Melon

Melon is a neuro-evolution library. Neuroevolution is the concept of using genetic algorithms to optimize neural networks. This library is both lightweight and implements many types of crossover, mutation, and selection to improve training.

Adding to your project

If you are using cmake, adding the library to your project is incredibly easy. Simply clone the repository and call add_subdirectory() on wherever it has been downloaded. There currently is no support for other build systems.

Example

#include <iostream>
#include <Melon/GeneticAlgorithm.hpp>

int main(){
    GeneticAlgorithm algo;
    
    // This is the neural network model of all neural networks in the genetic algorithm
    Layer model[] = {
        {
            .activationFunc = NO_ACTIVATION_FUNCTION,
            .numNodes = 1
        },
        {
            .activationFunc = SIGMOID,
            .numNodes = 3
        },
        {
            .activationFunc = SOFTMAX,
            .numNodes = 2
        }
    };

    // Add a neural networks to the genetic algorithm
    // Every neural network in the genetic algorithm must have the same model
    // Note: The reason the NeuralNetwork class is coded in a way that allows neural networks to have different models is to allow the user to use the NeuralNetwork class in other areas of their code.
    for(int i = 0; i < 200; i++){
        algo.brains.push_back(new NeuralNetwork(model, 3, RANDOM_WEIGHT_INITIALIZATION));
    }

    // Initialize the genetic algorithm
    algo.init();

    int numGenerations = 100;
    for(int i = 0; i < numGenerations; i++){
        // Here you would run a simulation for all neural networks in order to assign each of them a fitness value

        // This line runs the genetic algorithm(does selection, crossover, and mutation on the neural networks)
        algo.performGeneticAlgorithm();
    }

    // At this point neural networks should be trained. You would probably evalute the fitness values for all neural networks here and get the one with the highest fitness.

    // Destroy the genetic algorithm
    algo.destroy();

    return 0;
}