GLOBe is a collection of global optimization algorithms implemented in C++ and linked with Python. It also includes a set of analytical benchmark functions and a random function generator (PyGKLS) to test the performance of these algorithms.
- AdaLIPO+
- AdaRankOpt
- Bayesian Optimization
- CMA-ES
- Controlled Random Search
- DIRECT
- Every Call is Precious
- Multi-Level Single-Linkage
- Social Only Particle Swarm Optimization
- Langevin dynamics
- Stein Boltzmann Sampling
- Consensus Based Optimization
- Common noise variants of McKean-Vlasov dynamics
- Gradient Descent
- Multi-start Gradient Descent
- Pure Random Search
The documentation is available at gaetanserre.fr/GLOBe.
Install the package via pip from PyPI:
pip install globe-optiAlternatively, download the corresponding wheel file from the releases and install it with pip:
pip install globe-opti-<version>-<architecture>.whlMake sure you have CMake (≥ 3.28), a c++ compiler, and the eigen3 library installed. Then clone the repository and run:
pip install . -vIt should build the C++ extensions and install the package. You can also build the documentation with:
cd docs
pip install -r requirements.txt
make htmlThis package can be used to design a complete benchmarking framework for global optimization algorithms, testing multiple algorithms on a set of benchmark functions. See test_globe.py for an example of how to use it.
The global optimization algorithms can also be used independently. For example, to run the AdaLIPO+ algorithm on a benchmark function:
from globe.optimizers import AdaLIPO_P
from globe import create_bounds
f = lambda x: x.T @ x
opt = AdaLIPO_P(create_bounds(2, -5, 5), 300)
res = opt.minimize(f)
print(f"Optimal point: {res[0]}, Optimal value: {res[1]}")See test_optimizers.py for more examples of how to use the algorithms.
Contributions are welcome! Please see the CONTRIBUTING file for guidelines.
This is licensed under the GNU General Public License v3.0. See the LICENSE file for details.