vivalaakam_neuro_neat

A library for working with neuroevolutionary networks using the NEAT (NeuroEvolution of Augmenting Topologies) algorithm.

Purpose

• Evolving neural network topologies (NEAT)
• Flexible mutation, crossover, and structure configuration
• Genome serialization/deserialization support
• High performance and easy integration

Main Structures

• Config — evolution and mutation parameters
• Genome — network genome (nodes, connections, mutation and crossover methods)
• Node — network node (neuron)
• Connection — connection between neurons
• NeuronType — neuron type (Input, Hidden, Output)
• Organism — entity encapsulating genome and network
• Network — computational network built from genome

Usage Example

use vivalaakam_neuro_neat::{Config, Genome, Node, Connection, NeuronType};

let config = Config::default();
let nodes = vec![
    Node::new(NeuronType::Input, 0, 0.0, None, Some(1)),
    Node::new(NeuronType::Output, 1, 0.0, None, Some(2)),
];
let connections = vec![Connection::new(0, 1, 0.5)];
let genome = Genome::new(nodes, connections).unwrap();
let network = genome.get_network();
let output = network.activate(vec![1.0]);

Methods Reference

Genome

• new(nodes, connections) — Create a genome from nodes and connections.
• generate_genome(inputs, outputs, hidden, activation, config) — Generate a random genome.
• mutate(child, config) — Mutate genome (add node, add connection, weights, etc).
• mutate_add_node(config) — Add a new node via mutation.
• mutate_add_connection(config) — Add a new connection via mutation.
• mutate_node_bias(config) — Mutate node bias.
• mutate_node_activation(config) — Mutate node activation function.
• mutate_node_enabled(config) — Toggle node enabled/disabled.
• mutate_connection_weight(config) — Mutate connection weight.
• mutate_connection_enabled() — Toggle connection enabled/disabled.
• mutate_crossover(child) — Crossover with another genome.
• get_network() — Build a Network from the genome.
• get_nodes() / get_connections() — Get all nodes/connections.
• get_distance(child) — Levenshtein distance between hidden nodes.
• as_json() — Serialize genome to JSON.
• to_weights() / from_weights() — Convert genome to/from flat weights.

Node

• new(neuron_type, id, bias, activation, position) — Create a node.
• get_id() / get_type() / get_bias() / get_activation() / get_enabled() — Accessors.
• set_bias(bias) / set_activation(activation) / toggle_enabled() — Mutators.
• get_position() / set_position(position) — Node position in network.
• to_weights() / from_weights() — Convert node to/from weights.

Connection

• new(from, to, weight) — Create a connection.
• get_from() / get_to() / get_weight() / get_enabled() — Accessors.
• set_weight(weight) / set_enabled(enabled) / toggle_enabled() — Mutators.
• get_id() — Unique string id for the connection.
• to_weights() / from_weights() — Convert connection to/from weights.

Organism

• new(genome) — Create an organism from a genome.
• activate(inputs) — Run the network on input vector.
• activate_matrix(matrix) — Run the network on input matrix (batch).
• set_fitness(f32) / get_fitness() — Set/get fitness value.
• inc_stagnation() / get_stagnation() — Increase/get stagnation counter.
• mutate(child, config) — Mutate organism (delegates to genome).
• get_genotype() — Get genotype (hidden node ids).
• as_json() — Serialize genome to JSON.
• set_id(id) / get_id() — Set/get organism id.

Network

• new(neurons) — Build a network from neurons.
• activate(inputs) — Run the network on input vector.
• activate_matrix(matrix) — Run the network on input matrix (batch).

Link

• new(from_id, to_id, weight) — Create a link.
• get_from() / get_to() / get_weight() — Accessors.

NeuronType

• Enum: Input, Hidden, Output, Unknown.
• to_bytes() / from_bytes(byte) — Convert to/from byte.

Config

• All fields are public. See src/config.rs for details and defaults.

Memory Bank (Quick Reference)

• Config: mutation parameters, limits, probabilities (see src/config.rs)
• Genome:
  • Genome::new(nodes, connections) — create a genome
  • Genome::generate_genome(inputs, outputs, hidden, activation, &config) — generate a random genome
  • genome.mutate(...) — mutate genome
  • genome.get_network() — get computational network
  • genome.as_json() — serialize to JSON
• Node:
  • Node::new(NeuronType, id, bias, activation, position)
  • Methods: get_id(), get_type(), get_bias(), get_activation(), get_enabled()
• Connection:
  • Connection::new(from, to, weight)
  • Methods: get_from(), get_to(), get_weight(), get_enabled()
• Organism:
  • Organism::new(genome)
  • Methods: activate(inputs), set_fitness(f32), get_fitness()

Tests

See tests/genome.rs for usage examples and tests for mutation, crossover, and serialization.

Author: Andrey Makarov viva.la.akam@gmail.com

License: MIT