✨ mlflow UI on Analytical Platform

[mshodge]

Describe the feature request.

Hi there -

Data Scientists at MoJ routinely do Machine Learning work on the AP, but there is no centralised, structured way of them logging and recording their experiments and ML model performance. mlflow is a widely used tool by Data Scientists to record and assess performance of their ML models.

This request would be to help setup the backend storing of the model artefacts as well as helping setup the UI for AP users to look at their own, and others, ML performance and metrics.

Describe the context.

I am starting my role as MLOps coordinator from next week. Therefore I am highlighting MLOps needs as early as possible in the hopes that they can be made possible during my six month stint.

Value / Purpose

Without a dedicated tool like mlflow, the impact of Data Science at MoJ will be greatly reduced. It will save time for the Data Scientists by providing a centralised store of their own model performance and metrics, but also provide them with the opportunity to see what others are doing. Without this approach, ML model metrics will forever be kept on scrap bits of paper and likely, lost

User Types

Data Scientists doing ML

[mshodge]

Note: Might be redundant if SageMaker is the chosen approach.

[Ed-Bajo]

Meet to ascertain actual require and current state.

[michaeljcollinsuk]

Related support issue https://github.com/ministryofjustice/data-platform-support/issues/527

[mshodge]

Thanks @michaeljcollinsuk and @Ed-Bajo - bringing in @PriyaBasker23 to this as well. We've done a lot of work on this over the past few weeks on the sandbox and before we meet we will have a better ask for you and the team :)

[jacobwoffenden]

Closing as duplicate of #4275