How Adatis helped Rank Gaming productionise & automate the management of machine learning models in Azure
Rank Group approached Adatis Consulting Ltd in 2017 to help tackle a key issue their data science team were encountering – “How do you to gracefully transition from one machine learning model to another as models are retrained and rewritten?”
Rank Group is the owner of many popular gaming brands in the UK, including Grosvenor Casinos and Mecca Bingo. Rank use Machine Learning to optimise and influence business decisions across their enterprise. Models are deployed to identify customers churn, improve cross sale, enhance retention and most importantly to identify customers who are at risk of having a gambling addiction. These models are constantly being evaluated and retrained as gambling habits change and new games are introduced.
Adatis implemented a new advanced Machine Learning Model Management service based on the "Rendezvous" Architecture created by Ellen Friedman and Ted Dunning. Rendezvous handles the distribution of a single request to multiple models, scoring all in parallel, then decides on the most appropriate output to return. This is a massively scalable, flexible architecture that solves one of the key problems encountered by Data Science teams today.
In this session we will look at the original problem and the architecture which was used to solve it.