Laravel Reductor

A powerful Laravel package for detecting and reducing test redundancy using machine learning clustering algorithms. Reductor analyzes your test suite to identify similar tests that provide overlapping coverage, helping you maintain a leaner, more efficient test suite.

Features

• ML-Powered Analysis: Uses advanced clustering algorithms (DBSCAN, Hierarchical, K-means) to identify redundant tests
• Coverage-Based Detection: Analyzes code coverage patterns to find tests with similar execution paths
• Semantic Understanding: Combines coverage analysis with semantic similarity of test code
• Safety Validation: Prevents merging of semantically opposing tests (e.g., success vs failure tests)
• Multiple Output Formats: Markdown (default), JSON, YAML, and HTML reports
• IDF-Weighted Coverage: Uses Inverse Document Frequency to emphasize unique coverage patterns
• Configurable Thresholds: Adjust similarity thresholds to control reduction aggressiveness

Installation

1. Install the package via Composer:

composer require mottaviani-dev/laravel-reductor

2. Publish the configuration file:

php artisan vendor:publish --provider="Reductor\ReductorServiceProvider"

3. Run migrations to create the necessary database tables:

php artisan migrate

4. Ensure Python 3.8+ is installed with required packages (alpine based images):

apk add --no-cache build-base python3-dev py3-pip py3-wheel py3-setuptools py3-numpy cmake

pip install -r vendor/mottaviani-dev/laravel-reductor/requirements.txt

Quick Start

1. Generate Coverage Data

First, generate PHPUnit coverage in text format:

XDEBUG_MODE=coverage ./vendor/bin/phpunit --coverage-php=storage/coverage.cov

2. Ingest Coverage Data

Import the coverage data into Reductor:

php artisan reductor:ingest-coverage storage/coverage.cov

This will output a test run ID that you'll use for analysis.

3. Analyze for Redundancy

Run the redundancy detection (K-Means algorithm and Markdown output are defaults):

php artisan tests:reduce <test-run-id>

Usage Examples

Basic Usage with Different Algorithms

# Using K-means (default)
php artisan tests:reduce 46 --algorithm kmeans

# Using DBSCAN (good for varied density data)
php artisan tests:reduce 46 --algorithm dbscan

# Using Hierarchical clustering
php artisan tests:reduce 46 --algorithm hierarchical

Output Formats

# Markdown report (default, human-readable)
php artisan tests:reduce 46

# JSON format (for programmatic use)
php artisan tests:reduce 46 --format json

# HTML report
php artisan tests:reduce 46 --format html --output report.html

# YAML format
php artisan tests:reduce 46 --format yaml

Adjusting Similarity Thresholds

# Conservative (95% similarity required)
php artisan tests:reduce 46 --threshold 0.95

# Balanced (85% default)
php artisan tests:reduce 46

# Aggressive (70% similarity)
php artisan tests:reduce 46 --threshold 0.7

Combined Options

# Full example with all options
php artisan tests:reduce 46 \
  --algorithm dbscan \
  --threshold 0.85 \
  --format markdown \
  --output storage/reductor/analysis.md

Configuration

Edit config/reductor.php to customize default settings:

return [
    'analysis' => [
        // Default clustering algorithm
        'algorithm' => env('REDUCTOR_ALGORITHM', 'kmeans'),
        
        // Similarity thresholds
        'similarity_thresholds' => [
            'conservative' => 0.95,  // High confidence
            'balanced' => 0.85,      // Default
            'aggressive' => 0.75,    // More reduction
        ],
        
        // DBSCAN parameters
        'dbscan' => [
            'eps' => null,  // Auto-detect
            'min_samples' => 3,
            'metric' => 'euclidean',
        ],
    ],
    
    'coverage' => [
        // Coverage file search paths
        'auto_detect_paths' => [
            storage_path('coverage.txt'),
            storage_path('coverage.cov'),
            base_path('coverage/coverage.txt'),
        ],
    ],
];

Understanding the Algorithms

K-means (Default)

• Centroid-based clustering that automatically finds optimal number of clusters
• Most validated algorithm in research literature (35% of studies)
• Fast and deterministic with consistent results across runs
• Works well when test clusters have similar sizes

DBSCAN

• Density-based clustering that can identify noise points and outliers
• Good for varied density when test clusters have very different sizes
• Adaptive parameters based on dataset characteristics

Hierarchical Clustering

• Structure-aware clustering that builds a tree of clusters
• Useful for understanding hierarchical relationships between test groups
• Good alternative when you need dendrogram visualization

Interpreting Results

Redundancy Scores

• 95-100%: Nearly identical tests - safe to remove
• 85-95%: Very similar tests - review recommended
• 70-85%: Related tests - manual review required
• Below 70%: Different tests - keep both

Priority Levels

• High: Tests with >90% similarity in large clusters
• Medium: Tests with 70-90% similarity
• Low: Tests with <70% similarity or small clusters

Coverage Overlap

• Shows percentage of shared code coverage between tests
• 100% overlap doesn't always mean redundant (different assertions possible)
• Combined with semantic similarity for accurate detection

Advanced Features

Safety Validation

Reductor automatically prevents merging tests with opposing semantics:

• Success vs Failure tests
• Valid vs Invalid tests
• Create vs Delete operations
• Authorized vs Unauthorized tests

IDF-Weighted Coverage

Lines covered by many tests are weighted less than unique coverage patterns, improving discrimination between tests that share common initialization code.

Dimensionality Reduction

Automatically reduces high-dimensional vectors (640D) to manageable size (128D) while preserving 95%+ variance for accurate clustering.

Troubleshooting

"Pipeline failed" Error

• Check Python is installed: python3 --version
• Verify ML dependencies: pip list | grep -E "scikit-learn|numpy"
• Check logs: tail -50 storage/logs/laravel.log

Poor Clustering Results

• Ensure coverage data is comprehensive
• Try different algorithms (DBSCAN or Hierarchical)
• Adjust threshold based on your needs
• Check for semantic vector issues (all zeros)

Example Output (Markdown)

# Test Redundancy Analysis Report

## Summary
- Total Redundant Tests: 23
- High Priority: 1 findings
- Medium Priority: 2 findings
- Low Priority: 4 findings

## High Priority Findings

### Cluster 1
**Redundancy Score**: 96%
**Representative Test**: UserLoginTest::test_successful_login
**Redundant Tests** (3):
- UserAuthTest::test_user_can_login
- LoginControllerTest::test_login_success
- AuthenticationTest::test_valid_credentials

Contributing

Please see CONTRIBUTING.md for details.

License

The MIT License (MIT). Please see License File for more information.