AutoML Pipeline

A simplified machine learning automation framework that handles the complexities of ML workflows through configuration. The project provides an easy-to-use interface for training machine learning models while automatically managing optimization, visualization, and training processes.

Important

This framework is designed for rapid ML prototyping and experimentation. For production deployments, please review the performance metrics and model validation sections carefully.

🚀 Quick Start

from lightning_auto import AutoML
from config import get_classification_config

# Get configuration
config = get_classification_config()

# Initialize and train
auto_ml = AutoML(config)
auto_ml.fit(train_data, val_data)

Tip

Check the examples/ directory for complete notebook demonstrations of common use cases.

✨ Key Features

• Automated Training: Simplified training process with minimal user intervention
• Configuration Templates: Easy-to-use configuration templates for common ML tasks
• Visualization: Built-in visualization tools for performance analysis

📊 Performance Metrics

Note

These metrics represent baseline performance and may vary based on your dataset and configuration.

• Training Loss: 1.6422
• Validation Loss: 1.6169
• Learning Rate: 0.000896

📈 Visualization Analysis

1. Class Distribution

Note

The distribution plot shows:

• Class 2 dominates with ~50 samples
• Class 0 follows with ~27 samples
• Class 4 has the least representation with ~20 samples
• Clear sign of class imbalance that may need addressing

2. Classification Results

Note

Key observations from the confusion matrix:

• Strong diagonal pattern indicates good overall classification
• Class 2 shows the highest confidence with 10–14 correct predictions
• Some classifications between neighboring classes
• Class 3 shows room for improvement in discrimination

3. Training Dynamics

Note

The loss plot reveals:

• Training loss (blue) shows healthy fluctuation between 1.54–1.70
• Validation loss (orange) maintains stability around 1.62
• No significant overfitting as validation loss remains stable
• Good model convergence with occasional exploration spikes

4. Learning Rate Schedule

Note

The learning rate schedule demonstrates:

• Smooth cosine decay from 2e-3 to 9e-4
• Gradual learning rate reduction for fine-tuning
• Proper annealing behavior for optimization stability

📁 Project Structure

automl/
├── lightning_auto.py   # Core AutoML engine
├── config.py          # Configuration templates
├── train.py          # Training script
├── WriterSide/       # Documentation
└── examples/         # Example notebooks (Coming soon!)

⚙️ Configuration Templates

Warning

Always validate configuration parameters against your specific use case before training.

Linear Classification Example

config = {
    "model": {
        "type": "classification",
        "input_dim": 10,
        "output_dim": 5,
        "task": "classification"
    },
    "training": {
        "learning_rate": 0.002,
        "epochs": 30
    }
    # ... other parameters
}

🤝 Contributing

Caution

Before submitting large changes, please open an issue to discuss the proposed modifications.

Adding New Configuration Templates

1. Create a new function in config.py:

def get_custom_config():
    return {
        "model": {
            # model specifications
        },
        "training": {
            # training parameters
        }
    }

1. Add documentation and example usage
2. Submit a pull request

Backend Development

1. For new features:
   • Fork the repository
   • Create feature branch
   • Add tests
   • Submit pull request

🎯 Logo

📝 License

This project is licensed under the MIT License—see the LICENSE file for details.