ENGA — Enhanced Networked Genetic Algorithm

A Python implementation of networked genetic algorithms that use genotypic and phenotypic similarity to form scale-free inter-chromosomal links, helping balance exploration and exploitation during evolution.

Animated evolution of node rankings across generations. Node colors indicate fitness ranking: green (top 2), teal (next 10), blue-green (next 15), and dark blue (others). Labels show node scores.

This repository contains several GA variants (including ENGA, NGA, GA, CRGA, DCGA) and a small benchmarking harness to compare them on common test functions.

Quick start

1. Clone the repository:

git clone https://github.com/yazid-hoblos/ENGA.git
cd ENGA

2. Install requirements:

pip install -r requirements.txt

3. Run the main benchmark script:

python src/main.py --run all

Command-line options

• --run — select which benchmark(s) to execute. Accepts one or more of the following values: enhanced, networked, ga, crga, dcga, or all.

• --save-metrics — optional flag. When present, the benchmarks will save metrics (fitness, diversity, selection pressure, etc.) into logs/metrics/. This flag is off by default.

• --generations N — set the number of generations (overrides per-benchmark default). Default (if omitted): 2000.

• --population-size N — set the population size for all benchmarks. Default (if omitted): 125 (124 for CRGA in original config).

• --number-elites N — number of elites to keep. Default (per-benchmark): 0 for most algorithms, 10 for the plain GA.

• --crossover-prob P — crossover probability (0.0–1.0). Default: 0.95.

• --mutation-prob P — mutation probability (0.0–1.0). Default: 0.10.

• --nruns N — number of independent runs to perform. Default: 1.

• --no-verbose — disable verbose output. Default: verbose on.

Examples:

# run all benchmarks
python src/main.py --run all

# run only the enhanced networked GA
python src/main.py --run enhanced

# run multiple specified benchmarks (enhanced and plain ga)
python src/main.py --run enhanced ga

Files and project layout

Key files (located in src/):

• main.py — example entry point and benchmark orchestration.
• genetic_algorithm.py — base GA class and shared utilities.
• networked_genetic_algorithm.py, enhanced_networked_genetic_algorithm.py, networked_assignment.py — networked GA variants and helpers.
• benchmark.py, benchmark_functions.py — benchmarking harness and test functions.
• crossover.py, mutation.py, selection.py, networked_selection.py — GA operators.
• utils/ — small utility modules (math.py, drawable.py).

Output and logs:

• logs/metrics/ — saved metrics (fitness, diversity, selection pressure, ...)
• logs/results/ — benchmark run outputs and archived results

Metrics & logs

This project can save a rich set of metrics and visualizations from each benchmark run. Metrics are written under logs/metrics/ when enabled (see below). Keep in mind metrics (especially animated GIFs) can consume disk and I/O during long runs — enable them only when you need the outputs.

Key Components

• logs/metrics/fitness/ — PNG plots of average and best fitness over generations.
• logs/metrics/phenotype_diversity/ — PNG plots showing phenotype diversity over time.
• logs/metrics/genotype_diversity/ — PNG plots for genotype-space diversity.
• logs/metrics/selection_pressure/ — PNG plots tracking selection pressure each generation.
• logs/metrics/degree/, logs/metrics/degree_exponent/, logs/metrics/clustering_coefficient/, logs/metrics/path_length/ — network-related PNG visualizations.
• logs/metrics/number_of_nodes_to_remove/ — PNG plots for node-removal statistics.
• logs/metrics/animated_networks/ — optional animated network visualizations saved as GIF files.

How to enable saving

• From the command line: pass --save-metrics to src/main.py to enable metric saving for benchmark runs (this toggles the save_metrics flag when Benchmark objects are created).
• Programmatically: set save_metrics=True when constructing a Benchmark(...) or call DrawManager.draw_animated_network(history, save=True) to save an animation explicitly.

Network Animation

As highlighted in the animation above, in the network architecture, high‑fitness individuals migrate from the periphery to the center as evolution creates hubs — node color shows fitness and centrality reflects accumulated connections. This highlights selection pressure, hub formation, and convergence.

Typical usage example

from benchmark import Benchmark
from benchmark_functions import get_all_functions
from enhanced_networked_genetic_algorithm import EnhancedNetworkGeneticAlgorithm

benchmark_functions = get_all_functions()
enhanced_nga_benchmark = Benchmark(
    genetic_algorithm=EnhancedNetworkGeneticAlgorithm,
    number_of_runs=1,
    benchmark_functions=benchmark_functions,
    number_of_generations=2000,
    population_size=125,
    number_elites=0,
    probability_crossover=0.95,
    probability_mutation=0.1,
    verbose=True,
    save_metrics=True,
)

enhanced_nga_benchmark.run()

Requirements

See requirements.txt for Python package dependencies. The project expects Python 3.8+ (recommend 3.10+).

Contributing

Contributions are welcome. If you'd like to contribute:

1. Fork the repo and create a feature branch.
2. Add tests or examples for new behaviour when applicable.
3. Open a pull request describing your change.

If you plan to work on an issue, comment on it to avoid duplicated effort.

Authors

• @Murf-y
• @yazid-hoblos

Acknowledgements

• Numpy
• Networkx
• Pyvis