NimEAT

NEAT (NeuroEvolution of Augmenting Topologies) implementation in Nim.

Installation

Requires Nim 2.0+ and nimble.

nimble install

Quick Start

import genotype
import network
import population
import species
import params

# Configure parameters
param.setPopSize(150)

# Create initial genome (2 inputs, 1 output)
var g = newGenotype(2, 1)

# Create and spawn population
let p = newPopulation()
p.spawn(g)

# Evolution loop
for generation in 1..100:
    for organism in p.population:
        # Evaluate fitness
        let output = organism.net.predict(@[1.0, 0.0])
        organism.fitness = computeFitness(output)
    
    p.advanceGeneration()

Running the XOR Example

nimble runxor

Commands

nimble test      # Run all tests
nimble testFast  # Run unit tests only (faster)
nimble bench     # Run benchmarks
nimble clear     # Clean build artifacts

Parameters

Key parameters in params.nim:

Parameter	Default	Description
POP_SIZE	150	Population size
COMPAT_THRESHOLD	2.0	Species compatibility threshold
MUT_ADD_NODE_PROB	0.001	Probability of adding a node
MUT_ADD_LINK_PROB	0.15	Probability of adding a link
MUT_WEIGHT_PROB	0.9	Probability of mutating weights

Adjust via setters:

param.setPopSize(512)
param.setCompatThreshold(3.0)

Project Structure

.
├── genotype.nim      # Genome representation
├── network.nim       # Neural network phenotype
├── species.nim       # Species and organism types
├── population.nim    # Population management
├── mutations.nim     # Mutation operators
├── params.nim        # Configuration parameters
├── activation.nim    # Activation functions
├── serialization.nim # JSON serialization
└── tests/            # Test suite

Development

See GEMINI.md for tooling guidelines.

License

MIT