The comprehensive, production-focused ML course that teaches the 80-20 most important skills for ML engineers at companies

A hands-on learning system that teaches you to build production ML systems from scratch, covering both technical skills AND the business/collaboration skills that companies actually need. By the end, you'll be ready to contribute meaningfully to an ML team on day one.

Themes: Pokemon card type classification + price prediction Format: Interactive Marimo notebooks with exercises Duration: 20-24 hours over 3-4 weeks Outcome: Production-ready ML engineering skills + business acumen + team collaboration skills

• When to use ML (and when not to)
• ROI calculation and business metrics
• Stakeholder communication
• Setting realistic expectations
• Status: ✅ Complete with 5 real-world scenarios

• Data loading, validation, and quality analysis
• Pandera schema validation
• Data cleaning pipelines
• Pandas vs Polars performance comparison
• Status: ✅ Complete with exercises

• Comprehensive exploratory data analysis
• Feature engineering with domain knowledge
• Data leakage prevention
• Scikit-learn preprocessing pipelines
• Status: ✅ Complete with exercises

• Baseline models and model zoo
• Cross-validation and hyperparameter tuning
• Experiment tracking
• Systematic model selection
• Status: ✅ Complete with exercises

• Metrics beyond accuracy
• Confusion matrices and error analysis
• Confidence calibration
• Model cards and documentation
• Status: ✅ Complete with exercises

• Model serialization and versioning
• Production inference API
• Input validation and error handling
• Model serving patterns
• Status: ✅ Complete with interactive UI

• Production debugging runbook
• Data drift detection techniques
• Incident response procedures
• Monitoring strategies and alerts
• Status: ✅ Complete with 3 incident scenarios

• Git workflows for ML projects
• ML code review best practices
• Writing effective PR descriptions
• Working with existing ML codebases
• Status: ✅ Complete with 3 sample PR reviews

• End-to-end Pokemon card price prediction
• Integrates all concepts from Modules 0-7
• 9 phases: business → data → features → training → evaluation → deployment → monitoring → docs → reflection
• Self-assessment rubric (100 points)
• Status: ✅ Complete with detailed rubric

• ✅ ML Cheatsheet - Quick reference for all key concepts
• ✅ Progress Tracker - Track your learning journey
• ✅ CAPSTONE_RUBRIC.md - 100-point self-assessment rubric
• ✅ 16,000 Pokemon Cards Dataset - Comprehensive dataset with price_usd for regression tasks
• ✅ USING_PRESPLIT_DATA.md - Guide for using pre-split train/val/test datasets

• uv - Fast Python package installer (install here)
• Python 3.13+ (uv will handle this if needed)
• That's it! uvx will handle everything else

# 1. Navigate to the project directory
cd marimo-pokemon

# 2. Install dependencies with uv
uv sync

# 3. Start learning with Workspace Mode (RECOMMENDED!)
uvx marimo edit ./

🌟 Why Workspace Mode (uvx marimo edit ./)?

• Opens the entire project in one window
• File browser to easily switch between modules
• No need to remember file names
• See your progress across all modules
• This is the recommended way to use the course!

If you prefer to open individual notebooks:

# Start with Module 0 (business context)
uvx marimo edit 00_ml_in_business.py

# Or jump to a specific module
uvx marimo edit 03_model_training.py

# View-only mode (read without editing)
uvx marimo run 00_ml_in_business.py

# Generate comprehensive Pokemon card dataset (16,000 cards)
uv run python data/generate_comprehensive_dataset.py

# This creates:
#   - data/pokemon_cards.csv (16,000 cards)
#   - data/pokemon_cards_sample_1000.csv (1,000 card sample)

# Create professional 70/15/15 splits (stratified by type)
uv run python data/create_splits.py

# This creates data/splits/:
#   - pokemon_train.csv (11,200 cards)
#   - pokemon_validation.csv (2,400 cards)
#   - pokemon_test.csv (2,400 cards)

# See USING_PRESPLIT_DATA.md for usage guide

Why use pre-split data?

• ✅ No data leakage (splits done before any EDA)
• ✅ Consistent across all modules
• ✅ Stratified by type (balanced classes)
• ✅ Follows industry best practices (70/15/15)

# Run the setup test to verify everything works
uv run python test_setup.py

# This checks:
# - All dependencies are installed
# - Dataset exists
# - All 8 modules + 7 exercises are present
# - Libraries can be imported
# - Basic operations work

If test_setup.py passes: You're ready to start! 🎉 If it fails: See TROUBLESHOOTING.md for solutions.

💡 Pro Tip: Always use uvx marimo edit ./ from the project root to get the best experience!

Why uvx is awesome:

• No need to install marimo globally
• Automatically uses the right version from your environment
• Works from any directory
• Perfect for ephemeral environments (containers, Codespaces)
• No virtual environment activation needed

Phase 1: Business & Technical Foundations (8-10 hours)

1. Module 0: ML in Business Context → uvx marimo edit 00_ml_in_business.py
2. Exercises 0: Business case studies → uvx marimo edit exercises_00.py
3. Module 1: Data Engineering → uvx marimo edit 01_data_engineering.py
4. Exercises 1: Practice validation and pipelines → uvx marimo edit exercises_01.py
5. Module 2: EDA & Features → uvx marimo edit 02_eda_and_features.py
6. Exercises 2: Feature engineering competition → uvx marimo edit exercises_02.py

Phase 2: Model Development (8-10 hours) 7. Module 3: Model Training → uvx marimo edit 03_model_training.py 8. Exercises 3: CV and tuning → uvx marimo edit exercises_03.py 9. Module 4: Model Evaluation → uvx marimo edit 04_model_evaluation.py 10. Exercises 4: Metrics and error analysis → uvx marimo edit exercises_04.py 11. Module 5: Deployment → uvx marimo edit 05_inference_service.py

Phase 3: Production & Collaboration (6-8 hours) 12. Module 6: Production ML & Monitoring → uvx marimo edit 06_production_monitoring.py 13. Exercises 6: Incident response scenarios → uvx marimo edit exercises_06.py 14. Module 7: Team Collaboration → uvx marimo edit 07_collaboration.py 15. Exercises 7: Code review practice → uvx marimo edit exercises_07.py

Phase 4: Capstone (4-6 hours) 16. Module 8: End-to-End Project → uvx marimo edit 08_capstone.py 17. Reference: Capstone Rubric → CAPSTONE_RUBRIC.md 18. Reference: ML Cheatsheet → ml_cheatsheet.md

Open all notebooks at once with:

uvx marimo edit ./

This gives you a file browser and lets you switch between modules easily!

• ✅ Build end-to-end ML systems from data to deployment
• ✅ Write production-quality ML code with proper validation
• ✅ Engineer features using domain knowledge without data leakage
• ✅ Train and evaluate models systematically
• ✅ Deploy models with monitoring and error handling
• ✅ Debug production ML issues using runbooks
• ✅ Detect and respond to data drift

• ✅ Frame ML problems and calculate ROI
• ✅ Communicate with non-technical stakeholders
• ✅ Set realistic expectations and success metrics
• ✅ Know when NOT to use ML
• ✅ Translate business metrics to model metrics

• ✅ Use Git workflows for ML projects
• ✅ Review ML code effectively
• ✅ Write clear PR descriptions
• ✅ Document models with model cards
• ✅ Onboard to existing ML codebases

You'll be ready to contribute meaningfully to an ML team on day one!

marimo-pokemon/
├── README.md                          # This file
├── CAPSTONE_RUBRIC.md                # 100-point self-assessment rubric
├── prompt.md                          # Complete course specification
├── progress_tracker.md                # Track your learning
├── ml_cheatsheet.md                   # Quick reference guide
│
├── data/
│   ├── generate_dataset.py           # Dataset generator
│   ├── pokemon_cards.csv             # 800 Pokemon cards (with price_usd)
│   └── clean/                        # Cleaned data (generated)
│
├── 00_ml_in_business.py              # Module 0: Business context
├── exercises_00.py                    # Module 0 exercises
├── 01_data_engineering.py            # Module 1: Data engineering
├── exercises_01.py                    # Module 1 exercises
├── 02_eda_and_features.py            # Module 2: EDA & features
├── exercises_02.py                    # Module 2 exercises
├── 03_model_training.py              # Module 3: Model training
├── exercises_03.py                    # Module 3 exercises
├── 04_model_evaluation.py            # Module 4: Evaluation
├── exercises_04.py                    # Module 4 exercises
├── 05_inference_service.py           # Module 5: Deployment
├── 06_production_monitoring.py       # Module 6: Production & monitoring
├── exercises_06.py                    # Module 6 exercises
├── 07_collaboration.py               # Module 7: Team collaboration
├── exercises_07.py                    # Module 7 exercises
├── 08_capstone.py                    # Module 8: End-to-end capstone
│
└── models/                            # Saved models (generated)

• Code written to production standards
• Type hints, docstrings, error handling
• Industry best practices throughout

• 15+ exercises across all modules
• Interactive UIs in Marimo notebooks
• Real-world scenarios and challenges

• Full ML lifecycle from data to deployment
• Both classification and regression examples
• Tools: pandas, polars, scikit-learn, XGBoost

• Industry context in every module
• Real company examples (Netflix, Google, etc.)
• Common pitfalls and how to avoid them

1. Week 1: Modules 0-2 + exercises (8-10 hours)
   • Business context, data engineering, feature engineering
2. Week 2: Modules 3-5 + exercises (8-10 hours)
   • Model training, evaluation, deployment
3. Week 3: Modules 6-7 + exercises (6-8 hours)
   • Production monitoring, team collaboration
4. Week 4: Module 8 Capstone (4-6 hours)
   • End-to-end project applying all skills

• Start with Module 0 - business context is critical!
• Run every code cell and experiment
• Answer socratic questions before moving on
• Complete ALL exercises (they're where real learning happens!)
• Use progress_tracker.md to track completion
• Refer to ml_cheatsheet.md when stuck
• For capstone, use CAPSTONE_RUBRIC.md to self-assess

You've completed the course when you can:

• Frame ML problems with business context and ROI
• Load, validate, and clean data like a pro
• Engineer features without data leakage
• Train multiple models and pick the best systematically
• Evaluate models with appropriate metrics
• Deploy a model to production with monitoring
• Debug production ML systems using runbooks
• Collaborate effectively with ML teams
• Score 80+ on the capstone rubric

Ready for your first ML engineer role!

This course covers the 80-20 most important skills for ML engineers:

• Technical Core (60%): Data → Features → Models → Deployment
• Business Acumen (20%): ROI, stakeholder communication, when NOT to use ML
• Collaboration (20%): Git workflows, code reviews, documentation

Most courses only teach the technical core. This course teaches all three.

• Testing Results: TESTING_RESULTS.md - ✅ Complete test results (all tests passed!)
• Testing Guide: TESTING_GUIDE.md - How to test each module
• Review Summary: REVIEW_SUMMARY.md - Assessment criteria
• Module 1 Evaluation: MODULE_1_EVALUATION.md - Quality metrics

• Python 3.13+
• Marimo - Reactive notebooks (better than Jupyter for production)
• pandas - Data manipulation
• polars - High-performance data processing
• scikit-learn - ML algorithms
• XGBoost - Gradient boosting
• Pandera - Data validation
• Matplotlib/Seaborn - Visualization

Ready to become an ML engineer?

cd marimo-pokemon
uv sync
uvx marimo edit ./

This opens the entire project in one window with a file browser. Start with 00_ml_in_business.py and work through modules 0-8 in order.

Option 1: Start with Module 0

uvx marimo edit 00_ml_in_business.py

Start here to understand the business context before diving into code!

Option 2: Jump to specific modules

uvx marimo edit 03_model_training.py   # Jump to specific module
uvx marimo edit 08_capstone.py         # Jump to capstone (after 0-7)

Option 3: View-only mode

uvx marimo run 00_ml_in_business.py    # View without editing

Happy learning! 🎓

P.S. This course will prepare you for the 80-20 most important skills companies need in ML engineers. Use uvx marimo edit ./ for the best experience, and start with Module 0 - don't skip the business context!