https://predictiveanalyticsworld.de/de/berlin2017/programm/#session37621 AutoScout24 is the largest online car marketplace Europe-wide. With more than 2.4 million listings across Europe, AutoScout24 has access to large amounts of data about historic and current market prices and wants to use this data to empower its users to make informed decisions about selling and buying cars. We created a live price estimation service for used vehicles based on a Random Forest prediction model that is continuously delivered to the end user. Learn how automated verification using live test data sets in our delivery pipeline allows us to release model improvements with confidence at any time.

1. Predictive Analytics World Berlin | 14.11.2017 | Christian Deger | @cdeger
Predictive Analytics for Vehicle Price Prediction
Delivered Continuously at AutoScout24

2. Christian Deger
Chief Architect
christian@deger.eu
@cdeger

3. PL
S
RUS
UA
RO
CZ
D
NL
B
F
A
HR
I
E
BG
TR
18countries
2.4m+cars & motos
10m+users per
month

4. The task: A consumer-facing data product

5. The task: A consumer-facing data product

6. The task: A consumer-facing data product

7. The prediction model
Car listings of
last two years

8. The prediction model
Car listings of
last two years

9. The prediction model: Random forest
Volkswagen GolfCar listings of
last two years

10. How to turn an R-based prediction model
into a high-performance web application?
?

11. How to turn an R-based prediction model
into a high-performance web application?

12. How to turn an R-based prediction model
into a high-performance web application?

13. How to turn an R-based prediction model
into a high-performance web application?
 Continuous Delivery!

14. Application code in
one repository per
service.
Typical delivery pipeline

15. Application code in
one repository per
service.
CI
Deployment package
as artifact.
Typical delivery pipeline

16. Application code in
one repository per
service.
CI
Deployment package
as artifact.
CD
Deliver package to
servers
Typical delivery pipeline

17. Continuous delivery pipelines
Prediction Model Pipeline

18. Continuous delivery pipelines
Prediction Model Pipeline
Web Application Pipeline

19. The price for CD: Extensive model validation

20. The price for CD: Extensive model validation

21. Lessons learned
Form a cross-functional team of
data scientists & software engineers!
Software engineers
… learn how data scientists work
… and understand the quirks of a prediction model
Data Scientist
… learn about unit testing, stable interfaces, git, etc.
... get quick feedback about the impact of their work
 Model and product iterations become much faster!

22. Lessons learned
Generating gigabytes of Java code
is a challenge for the JVM
Use the G1 garbage collector
Turn off Tiered Compilation
 Do extensive warm-ups

23. Lessons learned – Warm up

24. Lessons learned
The approach of applying Continuous Delivery to
Data Science is useful independently of the tech
 Successfully applied similarly to a Python- and
Spark-based project
 Even more useful when quick model evolution
is required because of rapidly changing inputs
(e.g. user interaction)

25. Conclusions
 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.

26. Conclusions: Price evaluation everywhere

27. Conclusions: Price evaluation everywhere

28. Conclusions: Price evaluation everywhere

29. Conclusions: Price evaluation everywhere