loguso benchmarking experiments

This archive contains the source code, benchmark definitions, experimental results, plotting utilities, and manuscript files for the study of LOGUSO, a mixed-variable surrogate optimization method for discretized real, integer, and categorical design spaces. The repository implements the LOGUSO optimizer together with supporting modules for uncertainty scoring, MIPT-based space-filling sampling, Pareto-style utilities, categorical COCO benchmark construction, fANOVA-based grouped sensitivity analysis, and large-scale benchmark comparison against several baseline optimizers.

The archive also includes the categorical extension of the COCO bbob-mixint benchmark suite, experiment runners for benchmark campaigns in dimensions d=24 and d=48, grouped fANOVA study scripts, parameter-study utilities, and postprocessing scripts used to generate aggregate comparison tables and figures. Result folders contain raw run traces, merged benchmark outputs, summary CSV files, configuration files, and pre-generated diagrams used in the paper.

The paper/ subfolder contains the LaTeX source of the manuscript, bibliography, figure files, and the compiled PDF. Together, the archive is intended to support transparency, reproducibility, and reuse of both the proposed optimization method and the accompanying categorical benchmark experiments.

In particular:

• loguso.py and supporting Python modules: implementation of the LOGUSO optimizer and helper routines.
• coco_categorical.py, benchmarks.py: categorical COCO benchmark construction and benchmark definitions.
• mixed_parallel_benchmark_runner.py, fanova_loguso.py, related scripts: experiment execution and analysis pipelines.
• benchmark_results_*, fanova_*: raw and processed results, plots, and comparison summaries.
• paper/: manuscript source, figures, bibliography, and compiled article PDF.