ADR-EDA TFM Public Package

This repository is the public, cleaned reproducibility package associated with the Master's thesis:

ADR-EDA: una arquitectura adaptativa de estimacion de distribuciones para optimizacion multiobjetivo dinamica

The package contains the code and consolidated evidence needed to inspect the main experimental blocks of the thesis without exposing private working files, raw data dumps, tutor-review material, or copyrighted papers.

Scope

ADR-EDA is a reliability-aware controller for dynamic multiobjective estimation of distribution algorithms. The thesis evaluates it in three stages:

• dynamic multiobjective benchmark functions;
• electric free-floating carsharing, including RealCar-Munich;
• station-based bike-sharing with real observed trips.

This public package is intended for technical inspection, reproducibility, and future integration work. It is not a full historical workspace dump.

Repository Layout

• src/free_floating_robust/: simulator, planner, evaluator, ADR controller bridges, and carsharing domain code.
• tests/edas/: benchmark runners and collection scripts used for dynamic multiobjective comparisons.
• tests/model/: applied validation scripts and OSRM helper utilities.
• scripts/: data-preparation, campaign execution, and artifact-generation scripts for the applied studies.
• docs/benchmark_results/final_thesis_pack_2026_06_28/: consolidated benchmark evidence.
• docs/carsharing_tfm/final_thesis_pack_2026_06_02/: consolidated carsharing and RealCar-Munich tables.
• docs/realbike_dmoo/final_thesis_pack_2026_06_02/: final RealBike tables and figures used in the thesis.
• docs/realbike_dmoo/final_realbike_package_2026_05_25/: complementary event-replay RealBike evidence.
• external/edaspy_extension/: patch documenting the local EDAspy extensions used during the thesis experiments.
• docker/: minimal container setup for running the public package.

What Is Not Included

The public package deliberately excludes:

• raw mobility datasets;
• internal tutor-review notes;
• weekly progress reports;
• local backups and intermediate exploratory runs;
• downloaded papers and third-party binaries;
• private forms or administrative documents.

Some long campaigns therefore cannot be reproduced from scratch with only this repository. The frozen outputs used in the thesis are included so that the reported tables and figures can be inspected.

Basic Usage

Use Python 3.11 or newer. From the repository root:

PYTHONPATH=src python3 tests/edas/clean_dynamic_sota_comparison.py --help
PYTHONPATH=src python3 scripts/run_realcar_freefloating_mini_campaign.py --help
PYTHONPATH=src python3 scripts/run_real_bikeshare_station_benchmark.py --help

Some workflows require EDAspy, OSRM-derived matrices, or external raw datasets. See REPRODUCIBILITY.md for details.

EDAspy Extension

The EDAspy checkout used during the thesis is not vendored as a full third-party repository. Instead, the local thesis-side changes are provided as a patch in:

external/edaspy_extension/patches/adr_eda_edaspy_extension.patch

This is the recommended format for review or later incorporation into the CIG version of EDAspy.

Docker

A minimal Docker setup is included under docker/:

docker compose -f docker/docker-compose.yml up --build -d

The image installs the public package dependencies. EDAspy-backed workflows still require an EDAspy checkout to be mounted or installed separately.

Citation

If you use this material, please cite the thesis and, when relevant, the derived CAEPIA 2026 paper. Citation metadata is provided in CITATION.cff.