Distributed deployment example

[victusfate]

I wanted to share something I built on top of river for a continuous learning system.

It leverages item ratings in realtime through:

• redis pubsub
• updates a recent ratings ordered set
• does a full training hourly via an exported csv
• delta trains from the recent ratings after the models last event

I don't see any major issues unless the delta training set grows large vs the memory cost (many millions fit into a reasonable set of memory).

https://github.com/victusfate/concierge

[MaxHalford]

I love the smell of real-time in the morning.

That looks really fun, I love it! We definitely need more user examples like these. Thank you very much for sharing, I hope it will inspire others :)