Editable YAML specs

YAML is the recommended way to configure FuncCraft suites because it is easy to read, edit, review, and version. A YAML file describes what should be generated; FuncCraft turns it into runtime benchmark functions for a chosen dimension.

There are two main spec types:

SuiteSpec: A generator recipe for many functions. This is the usual entry point.
FunctionSpec: A complete recipe for one function. Use this when you want to hand-design a single benchmark function.

Suite YAML

Save a file such as my_suite.yaml:

supported_dimensions: any
base_functions: [1, 9, 10, 11, 12]
composition_base_functions: [9, 10, 11, 12]

coordinate_transforms:
  - kind: rotation
    probability: 0.5
  - kind: blockrotation
    probability: 0.5

value_transforms:
  - kind: none
    probability: 0.5
  - kind: osc
    probability: 0.5

compositions:
  - kind: cpmsum
    probability: 0.5
  - kind: dpmsoftmax
    probability: 0.5
    parameters: [0.005]

min_components: 2
max_components: 4
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 it in Python:

import funccraft as fc

dimension = 10
spec = fc.load_suite_spec("my_suite.yaml")
suite = fc.BenchmarkSuite(spec, dimension)

Load it in C++:

#include "funccraft.h"

int main() {
    const int dimension = 10;
    FuncCraft::SuiteSpec spec =
        FuncCraft::load_suite_spec("my_suite.yaml");
    FuncCraft::BenchmarkSuite suite(spec, dimension);
}

Suite fields

supported_dimensions: A comma-separated list such as "2,5,10,20" or any.
base_functions: Primitive functions that are included as mandatory single-function benchmarks in the generated suite. Values may be numeric IDs or names.
composition_base_functions: Primitive pool used inside composed functions and nested composed components.
coordinate_transforms, value_transforms, compositions: Choice tables. Each entry has kind, probability, and optional parameters. Probabilities in each table are fractions and should sum to one.
min_components and max_components: Range for the number of components in composed functions.
max_nested_composition_depth: 0 means composed functions use only primitive components. Larger values allow components to be composed functions.
nested_probability: Probability that a component becomes a nested composed function when nesting is allowed.
requested_number_of_functions: Number of generated functions requested from the suite.
max_number_of_functions: Optional hard cap. 0 means no explicit cap beyond the requested count.
master_seed: Seed for the suite generator.
lower_bound and upper_bound: Search-domain bounds for generated functions.
assigned_fopt: Assigned optimum value for generated functions.
xopt_domain_shrink_factor: Fraction of the domain used for generated optimum locations and DPM centers. 0.8 means the central 80 percent of the domain.
suite_label: Human-readable label stored with generated specs.

Function YAML

A function YAML describes one benchmark function directly:

dimension: 2
domain:
  dimension: 2
  lower_bound: [-5.0, -5.0]
  upper_bound: [5.0, 5.0]
assigned_xopt: [1.0, -2.0]
assigned_fopt: 0.0
scale_factor: 1.0
seed: 7
label: example-function
components:
  - base_function: Sphere
    coordinate_transform:
      kind: none
      input_dimension: 2
      output_dimension: 2
      assigned_xopt: [1.0, -2.0]
    value_transform:
      kind: none
  - base_function: Rastrigin
    coordinate_transform:
      kind: rotation
      input_dimension: 2
      output_dimension: 2
      assigned_xopt: [1.0, -2.0]
      seed: 17
    value_transform:
      kind: power
      parameters: [1.25, 1.0]
composition:
  kind: cpm-wsum

Load it in Python:

import funccraft as fc

spec = fc.load_function_spec("my_function.yaml")
f = fc.BenchmarkFunction(spec)
values = f.evaluate([[0.0, 0.0], [1.0, -2.0]])

Load it in C++:

FuncCraft::FunctionSpec spec =
    FuncCraft::load_function_spec("my_function.yaml");
FuncCraft::BenchmarkFunction f(spec);

Function fields

dimension and domain: Ambient/search dimension and bounds.
components: A list of primitive or nested components. A primitive component has base_function. A nested component has composed_function.
coordinate_transform: Maps the parent point into the component input. input_dimension is the parent dimension. output_dimension is the component dimension. For block rotation, selected_indices chooses the subspace.
value_transform: Optional scalar transform of the nonnegative component value.
composition: Combines component values. DPM compositions may also specify full dimensional centers and component biases.
assigned_xopt and assigned_fopt: Desired optimum location and value for the constructed function.
scale_factor: Final value scale. Set it to a positive number for direct control, or omit it to let FuncCraft estimate it.
seed, label, metadata: Optional generation and bookkeeping fields.

Names and aliases

Spec parsers normalize names before matching: case, spaces, hyphens, and underscores are ignored. These inputs are equivalent:

kind: dpm-bgsoftmax
kind: dpmbgsoftmax
kind: DPM BG Softmax

For the full list of mechanisms and parameter conventions, see Implemented mechanisms.