Note: This project is currently a work in progress.
This is a comprehensive rework of my original NBA prediction machine learning project, focusing on improved reliability, engineering practices, and prediction accuracy.
- Migration from SaaS solutions to open-source alternatives
- Using GitHub as primary data store (previously Hopsworks)
- Implementing MLFlow for experiment tracking (replacing Neptune)
- MLFlow for model registry management (replacing Hopsworks)
- Redundant deployment architecture
- Container-based deployment strategy
- Object-Oriented Programming implementation
- Enhanced modularity and component isolation
- Comprehensive logging system
- Robust error handling
- Extensive testing infrastructure
- Expanded data collection through advanced scraping
- Integration of ELO scoring system
- Advanced feature selection methodology
- Increased experimental iterations
- Target accuracy improvement to ~65%
- Detailed technical documentation
- Process discussion and methodology explanations
- Decision-making rationale
- Multi-provider deployment architecture
- Self-hosted solutions
- Cloud platform integration
- Redundant deployment systems
The original version of this project can be found [here][https://github.com/cmunch1/nba-prediction].