Feature Selection via Genetic Algorithm (FSGA)

A university project implementing feature selection using Genetic Algorithms, with evaluation and visualization tools.

Quick Start

# Installation
git clone https://github.com/straightchlorine/fsga.git
cd fsga
uv venv && source .venv/bin/activate
uv pip install -e .

# Run example
python experiments/run_experiment.py

Basic Usage

from fsga.core.genetic_algorithm import GeneticAlgorithm
from fsga.datasets.loader import load_dataset
from fsga.evaluators.accuracy_evaluator import AccuracyEvaluator
from fsga.ml.models import ModelWrapper

# Load data and setup
X_train, X_test, y_train, y_test, _ = load_dataset('iris', split=True)
model = ModelWrapper('rf', n_estimators=50, random_state=42)
evaluator = AccuracyEvaluator(X_train, y_train, X_test, y_test, model)

# Run GA
from fsga.selectors.tournament_selector import TournamentSelector
from fsga.operators.uniform_crossover import UniformCrossover
from fsga.mutations.bitflip_mutation import BitFlipMutation

ga = GeneticAlgorithm(
    num_features=X_train.shape[1],
    evaluator=evaluator,
    selector=TournamentSelector(evaluator, tournament_size=3),
    crossover_operator=UniformCrossover(),
    mutation_operator=BitFlipMutation(probability=0.01),
    population_size=50,
    num_generations=100,
    early_stopping_patience=10
)

results = ga.evolve()
print(f"Accuracy: {results['best_fitness']:.2%}")
print(f"Features: {results['best_chromosome'].sum()}/{X_train.shape[1]}")

Key Features

• Modular Design: Swappable operators, selectors, and evaluators
• Multiple Operators: 4 crossover operators, 4 selection strategies, 3 fitness functions
• Baseline Comparisons: Built-in RFE, LASSO, Mutual Information, Chi², ANOVA
• Statistical Testing: Wilcoxon, Mann-Whitney, Cohen's d, Jaccard stability
• Visualization: 9 plot functions for analysis and comparison
• Experiment Framework: ExperimentRunner for reproducible experiments
• Configuration: YAML-based configuration system

Architecture

fsga/
├── core/          # GA engine (genetic_algorithm, population)
├── operators/     # Crossover: uniform, single-point, two-point, multi-point
├── mutations/     # Mutation: bitflip
├── selectors/     # Selection: tournament, roulette, ranking, elitism
├── evaluators/    # Fitness: accuracy, F1, balanced accuracy
├── ml/            # Model wrappers (sklearn integration)
├── datasets/      # Dataset loaders (iris, wine, breast_cancer, digits)
├── analysis/      # Baselines + ExperimentRunner
├── visualization/ # 9 plot functions
└── utils/         # Config, metrics, serialization, logging

Documentation

• Getting Started - Installation and basic usage
• Tutorial - Step-by-step guide with examples
• Architecture - System design and extension points
• Project Plan - Status and roadmap
• Module READMEs - See fsga/*/README.md for component details

Example Results

Breast Cancer Dataset (30 features → 14 features):

• GA Accuracy: 98.3% with 48% of features
• All Features: 95.7% with 100% of features
• +2.6% accuracy, 52% dimensionality reduction

Iris Dataset (4 features → 2.6 features):

• GA Accuracy: 97.7% with 65% of features
• Selected: petal length, petal width

Wine Dataset (13 features → 7.2 features):

• GA Accuracy: 100% with 55% of features

Running Experiments

# Full analysis (iris, wine, breast_cancer; 6 plots per dataset)
python experiments/run_experiment.py

# Quick test (single dataset, fewer runs)
python experiments/run_experiment.py --quick

# Specific datasets only
python experiments/run_experiment.py --datasets iris wine

# Without visualizations (faster)
python experiments/run_experiment.py --no-plots

# Results saved to: results/{mode}/{dataset}/

Tests

# Run all tests
uv run pytest tests/ -v

# With coverage
uv run pytest tests/ --cov=fsga --cov-report=html

# Current: 290 tests (about 67% line coverage)

Configuration

Example config (configs/default.yaml):

population_size: 50
num_generations: 100
mutation_rate: 0.01
crossover_rate: 0.8
early_stopping_patience: 10

dataset:
  name: iris
  test_size: 0.2

Load with:

from fsga.utils.config import Config
config = Config.from_file('configs/default.yaml')

License

MIT License - see LICENSE file for details.