Building low latency gradient boosted tree predictors on distributed platform- György Móra |Crunch19 скачать в хорошем качестве

Building low latency gradient boosted tree predictors on distributed platform- György Móra |Crunch19

👉 Learn more about the talk and download the slides at https://crunchconf.com/speaker/Gyorgy... 📬 Sign up to our newsletter so you won't miss the updates about the next Crunch Data Conference: http://eepurl.com/dGwi1f At Ekata we evaluate identity information in online shopping transactions, to enable our customers (vendors) to protect against fraud. We make commitments to deliver pre-purchase verification predictions with very low latency. When we built the ML system powering the Identity Check Confidence Score we faced the following challenges: The pipeline that trains our model had to digest a large amount of transaction data We had to perform a large set of validation experiments to ensure each new release caused no disruptions for our customers At the same time, the model had to perform in a real time, low latency environment In this talk I discuss how we met these challenges by building a Spark- and XGBoost-based training and experimentation pipeline that delivers our models as a custom predictor library, to allow seamless integration with the production environment. Our embeddable library enables super-fast predictions real-time. - Captured Live on Ustream at https://www.ustream.tv/channel/JUMjvC...