicatefrommodel(gamodel, model)
            nummodels = length(gamodel.forest);
            numlayers = length(model.layers);
            for i = 1:nummodels
                for j = 2:numlayers
                    gamodel.forest{i}.layers{j}.W = model.layers{j}.W;
                    gamodel.forest{i}.layers{j}.b = model.layers{j}.b;
                end
            end
        end
    end
end