Machine Learning and Sagemaker at Zalando Mark… · Machine Learning and Sagemaker at Zalando...

Post on 26-May-2020

44 views 0 download

Transcript of Machine Learning and Sagemaker at Zalando Mark… · Machine Learning and Sagemaker at Zalando...

Kshitij Kumar,VP Data InfrastructureZalando SEKshitij.Kumar@zalando.de

Machine Learning and Sagemaker at Zalando

Suhas GuruprasadML Engineering LeadZalando SEsuhas.guruprasad@zalando.de

2

Put images in the grey dotted box "unsupported placeholder" - behind the orange box (left side stays white)

Write the quote in all capital letters

WE LOVE FASHION

3

Put images in the grey dotted box "unsupported placeholder" - behind the orange box (left side stays white)

Write the quote in all capital letters

WHAT STARTED AS A SIMPLE ONLINE SHOP…

4

Put images in the grey dotted box "unsupported placeholder" - behind the orange box (left side stays white)

Write the quote in all capital letters

…HAS BECOME THE EUROPEAN ONLINE PLATFORMFOR FASHION

5

Please write the title in all capital letters

Put images in the grey dotted box "unsupported placeholder"

WE OFFER A SUCCESSFUL AND CURATED ASSORTMENT

> 300,000articles from

~ 2,000international brands

15 privatelabels

HIGHLYEXPERIENCEDcategory management

> 500designers& stylistsLOCALIZATION

of the assortment

CURATEDSHOPPING

with Zalon

6

Please write the title in all capital letters

Put images in the grey dotted box "unsupported placeholder"

PLATFORM STRATEGY

BRANDS CONSUMERS

ENABLER

7

Put images in the grey dotted box "unsupported placeholder" - behind the orange box (left side stays white)

Write the quote in all capital letters

WE DRESS CODE

8

Please write the title in all capital letters

Put images in the grey dotted box "unsupported placeholder"

WE ARE CONSTANTLY INNOVATING

CLOUD-BASED,CUTTING-EDGE& SCALABLEtechnology solutions

> 2,000employees at

international tech locations8

HQsin Berlin

help our brand toWIN ONLINE

9

Please write the title in all capital letters

Put images in the grey dotted box "unsupported placeholder"Possible use cases of ML at an online retailer

An ML Driven Customer

Experience

10

ML driven real-time

recommendation engine

People who browsed this style also browsed these other styles…

11

Complete the look

Multi-dimensional ML driven product placement

Search

Recommended products

Complimentary items

Size (fit)

Delivery promise

12

ML Driven Supply Chain Management

What? ❏ Do we need to provide?

When? ❏ Do we need to provide it?

Where? ❏ Should it be available?

How much?

❏ Should it be available?

We use a myriad of tools

Nakadi

The ML JourneyDigital Foundation - Data

Explore

Fetch

Prepare

Train Model

Evaluate Model

Deploy to production

Monitor/ Evaluate Ready the Data

Prepare the models

Serve the models

16

Please write the title in all capital letters

Achieving the balance to run ML at Scale

Exploding new With the needs

Speed of Experimentation

Safe environment with metadata

Cost Efficiency

Number of User teams

Use cases

17

Please write the title in all capital letters

The ML pipeline – for a single use case

ML Use CaseNotebook/UI

creates workflows

Fetch Data

Extract Features

Prepare Data

Train Model Deploy Model

Serve

Monitor

Evaluate and Feedback

18

Please write the title in all capital letters

The ML pipeline – a couple of use cases

19

Please write the title in all capital letters

The ML pipeline – many use cases

20

Please write the title in all capital letters

Why SageMaker at Zalando

ML at scale, with cost efficiency

The ability to run hundreds of training jobs that are “serverless”. Trainings produce models and infrastructure is automatically shutdown.

3

ML at scale, with safety

The ability to understand metadata at every stage of the ML journey by just describing a training job at the call of

an API.

2

ML at scale, with speed

The ability to compose training jobs, tuning jobs and endpoints with ease, at the call of an API, and with algorithms available out of the box.

1

ProductionizingMLPipelinesAt Zalando

(Speed, Safety, Cost Efficiency)

22

Please write the title in all capital letters

Put images in the grey dotted box "unsupported placeholder"

An end to end pipeline in action

23

Please write the title in all capital letters

Put images in the grey dotted box "unsupported placeholder"

An end to end pipeline in action

24

Please write the title in all capital letters

Put images in the grey dotted box "unsupported placeholder"

Productionizing ML: Speed, with simplicity

src/lambdas/training_job.pysrc/lambdas/endpoint.pycf.yamlci-cd.yaml

CF:1. Step functions definition2. Trigger rule3. Roles

25

Please write the title in all capital letters

Put images in the grey dotted box "unsupported placeholder"

Productionizing ML: Speed, with simplicity

Container / script

26

Please write the title in all capital letters

Put images in the grey dotted box "unsupported placeholder"

Productionizing ML: Speed, with simplicity

{experiment_id_ts}-{build_number}

27

Please write the title in all capital letters

Put images in the grey dotted box "unsupported placeholder"

Productionizing ML: Speed, with simplicity

{experiment_id_ts}-{build_number}

28

Please write the title in all capital letters

Put images in the grey dotted box "unsupported placeholder"

Productionizing ML: Speed, with simplicity

29

Please write the title in all capital letters

Put images in the grey dotted box "unsupported placeholder"

Productionizing ML: Speed, with cost efficiency

30

Please write the title in all capital letters

Put images in the grey dotted box "unsupported placeholder"

ML pipelines should be safe and understandable

Did it run properly?

How many times did the pipeline run?

When?

Who has permissions to run the pipeline?

When was the pipeline created?

31

Please write the title in all capital letters

Put images in the grey dotted box "unsupported placeholder"

ML pipelines should be safe and understandable

What happened in each step of the pipeline?

32

Please write the title in all capital letters

Put images in the grey dotted box "unsupported placeholder"

ML pipelines should be safe and understandable

How long did it run?

When did it run?

Name?

Who had permissions to run it?

Did it run properly?

33

Please write the title in all capital letters

Put images in the grey dotted box "unsupported placeholder"

ML pipelines should be safe and understandable

What algorithm was used?

What did it run on?

How was the data loaded?

34

Please write the title in all capital letters

Put images in the grey dotted box "unsupported placeholder"

ML pipelines should be safe and understandable

What exact data was used for training?

What exact data was used for testing?

35

Please write the title in all capital letters

Put images in the grey dotted box "unsupported placeholder"

ML pipelines should be safe and understandable

How was the training monitored

What parameters were fed to the model

Where was the output model stored

36

Please write the title in all capital letters

Put images in the grey dotted box "unsupported placeholder"

ML pipelines should be safe and understandable

How did the training progress?

37

Please write the title in all capital letters

Put images in the grey dotted box "unsupported placeholder"

ML pipelines should be safe and understandable

Where is the model deployed?

When was the model deployed?

Is the model in use?

38

Please write the title in all capital letters

Put images in the grey dotted box "unsupported placeholder"

ML pipelines should be safe and understandable

What training job resulted in the deployment?Which model(s) was deployed?

What instances are the model(s) deployed?

How much traffic routed to which model?

39

Please write the title in all capital letters

Put images in the grey dotted box "unsupported placeholder"

ML pipelines should be safe and understandable

How is the model endpoint performing?

40

Please write the title in all capital letters

Put images in the grey dotted box "unsupported placeholder"

Scaling ML at Zalando

We’re hiring!

suhas.guruprasad@zalando.dekshitij.kumar@zalando.de