FuncCraft

FuncCraft is a Python package backed by a C++17 benchmark-function generator for black-box optimization research. It is designed for scalable benchmark suite generation across dimensions: one editable suite YAML file can generate hundreds, thousands, or practically unlimited numbers of distinct benchmark functions while controlling the constructed optimum location and optimum value.

Install

python -m pip install --upgrade funccraft
python -m pip install numpy scipy minionpy

YAML-First Workflow

Suite YAML is the easiest way to configure FuncCraft from files. The same structure can also be passed as a Python dictionary to BenchmarkSuite.

supported_dimensions: any
base_functions: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34]
composition_base_functions: [4, 8, 10, 11, 12, 13, 15, 16, 17, 18, 19, 20, 21, 23, 28, 30, 34, 33, 2, 3, 5, 9, 26, 27]
coordinate_transforms:
  - kind: none
    probability: 0.0
    parameters: []
  - kind: rotation
    probability: 0.34
    parameters: []
  - kind: affine
    probability: 0.33
    parameters: []
  - kind: subspace-rotation
    probability: 0.33
    parameters: []
value_transforms:
  - kind: none
    probability: 0.5
    parameters: []
  - kind: power
    probability: 0.25
    parameters: []
  - kind: oscillatory
    probability: 0.25
    parameters: []
  - kind: cosine-zero
    probability: 0.0
    parameters: []
compositions:
  - kind: cpm-wsum
    probability: 0.1
    parameters: []
  - kind: cpm-power-mean
    probability: 0.1
    parameters: [3.0]
  - kind: cpm-power-mean
    probability: 0.1
    parameters: [0.1]
  - kind: cpm-level-well
    probability: 0.2
    parameters: []
  - kind: dpm-softmax
    probability: 0.25
    parameters: [0.005]
  - kind: dpm-bgsoftmax
    probability: 0.25
    parameters: [0.005, 1.0, 0.01]
min_components: 2
max_components: 5
max_nested_composition_depth: 1
nested_probability: 0.1
requested_number_of_functions: 500
master_seed: 1
lower_bound: -100
upper_bound: 100
assigned_fopt: 100.0
xopt_domain_shrink_factor: 0.8
suite_label: my-suite

Load the YAML and evaluate a function. Function indices are one-based:

import funccraft as fc

dimension = 10
function_index = 1

suite = fc.BenchmarkSuite("my_suite.yaml", dimension=dimension)
f = suite.function(function_index)

points = [[0.0] * dimension, [1.0] * dimension]
values = f.evaluate(points)
print(values)

The packaged 2026 suite is also YAML-backed and exposed through a shortcut:

import funccraft as fc

dimension = 10
function_index = 1
year = 2026
version = 1
suite = fc.SuiteCollection(year=year, version=version).benchmark_suite(dimension)
f = suite.function(function_index)
values = f.evaluate([[0.0] * dimension])

All evaluations are batched: pass a list of points, not one flat point vector.

Optimization

SciPy:

import numpy as np
from scipy.optimize import differential_evolution
import funccraft as fc

dimension = 10
function_index = 1
year = 2026
version = 1
suite = fc.SuiteCollection(year=year, version=version).benchmark_suite(dimension)
f = suite.function(function_index)
domain = f.domain
bounds = list(zip(domain.lower_bound, domain.upper_bound))

def objective(x):
    return f.evaluate([np.asarray(x, dtype=float).tolist()])[0]

result = differential_evolution(objective, bounds, seed=1, polish=False)
print(result.x, result.fun)

MinionPy:

import funccraft as fc
import minionpy as mpy

dimension = 10
function_index = 1
year = 2026
version = 1
suite = fc.SuiteCollection(year=year, version=version).benchmark_suite(dimension)
f = suite.function(function_index)
domain = f.domain

optimizer = mpy.Minimizer(
    func=f.evaluate,
    x0=[
        [0.0] * dimension,
        [1.0] * dimension,
        [-0.5] * dimension,
    ],
    bounds=list(zip(domain.lower_bound, domain.upper_bound)),
    algo="ARRDE",
    maxevals=10000,
    callback=None,
    seed=None,
    options=None,
)
result = optimizer.optimize()
print(result.x, result.fun)

Mechanism Summary

FuncCraft builds functions from primitive benchmark landscapes, coordinate transforms, value transforms, and composition rules:

f(x) = assigned_fopt + scale_factor * psi(x, z_1(x), ..., z_m(x))
z_i(x) = component_scale_i * phi_i(g_i(T_i(x)) - f_i*)

Omitted component scales and the final scale_factor are estimated deterministically at construction time and materialized in exported specs.

Implemented mechanism families include:

• 34 primitive base functions, including Sphere, Rosenbrock, Ackley, Rastrigin, Griewank, Schwefel, Katsuura, Levy, BentCigar, HappyCat, HGBat, and StyblinskiTang.
• coordinate transforms: none, rotation, affine, subspace-rotation.
• value transforms: none, power, oscillatory, cosine-zero, huber, log, softplus-threshold, dead-zone, saturating, piecewise-power, noisy-smooth.
• compositions: none, cpm-wsum, cpm-power-mean, cpm-level-well, cpm-max, cpm-smoothmax, cpm-constraint-penalty, cpm-lexicographic, cpm-product, cpm-max-plus-mean, cpm-cvar, dpm-softmax, dpm-bgsoftmax.

Components can also be nested composed functions, up to the configured max_nested_composition_depth.

Names are parsed permissively: case, spaces, hyphens, and underscores are normalized before matching.

Exported Manifests

Input YAML is for configuration and editing. Exported YAML is a materialized record of what FuncCraft built:

f.export_yaml("function_materialized.yaml")
suite.export_manifest("suite_manifest.yaml")

Exported YAML records include generated matrices, selected subspaces, DPM centers and biases, assigned optima, scale factors, labels, and metadata.

Warning: the packaged 2026_v1 suite contains 1,000,000 functions. Exporting the full shipped suite manifest will write every generated function record and is usually unnecessary. To export a smaller reproducible subset, copy the collection configuration dictionary, reduce requested_number_of_functions, build the smaller suite, and export that suite:

import funccraft as fc

collection = fc.SuiteCollection(year=2026, version=1)
config = collection.config
config["requested_number_of_functions"] = 500

suite = fc.BenchmarkSuite(config, dimension=2)
suite.export_manifest("suite_2026_v1_first_500.yaml")

Links

• Documentation: https://funccraft.readthedocs.io/
• Python examples: docs/source/python/examples.rst
• C++ examples: docs/source/cpp/examples.rst
• Source: https://github.com/khoirulmuzakka/FuncCraft
• Issues: https://github.com/khoirulmuzakka/FuncCraft/issues