Implementation of IEEE Congress of Evolutionary (CEC) Bound-constrained Single Objective Numerical Optimization benchmarks in C++23.
Important
This is pre-alpha stage software. Only CEC BC-SOP 2017 is supported. See below table for supported benchmarks.
The {cecxx} is a novel implementation of IEEE CEC BC-SOP 1 benchmarks in C++23, which compared to the official implementation
does not:
- leak memory
- involves multiple IO operations during each objective function call.
Solving the above problems results in reliable and fast implementation that is ready to be used in your numerical experiments.
Note
This is not official implementation of CEC BC-SOP benchmarks. However, it can be used as a drop-in replacement. The
{cecxx} is extensively tested to be compliant with the official implementations by emplyoing property-based testing. See
test/compliance for details.
To build and install the {cecxx} library, the following dependencies are required:
- LLVM clang++ (>=v19) or GNU g++ (>=v14)
- cmake (>=v3.19)
To build and run tests, you need googletest. Note that only LLVM's compiler can be used to build compliance tests.
| Operating system | Status |
|---|---|
| Linux | ✔️ |
| macOS 2 | ✔️ |
| Windows | 🔜 |
To build the {cecxx} library, run the following commands:
mkdir build
cmake -B build -S .
cmake --build buildIf you are using Justfile, then you can type:
just init buildSee Justfile recipes for details.
To build unit or compliance tests, run cmake with -DWITH_UNIT_TESTS and -DWITH_COMPLIANCE_TESTS options.
To set up your CMake project with the {cecxx}, add the following lines to the CMakeLists.txt:
# {cecxx} requires at least C++23
set(CMAKE_CXX_STANDARD 23)
set(CMAKE_CXX_STANDARD_REQUIRED ON)
find_package(cecxx REQUIRED)
...
target_link_libraries(<your-target> cecxx::cecxx)
#include <print>
#include <ranges>
#include <vector>
#include <cecxx/mdspan.hpp>
#include <cecxx/benchmark/evaluator.hpp>
namespace rn = std::ranges;
namespace rv = std::ranges::views;
using namespace cecxx::benchmark;
auto main() -> int {
const auto dimensions = std::vector{10uz, 30uz, 50uz, 100uz};
// Create an evaluator object for the CEC2017 benchmark
auto cec_2017 = evaluator(cec_edition_t::cec2017, dimensions, DATA_STORAGE_PATH);
// Create a problem grid [problem_number X dimension]
const auto problem_grid = rv::cartesian_product(dimensions, rv::iota(1, 30));
// Evaluate given input on each optimization problem from CEC2017/D{10, 30, 50, 100}
for (const auto &[dim, fn] : problem_grid) {
// Prepare an input, i.e., matrix [dim x 2]
const auto input = rv::repeat(0.0) | rv::take(2 * dim) | rn::to<std::vector<double>>();
const auto output = cec_2017(fn, cecxx::mdspan{input.data(), dim, 2});
std::println("dim = {}, fn = {}, output[0] = {}, output[1] = {}", dim, fn, output[0], output[1]);
}
}You can find other examples in example directory.
| CEC editiion | Status |
|---|---|
| 2013 | 🔜 |
| 2014 | 🔜 |
| 2015 | 🔜 |
| 2017 | ✔️ |
| 2019 | 🔜 |
| 2021 | 🔜 |
| 2022 | 🔜 |
| 2023 | 🔜 |
| 2024 | 🔜 |
| 2025 | 🔜 |
If you encounter any bugs, feel encouraged to create an issue or raise a PR with the fix.
If you are using {cecxx} as a part of your academic research or other activities,
please star the repository.