Agile Data Science

1© All rights reserved to

Agile Analytics in Mobile Gaming:lessons learned

Volodymyr (Vlad) KazantsevHead of Data Science at Product Madness

2015

2

What we do?

3

Heart of Vegas in charts

iPad rankings, US iPad rankings, Australia

4

Data Impact Team

● Ad-hoc analytics and daily fires

● Deep dive analysis;Predictive analytics

● Data Engineering; R&D

Team of 6

5

Few Examples

A B

A/B TestsCustomer Lifetime Value

days

$ va

lue

Segmentation

group 1 group 2 group 3 group 4

6

Technology Stack

C++ETL

orchestration

Transformation& Aggregation

SQL

Data Products

Reports

Dashboards

+

7

Lessons

8

Lesson 1: Agile Philosophy for Data Science

1

9

Agile Manifesto

Individuals and interactions over processes and tools

Working software over comprehensive documentation

Customer collaboration over contract negotiation

Responding to change over following a plan

* agilemanifesto.org

10

Agile Data Science Manifesto

Individuals and interactions over processes and tools

Actionable insights over comprehensive reports

Customer collaboration over project negotiation

Responding to change over following a plan

11

“If a building doesn’t encourage [collaboration], you’ll lose a lot of innovation and the magic that’s sparked by serendipity” - Steve Jobs

Individuals and interactions over processes and tools

12

Individuals and interactions over processes and tools

Standing Desks + Easily Available Whiteboard

13

Agile Principles

Iterative, incremental and evolutionary

Efficient and face-to-face communication

Very short feedback loop and adaptation cycle

Quality focus

- iterations, timeboxed estimates

- no to tasks by email (with no face-to-face)

- daily standups, pair analysis

- verifiable, reproducible findings

14

Data Science Board

15

Scrum-Ban in Data Science @ProductMadness

● Weekly cycle

● Daily standup meeting @10am

● ToDo/WIP/Waiting buckets are kept small

● Disruptions to weekly plan are expected

● On-demand planning

16

Lesson 1: Agile methods in Data Science

1. co-location matter; whiteboard next to your desk

2. Work with decision maker; share preliminary findings

3. Make a research plan; pivot early

4. Book “Findings” meeting before project start

5. MVP for Data Products

6. Do Daily Stand-ups !

17

Lesson 2: Agile Velocity vs. Acceleration

2

18

What is Agile Acceleration

Waterfall Scrum

Units of WorkTime IntervalVelocity = ΔVelocity = Acceleration* ΔTime

VS.

19

a = Fm

I run SQL, copy-paste data to Excel and send it by email

I created a deep neural network to predict high spenders

20

Case Study: to Git or not to Git

Scripts (ruby, bash, python)Python AppsPython ModulesIPython NotebooksResearch Documents (word)Presentations (powerpoint)Spreadsheets (excel)

21

Case Study: Git or not to Git

Scripts (ruby, bash, python)Python AppsPython ModulesIPython Notebooks ?Research Documents (word)Slides (powerpoint)Spreadsheets (excel)

22

Case Study: Git or not to Git

Scripts (ruby, bash, python)Python AppsPython ModulesIPython NotebooksResearch Documents (word)Slides (powerpoint)Spreadsheets (excel)

23

Remove unnecessary weight

24

Lesson 2: find the lightest suitable tool1. IPython notebooks: Dropbox over Git2. Google Slides over Powerpoint

Google Slides over Email with images (>2 images)

3. Google Spreadsheets over Excel (for analytics)4. Podio over Jira (for analytics)5. Data Transformations in DWH in SQL over Hadoop6. Don’t copy-paste code in IPython notebooks; use functions;

don’t copy-paste functions in notebooks, use modules

25

Lesson 3: Focus on Closing the Loop

3

26

Analytics Loop

Spot Opportunity

Ask the Right Question

Make Decision

Improve the Business

Data Science @work

27

Analytics Spiral

Ideas & Questions

Data Analysis

Insights

Impact

28

Data Science Value Pyramid

Store & Query

Reports

Descriptive Analytics

Predictive Analytics

Data Products

* inspired by Agile Data Science, Russell Jurney, O'Reilly Media 2013

Record what Happened

Was it good or bad?

Why did it happen?

What will happen?

Affect the outcome

com

plex

ity

valu

e

29

Data Science Value Loop

Record what Happened

Was it good or bad?

Why did it happen?

What will happen?

Affect the outcome

30

Limit the number of Open Loops

90% 90%

75%80%

80%60%

100% 100%

100%100%

0% 0%

Always prefer to have: 90% of tasks are 100% complete

over 100% of tasks are 90% complete

VS.

31

Lesson 3: Focus on Closing the Loop1. Don’t build predictive models that you can’t act upon. Don’t

analyse stuff that does not help to make a decision

2. The best way to deal with Analytics Spiral is to avoid the spiral. Practise Crack a Case and “what if” method.

3. Climb the Data Value Pyramid fast.Once climbed - optimise the Data Value Loop.

4. Limit the number of “open loops”

32

Lesson 4: Reproducibility Matters

4

33

To the and back!

34

Why?

Boss: “Great! Can you run this for all monthly cohorts?”Because:

35

Why?

Because:Boss: “Sam is on holiday.Can you re-run his analysis?”

36

Few IPython Tips

37

Import all commonly used toolsin one line.

All access and security is abstracted away.Focus on SQL, not data access

formatting and publishing a .png in one line of code

PyCharm has great SQL editor

38

Lesson 4: Reproducibility

● Get rid of Windows and you get rid of Excel

● ipynb are always shared and versioned;Prefer simple cloud sharing to VCS

● Streamline data access functions

● Cache long-running code and queries

● Develop a common library

39

In Summary...

40

Summary

● Agile approach works well in Data Science

● Find the lightest suitable tool for a task

● Reproducibility is not negotiable

● Focus on closing the loop(s)

41

Questions?

We are Hiring !

volodymyrk

[email protected]