pateda — Python Algorithms for Estimation of Distribution Algorithms

pateda is a comprehensive Python library for Estimation of Distribution Algorithms (EDAs) supporting discrete, integer, continuous, and permutation-based optimization. It is a clean, PyPI-ready package extracted from the PATEDA research framework developed at the University of the Basque Country (EHU/UPV).

Features

Classical discrete EDAs

Continuous EDAs

Permutation EDAs

Supporting components

Benchmark functions

Discrete: OneMax, Deceptive-3, Trap, NK-landscape, SAT, UBQP, Ising, HP-Protein, Additive Decomposable, Contiguous Block
Continuous: Sphere, Rosenbrock, Rastrigin, Ackley, GNBG instances
Permutation: TSP, LOP, QAP

Installation

pip install pateda

Optional extras:

pip install "pateda[copula]"   # vine-copula EDAs (pyvinecopulib)
pip install "pateda[dev]"      # development tools
pip install "pateda[all]"      # everything

For neural-network-based EDAs (VAE, GAN, Diffusion, etc.) install the companion package:

pip install pateda-nn

Quick start

import numpy as np
from pateda import EDA, EDAComponents
from pateda.seeding import RandomInit
from pateda.learning import LearnUMDA
from pateda.sampling import SampleFDA
from pateda.selection import TruncationSelection
from pateda.stop_conditions import MaxGenerations
from pateda.functions.discrete.onemax import onemax

n_vars = 50
cardinality = np.full(n_vars, 2)

components = EDAComponents(
    seeding=RandomInit(),
    learning=LearnUMDA(),
    sampling=SampleFDA(),
    selection=TruncationSelection(),
    stop_condition=MaxGenerations(max_gen=100),
    selection_params={"selection_size": 150},
)

eda = EDA(
    pop_size=300,
    n_vars=n_vars,
    fitness_func=onemax,
    cardinality=cardinality,
    components=components,
    random_seed=42,
)
stats, cache = eda.run()
print(f"Best: {stats.best_fitness_overall}")

See examples/ for more complete demos covering discrete, continuous, permutation, and multi-objective problems.

Project structure

pateda/
├── src/
│   └── pateda/
│       ├── core/           # EDA engine (EDA, EDAComponents, Model)
│       ├── learning/       # Probabilistic model learning
│       ├── sampling/       # Model-based sampling
│       ├── selection/      # Selection operators
│       ├── seeding/        # Population initialisation
│       ├── mutation/       # Mutation operators
│       ├── crossover/      # Crossover operators
│       ├── replacement/    # Replacement strategies
│       ├── repairing/      # Constraint repair
│       ├── stop_conditions/# Termination criteria
│       ├── local_optimization/ # Local search wrappers
│       ├── functions/      # Benchmark objective functions
│       ├── inference/      # MAP inference
│       ├── statistics/     # Run-time statistics
│       └── knowledge_extraction/ # Model analysis tools
├── tests/
├── examples/
├── pyproject.toml
└── README.md

Citation

If you use pateda in academic work, please cite:

@misc{pateda,
  author  = {Roberto Santana},
  title   = {pateda: Python Algorithms for Estimation of Distribution Algorithms},
  year    = {2024},
  url     = {https://github.com/rsantana-isg/pateda},
}

License

MIT — see LICENSE.