How does Hydra help ML config management?

medium

Answer

$Facebook's config framework.$
$Compose configs from YAML files: defaults list + overrides.$
$\mathrm{python}\;\mathrm{train}\mathrm{py}\;\mathrm{optimizer} = \mathrm{adam}\;\mathrm{lr} = 0.001$
$\mathrm{python}\;\mathrm{train}\mathrm{py}\; - m\;\mathrm{lr} = 0.001, 0.01, 0.1$
$\mathrm{model} = \mathrm{resnet50}\;\mathrm{vs}\;\mathrm{vit}$
$Structured configs with dataclasses for type safety.$
$Automatic run dir per experiment.$
$Combined with Optuna sweeper for HPO.$
$Standard for research + production.$

Check yourself — multiple choice

Random
Facebook config framework: compose YAML + CLI overrides + multi-run sweep + groups swap subsystems (model=resnet50 vs vit) + structured dataclass + auto run dir + Optuna sweeper HPO
Same as argparse
Not real

Hydra: compose + override + multi-run + groups + structured.

#reproducibility#pipeline

215 interview questions in this topic.