tonykero/moe

[OLD] Moe is a C++14 header-only dependency-free library providing generic implementations of some metaheuristic algorithms

Moe

GCC / Clang (Linux)	VS 14/15 (Win 32/64)	Coverage (g++5)

Moe is a C++14 header-only dependency-free library providing generic implementations of some metaheuristic algorithms

Quick Overview

The main goal of Moe is to provide a generic and easy way to implement fast solving systems from any complexity with C++.

Moe stands for Maybe Overpowered Entity, but isn't limited to it.

Moe gives a way for the user to completely define the behavior of algorithms, all parameters are adjustable, and the structure allows to easily add features or custom Mutations/Crossover/Algorithms.

Take a look at a base sample using Moe (examples/square_root.cpp):

This example illustrates how Differential Evolution can be used to search for the square root of a number (in this case 2261953600), the known answer is 47560.

#include <moe/moe.hpp>
#include <iostream>
#include <string>

int main()
{
    DifferentialEvolution<int> moether(20);

    long long n = 2261953600;

    moether.setFitnessFunction( [n](auto moe) -> double
    {
        long long genotype = moe.genotype[0];
                
        double error = std::abs(n - genotype*genotype);

        return 1/(error+1);
    });

    moether.run( 50 );  // run 50 generations

    long long genotype = moether.getBestMoe().genotype[0];

    std::cout   << "genotype: "     << genotype << "\n"
                << "fitness: "      << moether.getBestMoe().fitness << std::endl;
}

Examples

Examples can be found here

Features

Moe contains the following features:

• Algorithms:

  • Genetic Algorithm
    • Crossovers:
      • One Point
      • Two Point
      • Uniform
    • Mutations:
      • Substitution
      • Insertion
      • Deletion
      • Translocation
  • Differential Evolution
  • Particle Swarm Optimization
  • • Abstract Classes

• Planned:

  • Performance:
    • Parallel Implementation
  • Better Parameters handling
    • Serialization

How To Use

Add the "include" folder to your include paths and write

#include <moe/moe.hpp>

and you're ready to go !

Building

Moe uses CMake, options are available:

Options	Description	Default Value
BUILD_EXAMPLES	Builds Examples	ON
BUILD_TESTS	Builds Catch Unit Tests	OFF
DEBUG	Enables debugging symbols	OFF

Moe is successfully tested against:

• GCC:
  • 5 (5.4.1)
  • 6 (6.2)
  • MinGW:
    • GCC 5.3
• Clang:
  • 3.6 (3.6.2)
  • 3.7 (3.7.1)
  • 3.8 (3.8.0)
  • 3.9 (3.9.1)
• MSVC:
  • Visual Studio 14 2015
  • Visual Studio 15 2017

Moe does not compile anymore on GCC 4.9

Compiling:

The following clone Moe and generates with CMake

git clone https://github.com/tonykero/Moe.git
cd Moe
mkdir build && cd build && cmake ..

then under linux

make

or

cmake --build .

or (if with MSVC, and -DDEBUG=1 was not specified)

cmake --build . --config Release

Contributing

Please read Contributing

License

The MIT License

Note: 'tests' folder contains catch.hpp which comes from https://github.com/philsquared/Catch and is licensed under the Boost Software License terms.

moe's Issues

Please fix compilation bugs

It would be great to see what perhaps are minor compilation bugs in this repo fixed.

The includes at the top of moe/base/algorithms/AlgorithmImpl.hpp seem incorrect, but even when I fixed those I had a fair number of compilation issues.