hydra-experiment-config

What problem does it solve?

Centralizes and clarifies Hydra-based experiment configuration so runtime constants are not hardcoded, overrides compose predictably, and experiments are reproducible across runs and environments.

Core Features & Use Cases

• Config-as-source-of-truth: Ensure all runtime constants live in YAML and avoid duplicated module-level constants.
• Hierarchical groups with flat aliases: Organize by semantic groups while providing flat aliases for ergonomic CLI overrides.
• Defaults list, derived values, and CLI patterns: Provide patterns for defaults, computed configuration values, multirun sweeps, and safe override behavior.
• Use Cases: Bootstrapping a new project's conf layout, adding model/training/data variants, debugging override resolution, and preparing configs for experiment tracking.

Quick Start

Convert this project's hardcoded training hyperparameters into a Hydra conf/ layout with hierarchical groups, flat aliases, and a defaults list so overrides resolve predictably.