Cartesian Genetic Programming toolbox for Matlab

This is an implementation of a cartesian genetic programming toolbox for Matlab.

Get Started

Just include the startup file that is on the root, and you are good to go.

You can follow along with the example of a symbolic regression provided here.

Configuration

The first step is to instantiate the cgp toolbox and provide the configuration. Example:

cgp = cgptoolbox.CGP(struct( ...
    'rows', 1, ...
    'columns', 10, ...
    'levels_back', 10, ...
    'output_type', 'random', ...
    'runs', 100, ...
    'outputs', 1, ...
    'generations', 8000, ...
    'offspring', 10, ...
    'mutation', 0.02, ...
    'fitness_solution', 0.01, ...
    'fitness_operator', '<=' ...
));

Inputs

You should then provide the inputs to the program as a struct, naming the input and providing the value. In this case, we are passing a points input which is itself a struct:

a = zeros(1, 50);
b = zeros(1, 50);
index = 1;
for x = -1:2/49:1
    a(index) = x^6 - 2*(x^4) + x^2;
    b(index) = x;
    index = index + 1;
end

cgp.addInputs( ...
    struct( ...
        'points', struct('x', b, 'y', a) ...
    ) ...
);

Parameters

You can optionally provide parameters to the genotype. The initialize function is executed in order to generate the value for that parameter. the mutate function is called every time there is a mutation on that gene. The mutate function receives the previous value as argument, so you can easily mutate the gene based on his previous value:

% you can add parameters to the genotype
cgp.addParameters( ...
   struct(                                ...
       'name', 'interval',                ...
       'initialize', @()rand(),           ...
       'mutate', @(x)x + rand()           ...
   ), ...
   struct(                                ...
       'name', 'interval',                ...
       'initialize', @()rand(),           ...
       'mutate', @(x)x + rand()           ...
   ) ...
);

Callbacks

If you want to know what the hell is going on behind the scenes, you can use callbacks. There are a number of callbacks available for you to set up. They are defined as a struct, with function handles (pointers to functions). If you have any doubts, check the Matlab documentation on handles

cgp.addCallbacks(struct(
      'CONFIGURATION', @my-configuration-callback-fn,
      'RUN', @my-run-callback-fn,
      'GENERATION', @my-generation-callback-fn,
      'GENERATION_ENDED', @(args)fprintf('generation: %d fitness: %.6f\n', args.generation, args.fitness), ...
      'GENOTYPE', @my-genotype-callback-fn,
      'GENOTYPE_MUTATED', @my-genotype-mutated-callback-fn
      'RUN_ENDED', @reportFitness ...
      'FITTEST_SOLUTION', @(args)fprintf('generation %d - %.12f\n',args.generation, args.fitness), ...
      ))

Function Set

To define the function set, create a directory, and place all the function-set functions inside that directory. Then, provide the path to that directory to the cgp toolbox:

cgp.addFunctionsFromPath('function-set/');

NOTE: Please remember to add the function-set to your Matlab startup script, so the functions are loaded in memory. Check the startup.m

Fitness Function

The fitness function is also set as an handle. The function receives the following arguments:

struct( ...
    'config', vararg.config, ...
    'genes', vararg.genes, ...
    'activeNodes', vararg.activeNodes, ...
    'programInputs', vararg.programInputs, ...
    'functionSet', {vararg.functionSet}, ...
    'run', vararg.run, ...
    'generation', vararg.generation ...
)

With those arguments, you can calculate the new fitness, using your fitness function. See the symbolic-regression example for more details.

Example:

function fitness = myFitnessFunction(vararg)
  % calculate your new fitness based on the `vararg` genes that you received
  % return the new fitness value
  fitness = 2;
end

cgp.addFitnessFunction(@myFitnessFunction);

Run

The last step is to tell the toolbox to run the program:

  cgp.run();

Notes

This toolbox was created for the following thesis and research papers:

• Evolution of classifiers for pitch estimation of piano music using cartesian genetic programming

• Cartesian genetic programming applied to pitch estimation of piano notes

• CGP4Matlab - A Cartesian Genetic Programming MATLAB Toolbox for Audio and Image Processing

• CGP4Matlab - A Cartesian Genetic Programming MATLAB Toolbox for Audio and Image Processing

• CGP4Matlab - A Cartesian Genetic Programming MATLAB Toolbox for Audio and Image Processing