BasinGraph is a geometry-controlled basin-graph optimization framework for nonconvex mathematical models.
basingraph/: proposed optimizer.baselines/: standardized benchmark implementations and wrappers.experiments/: COCO/BBOB and benchmark runners.scripts/: reproducible experiment and analysis entry points.protocols/: frozen benchmark protocols and manifests.reproducibility/: lightweight Source Data, summary tables, and metadata.environments/: dependency locks and CUTEst environment information.
Large COCO observer logs, CUTEst run-level histories, and full raw archives are not committed to GitHub. They are deposited on Zenodo at https://doi.org/10.5281/zenodo.20765047.
- Official noiseless COCO/BBOB: 24 functions, dimensions 2, 3, 5, 10 and 20, 15 instances, and a budget of 1,000d evaluations.
- Pre-registered CUTEst: 50 instances, seven algorithms, and 30 paired seeds.
- Independent MATLAB full-parallel validation and applied-mathematics tests.
Create the COCO environment from the archived dependency lock or install:
python -m pip install numpy scipy pandas matplotlib cma coco-experiment cocoppCUTEst requires SIFDecode, CUTEst, MASTSIF and PyCUTEst. See
protocols/ and environments/ for the frozen toolchain information.
Experiment entry points are documented in scripts/. The final public release
will contain a DOI-backed Zenodo archive with raw logs and convergence histories.
See CITATION.cff.
BasinGraph is released under the BSD-3-Clause license. See LICENSE.
- Version 1.0.0 DOI: https://doi.org/10.5281/zenodo.20765884
- All-versions DOI: https://doi.org/10.5281/zenodo.20765883
- Reproducibility Dataset DOI: https://doi.org/10.5281/zenodo.20765047
The official COCO/BBOB logs, pre-registered CUTEst results, MATLAB validation, Source Data and frozen protocols are openly available on Zenodo:
- Version-specific Dataset DOI: https://doi.org/10.5281/zenodo.20765047
- All-versions Dataset DOI: https://doi.org/10.5281/zenodo.20765046
The version-specific DOI identifies the exact files supporting the reported study.