Upload
christian-deger
View
192
Download
0
Embed Size (px)
Citation preview
PredictiveAnalyticsWorldLondon|Oct12,2016|ArifWider&ChristianDeger
PredictiveAnalyticsforVehiclePricePredictionDeliveredContinuouslyatAutoScout24
Dr.ArifWiderSeniorConsultant/[email protected]@arifwider
[email protected]@cdeger
The task: A consumer-facing data product
5Predictive Analytics World London 2016 Predictive Analytics Delivered Continuously – A. Wider & C. Deger
The task: A consumer-facing data product
6Predictive Analytics World London 2016 Predictive Analytics Delivered Continuously – A. Wider & C. Deger
The task: A consumer-facing data product
7Predictive Analytics World London 2016 Predictive Analytics Delivered Continuously – A. Wider & C. Deger
The prediction model: Random forest
8
Volkswagen GolfCar listings oflast two years
Predictive Analytics World London 2016 Predictive Analytics Delivered Continuously – A. Wider & C. Deger
How to turn an R-based prediction model into a high-performance web application?
9
?
Predictive Analytics World London 2016 Predictive Analytics Delivered Continuously – A. Wider & C. Deger
How to turn an R-based prediction model into a high-performance web application?
10Predictive Analytics World London 2016 Predictive Analytics Delivered Continuously – A. Wider & C. Deger
How to turn an R-based prediction model into a high-performance web application?
11Predictive Analytics World London 2016 Predictive Analytics Delivered Continuously – A. Wider & C. Deger
How to turn an R-based prediction model into a high-performance web application?
12
à Continuous Delivery!
Predictive Analytics World London 2016 Predictive Analytics Delivered Continuously – A. Wider & C. Deger
Applicationcode inonerepositoryper
service.
CI
Deploymentpackageasartifact.
CD
Deliverpackagetoservers
Predictive Analytics World London 2016 Predictive Analytics Delivered Continuously – A. Wider & C. Deger
Typical delivery pipeline
Continuous delivery pipelines
14
Prediction Model Pipeline
Predictive Analytics World London 2016 Predictive Analytics Delivered Continuously – A. Wider & C. Deger
Continuous delivery pipelines
15
Prediction Model Pipeline
Web Application Pipeline
Predictive Analytics World London 2016 Predictive Analytics Delivered Continuously – A. Wider & C. Deger
The price for CD: Extensive model validation
16Predictive Analytics World London 2016 Predictive Analytics Delivered Continuously – A. Wider & C. Deger
The price for CD: Extensive model validation
17Predictive Analytics World London 2016 Predictive Analytics Delivered Continuously – A. Wider & C. Deger
Lessons learned
18
§ Form cross-functional teams of data scientists & software engineers!
§ Set up usage reporting early to improve your data product in a data-driven way.
§Generating gigabytes of Java code is a challenge for the JVMà Use the G1 garbage collector
à Do extensive warm-ups
àTurn off Tiered Compilation
Predictive Analytics World London 2016 Predictive Analytics Delivered Continuously – A. Wider & C. Deger
Conclusions
19
§Continuous Delivery allows us to bring prediction model changes live very quickly.
§Only extensive automated end-to-end tests provide confidence to deploy to production automatically.
§ Java code generation allows for very low response times and excellent scalability for high loads but requires plenty of memory.
Predictive Analytics World London 2016 Predictive Analytics Delivered Continuously – A. Wider & C. Deger