Data Science - Delivered Continuously - XConf 2017

Arif Wider & Christian Deger

DATA SCIENCE,DELIVERED CONTINUOUSLY

A CONFERENCE ALL ABOUT TECHNOLOGY

Christian DegerChief [email protected]@cdeger

Arif WiderSenior Consultant/[email protected]@arifwider

PL

S

RUS

UA

RO

CZ

D

NL

B

FA

HRI

E

BG

TR

18countries

2.4m+cars & motos

10m+users per

month

The task: A consumer-facing data product

6XConf ’17 Hamburg/Manchester Data Science, Delivered Continuously – A. Wider & C. Deger





The prediction model: Random forest

9

Volkswagen GolfCar listings oflast two years

XConf ’17 Hamburg/Manchester Data Science, Delivered Continuously – A. Wider & C. Deger

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

10

?







13

Continuous Delivery!


Application code in one repository per

service.

CI

Deployment packageas artifact.

CD

Deliver package to servers

Typical delivery pipeline


Continuous delivery pipelines

15

Prediction Model Pipeline


Continuous delivery pipelines

16

Prediction Model Pipeline

Web Application Pipeline


The price for CD: Extensive model validation


The price for CD: Extensive model validation


Lessons learned

19

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!


Lessons learned

20

Generating gigabytes of Java code

is a challenge for the JVM

Use the G1 garbage collector

Turn off Tiered Compilation

Do extensive warm-ups


Lessons learned – Warm up


Lessons learned

22

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)


Conclusions

23

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.


Conclusions: Price evaluation everywhere


Conclusions: Price evaluation everywhere


QUESTIONS?

THANK YOU

Arif Wider & Christian Deger

@arifwider @cdeger

Data & Analytics

Data Science - Delivered Continuously - XConf 2017