Crate neuralneat

source ·
Expand description

neuralneat is an implementation of the NeuroEvolution of Augmenting Topologies (NEAT) described in the June 2002 paper titled “Evolving Neural Networks through Augmenting Topologies” by Kenneth O. Stanley and Risto Miikkulainen.

Much of this implementation was also guided by the NEAT 1.2.1 source code.

Basic usage:

use neuralneat::{Genome, Pool, Trainer};
use neuralneat::evaluation::TrainingData;

// To do something useful, you need to decide what your training data is!
fn load_training_data() -> Vec<TrainingData> {
    return vec![];
}

fn main() {
    let input_nodes = 5;
    let output_nodes = 1;
    // Create an initial pool of Genomes
    let mut gene_pool = Pool::with_defaults(input_nodes, output_nodes);
     
    // Load the data that will be used to train and evolve the Genomes
    let training_data: Vec<TrainingData> = load_training_data();
     
    // A Trainer can manage the process of training a population of Genomes
    // over successive generations.
    let mut trainer = Trainer::new(training_data);
     
    trainer.train(
        &mut gene_pool,
        // Train for 100 generations
        100,
    );

    // The winner!
    let best_genome = gene_pool.get_best_genome();
}

Modules

  • defaults
    The defaults module contains the default values of all of the constants used by the Pool to create, mutate, and mate Genomes.
  • evaluation
    The evaluation module contains functions and types related to the process of evaluating and activating the neural network of a Genome, such as basic activation functions that can be used with hidden and output layers of a Genome.

Structs