ML Debugging Notebooks

A comprehensive collection of Jupyter notebooks demonstrating how to debug common machine learning problems. Each notebook provides hands-on examples, visualizations, and practical solutions to help you identify and fix issues in your ML models.

📚 Notebooks

1. Overfitting and Underfitting Detection

Learn to identify and fix the bias-variance tradeoff problems:

• Topics Covered:
  • Detecting overfitting vs underfitting using learning curves
  • Impact of model complexity on performance
  • Regularization techniques (L1, L2, Ridge)
  • Proper model selection and validation
• Key Techniques: Learning curves, cross-validation, regularization

2. Data Leakage Detection and Prevention

Understand and prevent data leakage that inflates model performance:

• Topics Covered:
  • Target leakage identification
  • Train-test contamination
  • Temporal leakage in time-series
  • Proper data splitting and preprocessing
• Key Techniques: Correlation analysis, pipeline usage, temporal validation

3. Vanishing and Exploding Gradients

Debug gradient problems in deep neural networks:

• Topics Covered:
  • Identifying vanishing/exploding gradients
  • Weight initialization strategies
  • Batch normalization
  • Gradient clipping
• Key Techniques: Gradient monitoring, proper initialization (He/Xavier), ReLU activation

4. Class Imbalance Issues

Handle imbalanced datasets effectively:

• Topics Covered:
  • Detecting class imbalance problems
  • Resampling techniques (SMOTE, undersampling)
  • Class weights and cost-sensitive learning
  • Appropriate metrics for imbalanced data
• Key Techniques: SMOTE, class weights, ROC-AUC, precision-recall curves

5. Feature Scaling and Normalization Problems

Master feature scaling for better model performance:

• Topics Covered:
  • When and why to scale features
  • Different scaling techniques (StandardScaler, MinMaxScaler, RobustScaler)
  • Impact on various algorithms
  • Common scaling mistakes
• Key Techniques: StandardScaler, MinMaxScaler, RobustScaler, proper pipeline usage

🚀 Getting Started

Prerequisites

• Python 3.8 or higher
• pip package manager

Installation

1. Clone this repository:

git clone https://github.com/macanderson/MLLanguageModels.git
cd MLLanguageModels

2. Install required dependencies:

pip install -r requirements.txt

3. Launch Jupyter Notebook:

jupyter notebook

4. Open any notebook from the notebooks/ directory and start learning!

📋 Requirements

The notebooks use the following main libraries:

• NumPy - Numerical computing
• Pandas - Data manipulation
• Scikit-learn - Machine learning algorithms
• TensorFlow - Deep learning (for gradient problems)
• Matplotlib & Seaborn - Visualization
• imbalanced-learn - Handling imbalanced datasets

See requirements.txt for the complete list.

🎯 Learning Path

Recommended order for beginners:

1. Start with Overfitting/Underfitting to understand model performance basics
2. Learn Feature Scaling to prepare data properly
3. Study Data Leakage to avoid common pitfalls
4. Tackle Class Imbalance for real-world scenarios
5. Explore Gradient Problems for deep learning applications

For experienced practitioners:

• Jump to any notebook based on your current debugging needs
• Each notebook is self-contained with complete examples

🔍 What You'll Learn

Each notebook follows a consistent structure:

1. Problem Overview - What the issue is and why it matters
2. Symptoms - How to recognize the problem
3. Hands-on Examples - Code demonstrating the problem
4. Solutions - Multiple approaches to fix the issue
5. Best Practices - Guidelines to prevent future occurrences
6. Debugging Checklist - Quick reference for troubleshooting
7. Exercises - Practice problems to reinforce learning

💡 Use Cases

These notebooks are perfect for:

• Students learning machine learning fundamentals
• Data Scientists debugging model performance issues
• ML Engineers implementing production-ready models
• Researchers understanding common pitfalls
• Interview Preparation for ML/DS roles

🤝 Contributing

Contributions are welcome! If you have suggestions for:

• Additional debugging scenarios
• Improved explanations
• New visualization techniques
• Bug fixes

Please open an issue or submit a pull request.

📝 License

This project is open source and available under the MIT License.

📧 Contact

For questions or feedback, please open an issue on GitHub.

---

Happy Debugging! 🐛🔧