What is a model registry?
Options
- A list of the hyperparameters of one model
- A log of all prediction requests
- A cloud bucket for raw datasets
- A catalogue of model versions with their metadata (metrics, data and code used, stage), so you know what is approved and what is deployed (Correct option)
Answer
A catalogue of model versions with their metadata (metrics, data and code used, stage), so you know what is approved and what is deployed
Why
The registry is the shared record of the models a team has built. Each entry points to the model files and says which run created it, how it scored and whether it is in staging or production.
It lets deployment tools fetch "the approved version" instead of a file someone copied by hand.
It usually supports stages or aliases and an audit trail of who promoted what.
Remember
Registry = versions + metadata + stage.