Swarm Seek

Literature-validated Artificial Bee Colony (ABC) optimization for Python.

Swarm Seek implements Original ABC and principal peer-reviewed variants (GABC, qABC, MABC, CABC) with a NumPy core, optional Numba/JAX backends, an ask/tell API, scikit-learn / Optuna adapters, and automated checks against published benchmark figures.

Status: 0.2.0 (beta). Docs: https://swarm-seek.readthedocs.io/

Install

Requires Python 3.10+.

pip install swarm-seek

Optional extras:

pip install 'swarm-seek[numba]'    # Numba backend
pip install 'swarm-seek[jax]'      # JAX backend
pip install 'swarm-seek[sklearn]'  # ABCSearchCV
pip install 'swarm-seek[optuna]'   # ABCSampler

Editable install for development:

pip install -e ".[dev]"

Quick start

from swarm_seek import ABC, ContinuousSpace
from swarm_seek.benchmarks import sphere

space = ContinuousSpace([(-5.0, 5.0)] * 5)
colony = ABC(space, variant="original", pop_size=20, limit=100, max_evals=5_000, seed=0)
best = colony.minimize(sphere)
print(best.fitness)

Ask/tell loop:

import numpy as np

from swarm_seek import ABC, ContinuousSpace
from swarm_seek.benchmarks import sphere

space = ContinuousSpace([(-5.0, 5.0)] * 5)
colony = ABC(space, variant="gabc", pop_size=20, limit=100, max_evals=5_000, seed=0)

while not colony.converged:
    candidates = colony.ask()
    if candidates.size == 0:
        colony.tell([])
        continue
    colony.tell(np.asarray(sphere(candidates), dtype=np.float64))

print(colony.best.fitness)

Combinatorial ABC (permutation / TSP-style):

from swarm_seek import ABC, PermutationSpace
from swarm_seek.benchmarks import random_cities, tour_length

cities = random_cities(12, seed=0)
space = PermutationSpace(12)
best = ABC(space, variant="cabc", pop_size=16, max_evals=2_000, seed=0).minimize(
    lambda x: tour_length(x, cities)
)
print(best.fitness)

Runnable notebooks with stated pass criteria live in examples/. Backend timing: python scripts/compare_backends.py. Full documentation builds from docs/source/.

License

Apache License 2.0. See LICENSE.

Citation

See CITATION.cff for citation metadata. Archive / DOI: 10.5281/zenodo.21630532 (concept; version v0.1.1 is 10.5281/zenodo.21630533).