Slide 1
Slide 1 text
Holiday Extras Case Study: Employing Google Cloud
Machine Learning Engine to develop models in
production
REBECCA VICKERY | DATA SCIENTIST
@vickdata
Slide 2
Slide 2 text
Rebecca Vickery
Data Scientist, Holiday
Extras
@vickdata
medium.com/rebeccavickery
Slide 3
Slide 3 text
Scaling machine learning with
Google Cloud
Slide 4
Slide 4 text
A data science journey
@vickdata
Slide 5
Slide 5 text
No content
Slide 6
Slide 6 text
No content
Slide 7
Slide 7 text
No content
Slide 8
Slide 8 text
COLLECT,
MOVE, STORE
Slide 9
Slide 9 text
COLLECT,
MOVE, STORE
TRANSFORM,
AGGREGATE,
LABEL
Slide 10
Slide 10 text
TEST, LEARN,
PREDICT
TRANSFORM,
AGGREGATE,
LABEL
COLLECT,
MOVE, STORE
Slide 11
Slide 11 text
Data science team purpose:
To optimise customer experience and
business performance using
automated decision making leading to
greater profitability
Slide 12
Slide 12 text
NOW
Simple Machine
Learning
FUTURE
AI, Deep
Learning
Slide 13
Slide 13 text
Deploying machine learning is hard
@vickdata
Slide 14
Slide 14 text
Tools - Data Science
Open source Lack software
development
expertise
Mainly python
c Flaticon
Slide 15
Slide 15 text
Tools - Software Engineers
Different tools
Lack ML/Data
expertise
Mainly javascript
c Flaticon
Slide 16
Slide 16 text
Data science process
The wrong
kind of
independence
c Flaticon
Slide 17
Slide 17 text
People
Small data science team
Science + software experts are rare
c Flaticon
Slide 18
Slide 18 text
Two types of deployment
Slide 19
Slide 19 text
Bespoke Solutions
“Ideas are
worth nothing
unless
executed”,
Derek Sivers
c Daniel Moyo
Slide 20
Slide 20 text
Unused Models
Many models never make it to
production
“Ideas are worth nothing unless
executed”,
Derek Sivers
Slide 21
Slide 21 text
Time to model deployment
Model development
= days to weeks
Model deployment =
weeks to never!
“Ideas are worth
nothing unless
executed”,
Derek Sivers
Slide 22
Slide 22 text
The Google Way
@vickdata
Slide 23
Slide 23 text
Machine Learning Engine
A technical solution but also a
strategic solution
c Google
Slide 24
Slide 24 text
The right kind of independence
c flaticon
Data Scientists can
use preferred tools
Slide 25
Slide 25 text
The right kind of independence
c flaticon
Repeatable process
for deployment of most
models
Slide 26
Slide 26 text
Faster time to production
c flaticon
Fully managed service
Slide 27
Slide 27 text
Faster time to production
c flaticon
Fast response from
Google Cloud
Support
Slide 28
Slide 28 text
Extensibility
c flaticon.com
Multiple models and
versions
Slide 29
Slide 29 text
Extensibility
c flaticon.com
Regularly release
support for newer
versions of tools
Slide 30
Slide 30 text
c Flaticon
__init__.py
task.py
setup.py
model.py
Model Package
Slide 31
Slide 31 text
Repeatable, reusable process
__init__.py
task.py
setup.py
model.py
Model Package
Slide 32
Slide 32 text
No content
Slide 33
Slide 33 text
No content
Slide 34
Slide 34 text
Collaborative Project
Slide 35
Slide 35 text
ML Proxy (bespoke ML microservice)
Slide 36
Slide 36 text
Monitoring - model performance
Slide 37
Slide 37 text
Monitoring - ML Engine performance
Slide 38
Slide 38 text
No content
Slide 39
Slide 39 text
Three key benefits
@vickdata
Slide 40
Slide 40 text
The right kind
of
independence
Slide 41
Slide 41 text
Collaboration
for data
scientists
Slide 42
Slide 42 text
Data scientists
deliver more
value
Slide 43
Slide 43 text
Thank you
@vickdata