When it comes to Machine Learning (ML) development there are many tools created to help assist developers. Model registries are in ML development. Although they seem similar to version control they have different use cases.

What is a Model Registry

Yunna Wei defines a model registry as a centralised place to store all your ML artefacts along with their metadata from early-stage experiments to production-ready models. This allows ML teams to collaborate on models by providing model organisation, discovery, versioning, the ability to trace the origin of the model and the ability to manage production statuses of your models.

Importance of Model registry

Dominick Rocco states that without a model registry, data scientists and machine learning engineers are more likely to cut corners or make costly mistakes such as:

Mislabeled model artefacts and difficulties tracking files from training jobs.Lost or deleted dataMissing source code or unknown versions as even good models sometimes have errors.Undocumented model performance which makes it hard to compare different versions

On top of avoiding the negatives of not having one there are many positives to having a model registry. Stephen Oladele further elaborates on this by listing the positives of having a model registry:

Having a model registry enables faster deployment of your models by bridging the gap between experiment and production activities resulting in a faster production model rollout. It also stores trained models to be quickly retrieved.Having a model registry simplifies model lifecycle management by simplifying the management of your model lifecycleHaving a model registry enables production model governance by centralising models and organising their relevant detailsHaving a model registry can help improve the security of your model by managing specific versions of the packages that you can scan and remove security vulnerabilities that may pose a threat to the system.

In summary, version control software is used for managing changes to the codebase and other files in a machine learning project, while a model registry is used to manage and deploy machine learning models. We at Eden AI are more happy to set up model registries for production workload. Contact us @helloworld@edenai.co.za and we will help you.

​Stories by Eden AI on Medium  

Read More