machinable

machinable is a modular configuration system for machine learning research. Using straight-forward conventions and a powerful configuration engine, it can help structuring your projects in a principled way to move quickly while enabling reuse and collaboration.

Explore key features at a glance →

Ready to start?

$ pip install machinable

Features

Powerful configuration

• YAML-based project-wide configuration files with expressive syntax
• Efficient configuration manipulation
• Modular code organisation to allow for encapsulation and re-use
• Import system to use 3rd party configuration and code without overhead
• 'Mixins' for horizontal inheritance structure

Efficient execution

• Works with existing code
• Support for seamless cloud execution
• Automatic code backups
• Managed randomness and reproducibility
• Advanced hyperparameter tuning using Ray Tune

Effective result collection and analysis

• Logging, tabular record writer and storage API
• File system abstraction (in-memory, AWS S3, and more)
• Flat-file result database with SQL-like query interface
• Convenient configuration and result retrieval

Documentation

Read the user guide to get started.