Evolutionary Algorithms & Neuroevolution

Clear, modular & accessible evolution in Python

EvoLib bridges the gap between overly simple GA libraries and overly complex research frameworks - designed for clarity, accessibility, and easy adoption in both research and education .

→ View on GitHub 📘 Read the Docs
Installation 
pip install evolib
Minimal example 
from evolib import Population

pop = Population("examples/01_basic_usage/population.yaml")
pop.run()

Why EvoLib?

Modular by Design

EvoLib combines classical evolutionary strategies (μ+λ, adaptive mutation, crossover) with structural neuroevolution in one consistent framework.

Config-Driven Experiments

Define experiments via YAML with automatic validation - explicit, reproducible, and easy to adapt. Swap strategies or structures without touching core code.

Neuroevolution Built-in

Evolve weights, structure (connections, neurons, recurrence), and activations - combined with classical ES/GA methods.

Key features

Config-Driven Experiments

Define runs via YAML with automatic validation - no code changes needed.

Clear & Accessible

Clean API, transparent configs, examples that are easy to follow.

Classical Strategies

(μ, λ), (μ+λ), steady-state; tournament/roulette/rank/SUS selection.

Adaptive Operators

Constant, exponential decay, global & individual adaptation.

Neuroevolution

Evolve neural networks (EvoNet, structural growth).

Open & Extensible

MIT-licensed, flat API (from evolib import ...), easy to extend.

Quickstart

1) Setup

Install

pip install evolib

Create quickstart.yaml

# quickstart.yaml
parent_pool_size: 10
offspring_pool_size: 30
max_generations: 20
num_elites: 1
random_seed: 42

evolution:
  strategy: mu_plus_lambda

modules:
  main:
    type: vector
    dim: 8
    bounds: [-1.0, 1.0]
    initializer: random_vector
    mutation:
      strategy: constant
      probability: 1.0
      strength: 0.05

2) Create Python Script run_quickstart.py

# run_quickstart.py
import numpy as np
from evolib import Indiv, Population, sphere, plot_fitness

def my_fitness(indiv: Indiv) -> None:
    x = indiv.para["main"].vector
    indiv.fitness = sphere(x)

pop = Population("quickstart.yaml", fitness_function=my_fitness)
pop.run(verbosity=1)

# Basic plotting (default metrics: best, mean, and median fitness)
plot_fitness(pop, show=True)

3) Run

python run_quickstart.py

Showcases

Neuroevolution

Evolving network topology (add neurons/edges). Small, didactic, visual.

Image Approximation Demo

Image Approximation

Evolving an EvoNet to reproduce a target image. Great for lectures and demos.

Image Approximation Demo

Function Approximation

Classic continuous benchmark with adaptive mutation strategies.

Image Approximation Demo