Genetic Algorithm Test Generator

A Python tool that uses genetic algorithms to generate and evolve date-validation test cases. It runs candidate dates through selection, crossover, and mutation to maximize category coverage, then refines the result with a local search pass.

Features

• Generates date-validation test cases with a genetic algorithm.
• Supports seeded and non-seeded population strategies.
• Uses frequency-based fitness scoring to reward rare and uncovered categories.
• Applies local search after the GA phase to improve final coverage.
• Covers 21 valid, boundary, and invalid date categories.
• Exports evolved test cases to CSV.
• Prints coverage progress in the terminal.
• Plots GA and local-search coverage with Matplotlib.

How It Works

Phase	Description
Initialization	Builds a population of (day, month, year) candidates
Fitness	Scores candidates by category rarity and coverage value
Selection	Uses rank-based selection to preserve stronger candidates
Crossover	Swaps month and year genes between parent pairs
Mutation	Applies small random changes to candidate dates
Injection	Adds random dates during evolution to keep population diversity
Local Search	Refines the best population after GA convergence
Export	Writes the final sorted test cases to CSV

Date Categories

The generator targets 21 categories across boundary, valid, and invalid dates.

Boundary - 01/01/0000, 31/12/9999, 29/02/2020, 28/02/1900, 31/01/0000
Valid    - 31-day month, 30-day month, 28-day Feb, 29-day Feb (leap), other
Invalid  - day < 1, day > 31, month < 1, month > 12, year out of range,
           exceeded month length, exceeded February rules, and other invalid dates

Tech Stack

Part	Tech
Language	Python
Algorithm	Genetic Algorithm + Local Search
Plotting	Matplotlib
Output	CSV
Domain	Search-Based Software Testing

Screenshots

Seeded Genetic Algorithm Run

Coverage Progress Graph

Generated Test Cases CSV

Project Structure

.
|-- seeded_genetic_algorithm_date_tests.py       # Boundary-seeded GA run
|-- non_seeded_genetic_algorithm_date_tests.py   # Fully random GA run
|-- requirements.txt                             # Python dependencies
|-- sample_output/
|   `-- seeded_evolved_test_cases.csv            # Example CSV output
|-- assets/                                      # README screenshots
`-- README.md

Install Dependencies

pip install -r requirements.txt

Run Locally

Run the seeded version:

python seeded_genetic_algorithm_date_tests.py

Run the non-seeded version:

python non_seeded_genetic_algorithm_date_tests.py

Both scripts will:

1. Print generation-by-generation coverage.
2. Run a local search refinement pass.
3. Print the top evolved test cases with fitness scores.
4. Export the final evolved test suite to CSV.
5. Display a Matplotlib coverage chart.

Sample Output

Terminal:

Generation   0 => Coverage: 38.10%
Generation   1 => Coverage: 61.90%
Generation   2 => Coverage: 76.19%
Generation   3 => Coverage: 90.48%
Generation   4 => Coverage: 95.24%

Stopped at generation 5 with coverage: 100.00%

Top 15 Evolved Test Cases:
 1. 01/01/0000 => Boundary: 01/01/0000 (fitness=10.500)
 2. 31/12/9999 => Boundary: 31/12/9999 (fitness=10.500)
 3. 29/02/2020 => Boundary: 29/02/2020 (fitness=10.500)

CSV:

Date,Category,Fitness
01/01/0000,Boundary: 01/01/0000,10.500
31/12/9999,Boundary: 31/12/9999,10.500
29/02/2020,Boundary: 29/02/2020,10.500
31/01/2023,Valid: 31-day Month,10.250
00/06/2021,Invalid: Day < 1,10.333

See sample_output/seeded_evolved_test_cases.csv for a full example.

Configuration

Both scripts expose tuning values inside run_genetic_algorithm() and local_search().

Parameter	Seeded Default	Non-Seeded Default	Description
pop_size	500	500	Number of individuals per generation
coverage_target	0.95	0.97	Stop GA when this coverage ratio is reached
max_generations	1000	100	Hard cap on GA iterations
iterations	500	1000	Local search refinement steps

Seeded vs Non-Seeded

	Seeded	Non-Seeded
Starting population	Random dates plus known boundaries	Fully random dates
Fitness bonus	Rewards rare categories	Rewards rare categories more aggressively
Coverage target	95%	97%
Convergence speed	Faster	Slower
Use case	Boundary-aware testing	Black-box discovery