Building Prediction Pipelines that Rock in the Real World
You need to be signed in to add a collection
We are living in the world of (micro)services. Every day we are facing challenges with building scalable and performant applications in relatively short period of time. Cloud became an environment where our services operates. Picking the best technology which is performant, supports development and gives us good forecast for future is crucial. At mobile.de we were challenged with developing pipeline for our prediction models. Similar to “standard” services we want to remain short time-to-market, great performance and prevent mistakes by executing tests. I will talk from experience: - why we prefer H2O over python and R - how we have built pipeline using H20, scala and akka-http to predict in real time on milliseconds speed - what we do to optimize prediction model This talk is for software (data) engineers interested in building fast scalable pipelines.
Transcript
We are living in the world of (micro)services. Every day we are facing challenges with building scalable and performant applications in relatively short period of time. Cloud became an environment where our services operates. Picking the best technology which is performant, supports development and gives us good forecast for future is crucial.
At mobile.de we were challenged with developing pipeline for our prediction models. Similar to “standard” services we want to remain short time-to-market, great performance and prevent mistakes by executing tests.
I will talk from experience:
- why we prefer H2O over python and R
- how we have built pipeline using H20, scala and akka-http to predict in real time on milliseconds speed
- what we do to optimize prediction model
This talk is for software (data) engineers interested in building fast scalable pipelines.