machinable 2.8.1

A modular configuration system for machine learning research

machinable

machinable provides a system to manage configuration of machine learning projects more effectively. Using straight-forward conventions and a powerful configuration engine, it can help structure your projects in a principled way so you can move quickly while enabling reuse and collaboration.

Read the user guide to get started.

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 convenient query syntax