Estimation of Distribution Algorithms Framework (EDAF)

EDAF is a research-grade Java framework for Estimation of Distribution Algorithms (EDAs) and adjacent evolutionary optimization workflows. It is designed for teams that need reproducibility, deep telemetry, algorithmic extensibility, and publishable benchmarking outputs.

Documentation portal: https://edaf.readthedocs.io/
Repository docs index: docs/index.md

Why EDAF

• Strongly typed, modular architecture across algorithms, models, representations, problems, execution, persistence, reporting, and web.
• Reproducible by design: deterministic seed handling, component-level RNG streams, checkpoint/resume.
• Research observability: latent model insights, drift/diversity metrics, adaptive-event timeline, experiment-level analytics.
• Production-grade operational tooling: CLI, structured logging, DB-backed run explorer, static reports, Docker stack.
• Extensible plugin system (ServiceLoader) for custom algorithms, models, representations, and problems.

Architecture At A Glance

graph LR
    CFG["YAML Config"] --> CLI["edaf CLI"]
    CLI --> RUNNER["Experiment Runner"]
    RUNNER --> REG["Plugin Registry"]

    REG --> ALG["Algorithm"]
    REG --> MODEL["Model"]
    REG --> PROB["Problem"]
    REG --> REP["Representation"]

    ALG --> BUS["Event Bus"]
    BUS --> SINKS["Console / File / JSONL / DB Sinks"]
    SINKS --> DB[("SQLite / PostgreSQL")]

    DB --> WEB["Web Dashboard + API"]
    DB --> REPORT["HTML / LaTeX Reporting"]

Loading

Supported Optimization Scope

Representations

• BitString
• IntVector (bounded)
• CategoricalVector
• MixedDiscreteVector
• RealVector (bounded/unbounded)
• MixedRealDiscreteVector
• PermutationVector
• VariableLengthVector
• GrammarBitString (grammar-based GP)

Models

• Discrete: Bernoulli/frequency vectors, dependency-tree and Bayesian-network families
• Continuous: diagonal and full Gaussian, mixture/KDE/copula/flow families
• Permutation: edge histogram and ranking-based models
• Strategy-model hybrids: NES and CMA-style updates

Algorithm Families

• Discrete EDAs: UMDA, PBIL, cGA, BMDA, MIMIC, BOA, hBOA, EBNA, Chow-Liu / dependency-tree variants, factorized variants
• Continuous EDAs/strategies: Gaussian EDA, GMM-EDA, KDE-EDA, Copula-EDA, Flow-EDA, CEM, IGO-style drivers, sNES, xNES, CMA-ES
• Permutation EDAs: EHM/EHBSA-style, Mallows and Plackett-Luce variants
• Advanced lines: noisy/dynamic aliases, multi-objective skeletons, adaptive runs with latent-knowledge triggers

Built-in Problem Coverage

• Continuous: Sphere, Rosenbrock, Rastrigin, CEC suite adapters, COCO/BBOB integration
• Discrete/combinatorial: OneMax, Knapsack, MAX-SAT, TSP/TSPLIB, disjunct matrix family (DM/RM/ADM)
• Multi-objective: ZDT and DTLZ baseline suites
• Symbolic/grammar: boolean tasks, regression from CSV, multiclass classification from CSV
• Crypto/boolean-function optimization suite

Quick Start

1) Build and test

mvn -q clean test

2) Run a first experiment

./edaf run -c configs/docs/web-screenshot-onemax.yml

3) Start local web dashboard

./scripts/run-web-local.sh

Open http://localhost:7070.

4) Generate run report

./edaf report --run-id <run-id> --out reports --db-url jdbc:sqlite:edaf.db

Web Dashboard Preview

Run explorer

Experiment analytics

Representation insights panel

Grammar/tree visualization

Research-Grade Analytics Out Of The Box

EDAF persists both run-level and experiment-level analytics, including:

• convergence summaries with confidence intervals
• success rate and success-vs-budget curves
• time-to-target histograms
• final-fitness box plots and ECDF
• diversity and drift diagnostics
• latent model insights (entropy/fixation/dependencies, permutation structure, real-valued uncertainty)
• adaptive-event timeline (what triggered, what action was applied)
• statistical comparison blocks (Wilcoxon/Friedman/Holm where configured)

Grammar-Based Symbolic Optimization

EDAF includes grammar-based symbolic regression/classification with:

• grammar.mode: auto and grammar.mode: custom (BNF)
• deterministic encoding from derivation decisions
• ephemeral random constants (ERC) with reproducible sampling
• tree rendering in web UI (AST + infix + DOT + LaTeX export)
• config suites in configs/grammar_gp_suite/

Guide: docs/foundations/grammar-based-gp.md

Operability and Reproducibility

• JSONL/CSV/file/DB sinks with structured payloads
• asynchronous persistence path for high-throughput runs
• cooperative stop controls for run/experiment from UI/API
• batch orchestration and campaign workflows
• deterministic checkpoint/resume
• Docker-based local stack (runner + DB + web)

Use EDAF As A Dependency

<dependencyManagement>
  <dependencies>
    <dependency>
      <groupId>io.github.karloknezevic</groupId>
      <artifactId>edaf-parent</artifactId>
      <version>${edaf.version}</version>
      <type>pom</type>
      <scope>import</scope>
    </dependency>
  </dependencies>
</dependencyManagement>

<dependencies>
  <dependency>
    <groupId>io.github.karloknezevic</groupId>
    <artifactId>edaf-core</artifactId>
  </dependency>
  <dependency>
    <groupId>io.github.karloknezevic</groupId>
    <artifactId>edaf-algorithms</artifactId>
  </dependency>
  <dependency>
    <groupId>io.github.karloknezevic</groupId>
    <artifactId>edaf-problems</artifactId>
  </dependency>
</dependencies>

Integration guide: docs/guides/using-edaf-as-package.md

Academic Positioning and References

EDAF is built for reproducible optimization research pipelines and comparative studies. It supports benchmark protocol workflows (COCO/BBOB and suite-style campaigns), run replication, and exportable report artifacts suitable for publications.

Selected references:

1. Larrañaga, P., & Lozano, J. A. (Eds.). Estimation of Distribution Algorithms. Springer.
2. Mühlenbein, H., & Paass, G. From recombination of genes to the estimation of distributions I.
3. Mühlenbein, H., Bendisch, J., & Voigt, H. M. From recombination of genes to the estimation of distributions II.
4. Pelikan, M., Goldberg, D. E., & Cantú-Paz, E. BOA: the Bayesian optimization algorithm.

Full bibliography: docs/references/bibliography.md

Documentation Map

Core and architecture

• docs/index.md
• docs/foundations/architecture.md
• docs/foundations/configuration.md
• docs/foundations/cli-reference.md
• docs/foundations/algorithms.md
• docs/foundations/representations.md
• docs/foundations/extending-the-framework.md

Runtime and analytics

• docs/runtime/web-dashboard.md
• docs/runtime/database-schema.md
• docs/runtime/latent-insights.md
• docs/runtime/logging-and-observability.md
• docs/runtime/metrics-and-results.md

Benchmarks and suites

• docs/benchmarks/problem-suites.md
• docs/benchmarks/coco-integration.md
• docs/benchmarks/disjunct-matrix-problems.md
• docs/benchmarks/adm-paper-suite.md
• docs/benchmarks/crypto-boolean-problems.md
• docs/benchmarks/benchmark-comparisons.md
• docs/benchmarks/complexity-and-performance.md

Engineering and release

• docs/guides/getting-started.md
• docs/guides/usage-guide.md
• docs/guides/docker.md
• docs/engineering/testing-and-release.md
• docs/engineering/release-and-publishing.md
• docs/engineering/improvements.md
• docs/release-notes/index.md

API docs

• docs/api/javadoc-api.md

Author

Karlo Knezevic (2013). Evolucijski algoritmi temeljeni na vjerojatnosnim razdiobama (Croatian).
Master thesis, Nr. 540, Faculty of Electrical Engineering and Computing, University of Zagreb.
Google Scholar profile

License

This project is licensed under Apache License 2.0.