Categorygithub.com/JoshPattman/goevo
repositorypackage
0.5.0
Repository: https://github.com/joshpattman/goevo.git
Documentation: pkg.go.dev
Overview
Dependencies
13
Dependents
9

# README

WARNING: This package is currently under a period of intense development and will be extremely unstable until this message is removed. If you would like to use this package, I would highly recommend using a specific tagged version.

Go Reference Go Report Go Go Coverage Stability

GoEvo - Extensible Evolutionary Algorithms

GoEvo is an evolutionary algorithms package for Go that provides both flexible and fast implementations of many genetic algorithms.

Some Key Features:

  • Many Algorithms: Support for many types of evolutionary algorithms, from basic hill-climbers to full NEAT.
  • Optimize Anything: NEAT genotypes, slices of floats, or any type that you can perform crossover, mutation, and fitness evaluation on are supported by this package.
  • Flexible for Your Use-Case: As long as your components (such as mutation functions, selection functions, etc) implement the easy-to-understand interfaces specified, you can implement interesting and unique custom behavior.

Documentation

Currently, the only documentation can be found on the pkg.go.dev website. However, examples and possibly a wiki will be coming soon.

Components

GoEvo is broken down into distinct components, each of which may have multiple implementations to provide various behaviors. Below are the different types of components in GoEvo:

  • Genotype - A structure representing an organism's DNA, it can be anything from an array of floats to a NEAT neuron graph
  • Phenotype - Some genotypes, such as neat, require a sort of compiling step to allow them to be used, a phenotype is the result of this compiling
  • Reproduction - How can some number of organisms be combined and mutated to create a new child
  • Mutation - Genotype-specific, it represents a strategy to make some mutations on a Genotype
  • Crossover - Genotype-specific, it represents a strategy to combine some number of parents to create a child (without mutation)
  • Selection - Given some number of agents, each with their own fitness, choose one to be a parent
  • Population - A higher level concept that stores some number of agents, and is capable of creating a new generation

Built-In Components List

Below are the components that GoEvo currently ships with. If you require one that is not included, feel free to create it and make a pull request!

Genotypes, Mutations, and Crossovers

  • NeatGenotype - Provides a NEAT (Neuro Evolution of Augmenting Topologies) gene graph
    • NeatCrossoverAsexual - Crossover to clone one parent
    • NeatCrossoverSimple - Clones the topology of one parent but randomly chooses weights from the other
    • NeatMutationStd - Mutates a genotype with normally distributed values
  • ArrayGenotype - Provides a genotype that is a slice of values
    • ArrayCrossoverAsexual - Crossover to clone one parent
    • ArrayCrossoverKPoint - K-Point crossover
    • ArrayCrossoverUniform - Uniform crossover
    • ArrayMutationStd - Mutates with normal distribution for float arrays
    • ArrayMutationRandomBool - Randomly switches bool values
    • ArrayMutationRandomRune - Randomly switches rune values

Selections

  • TournamentSelection - N-sized tournament selection
  • EliteSelection - Always pick the best agent

Populations

  • SimplePopulation - One species generational population
  • SpeciatedPopulation - Generation population with multiple species
  • HillClimberPopulation - Population with two agents that perform hill climbing

TODO pre v1.0

  • NEAT Population
  • Dense genotype using Gonum
  • Add an example dir and/or a wiki
  • Possibly add a pre-made evolutionary loop