A Python Library for Classical and Dynamical Low-Rank CMA-ES

---

General 🌎

seamaze is a Python library for classical and Dynamical Low-Rank (DLR) CMA-ES variants. It is designed to navigate complex, high-dimensional fitness landscapes by iteratively adapting a multivariate Gaussian search space to the objective's local topography. By leveraging DLR approximations, seamaze remains computationally efficient even on ill-conditioned or rugged black-box problems. This implementation further extends to the integration of first-order information, constraints, and robust restart mechanisms.

Installation 💻

Python distribution

You can install the latest distribution via:

pip install seamaze

Source code

You can check the latest source code via:

git clone https://github.com/pyanno4rt/seamaze.git

Usage

seamaze has two main classes which provide a classical and a dynamical low-rank CMA-ES variant:

Classical CMA-ES

from seamaze.optimizers.evolutionary import CMAES

Dynamical low-rank CMA-ES

from seamaze.optimizers.low_rank import DLRCMAES

Dependencies

Name	Version
python	>=3.11, <4.0
numpy	>=2.4.4
scipy	>=1.17.1
numba	>=0.65.0
matplotlib	>=3.10.8
seaborn	>=0.13.2

Development 🚀

Important links

• Github
• PyPI
• Coveralls
• Issue tracker

Help and Support 👥

Resources

• Github Discussions
• Github Issues

Contact

• Github Page
• Mail
• LinkedIn

Citation

To cite seamaze, either use the link in the right sidebar of the Github landing page labeled "Cite this repository" or copy the short-form bib-style paragraph below:

@software{seamaze,
  title = {{seamaze}: a python library for classical and dynamical low-rank CMA-ES},
  author = {Ortkamp, Tim and Patwardhan, Chinmay and Stammer, Pia},
  version = {0.0.2},
  license = {MIT},
  year = {2026},
  publisher = {GitHub},
  url = {https://github.com/pyanno4rt/seamaze}
}