By Jorge Torres Adam Carrigan Costa Tin Erik Bovee George Hosu
Classical machine learning workflows are complex. It takes weeks to months to validate, test, and deploy ML models and integrate them into an application, not to mention how expensive it is. But what if machine learning was a standard functionality of data-layer tools, like Redis? Furthermore, what if ML models could learn and predict/forecast from the data contained in Redis automatically?
MindsDB achieved significant progress in bringing machine learning into popular databases, and now we are doing it for Redis! MindsDB’s open-source integrations make it possible to implement ML projects in a matter of hours, and only require basic database skills.
In this journey, we discovered and solved some machine learning problems that are hard even for ML engineers but are common for Redis, such as forecasting over time-series data streams while taking into account multiple parameters (model features) at once. For example, real-time forecasting of anomalies in inventory for all products in all stores from a stream containing inventory updates over time.
We have made incredible progress in solving these time-series problems, and now we bring these capabilities to Redis Streams, We would like to share with you our solutions and discuss some exciting ideas that have occurred in the process.