NBA Pick'em

NBA game prediction system using machine learning to predict game winners.

Architecture

┌─────────────────────────────────────────────────────────────┐
│  EXTRACTION PIPELINE                                   │
│  Scripts or Scrapy collect:                           │
│  - Team data                                      │
│  - Game results                                  │
│  - Matchup schedule                               │
│  - Betting odds                                  │
└─────────────────────────────────────────────────────────────┘
                            ↓
┌─────────────────────────────────────────────────────────────┐
│  STORAGE                                              │
│  DuckDB (data/nba_pickem.duckdb)                       │
│  Tables: teams, games, game_predictions, moneyline_odds  │
└─────────────────────────────────────────────────────────────┘
                            ↓
┌─────────────────────────────────────────────────────────────┐
│  MODELING PIPELINE                                     │
│  Jupyter notebooks for:                               │
│  - Data exploration                                 │
│  - Feature engineering                             │
│  - Model training                                 │
│  - Evaluation                                     │
└─────────────────────────────────────────────────────────────┘
                            ↓
┌─────────────────────────────────────────────────────────────┐
│  PREDICTION PIPELINE                                   │
│  Run predictions for any date:                      │
│  nba-pickem-predict 2025-04-20                    │
└─────────────────────────────────────────────────────────────┘

Quick Start

# Install package in editable mode
make install

# Initialize database
make init

# Or with seed data from CSV
make init-load

# Generate predictions
make predict

CLI Commands

Command	Description
nba-pickem	Show version and available commands
nba-pickem-init	Initialize empty DuckDB
nba-pickem-init --load	Initialize with seed data from CSV
nba-pickem-extract	Extract data (placeholder)
nba-pickem-predict	Generate predictions for today
nba-pickem-predict 2025-04-20	Predict specific date
nba-pickem-train	Train prediction model
python -m nba_pickem	Run as module

Or use make targets:

make install    # Install dependencies
make init      # Initialize empty DB
make predict   # Generate predictions
make jupyter  # Start Jupyter server
make clean    # Remove DB and cache files

Project Structure

nba-pickem/
├── src/nba_pickem/         # Python package
│   ├── __init__.py       # Exports
│   ├── __main__.py       # python -m entry
│   ├── config.py       # Paths (PACKAGE_ROOT, DATA_DIR, DB_PATH)
│   ├── dataloader.py   # DuckDB data layer
│   └── scripts/       # CLI scripts
│       ├── setup.py
│       ├── run_extraction.py
│       ├── run_prediction.py
│       └── train_model.py
├── scraping/           # Scrapy spiders (separate project)
├── data/             # Data files (duckdb, models, CSVs)
│   └── raw/         # Raw seed data
├── worker/           # Trained model
│   └── nba_model.pkl
├── notebooks/        # Model training pipeline
└── Makefile

Requirements

• Python 3.11 - 3.13 (not 3.14 yet due to PyO3 compatibility)
• uv for package management
• scikit-learn for ML
• DuckDB for storage
• Scrapy for data collection (optional)

Cron Usage

# Daily prediction at 8am
0 8 * * * cd /path/to/nba-pickem && nba-pickem-extract && nba-pickem-predict

License

MIT