the data science andai lifecycle - deutsche messe...

13
The Data Science and AI lifecycle Stephan Reimann IBM Deutschland [email protected] stereimann IBM Cloud / DEVELOPERS @ CEBIT - Data Science and AI Lifecycle / June 2018 / © 2018 IBM Corporation

Upload: others

Post on 29-Aug-2020

2 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: The Data Science andAI lifecycle - Deutsche Messe AGfiles.messe.de/abstracts/88812_GB_CeBit_Meetup_IBM_AI... · 2018. 6. 18. · 1 Prepare& understandthe data & 2 Buildthemachinelearning

The Data Science and AI lifecycle

Stephan ReimannIBM [email protected]

stereimann

IBM Cloud / DEVELOPERS @ CEBIT - Data Science and AI Lifecycle / June 2018 / © 2018 IBM Corporation

Page 2: The Data Science andAI lifecycle - Deutsche Messe AGfiles.messe.de/abstracts/88812_GB_CeBit_Meetup_IBM_AI... · 2018. 6. 18. · 1 Prepare& understandthe data & 2 Buildthemachinelearning

Our journey today

CRISP-DMAn industry standard process fordata science

2 Build the machine learning model

3 Deploy the machine learning model

4 Learning never stops – continuous learning

1 Prepare & understand the data Usually 80%of the effort

The Cool stuffeverybody talks about

Just developerstuff?

The lifecyclehas more steps

IBM Cloud / DEVELOPERS @ CEBIT - Data Science and AI Lifecycle / June 2018 / © 2018 IBM Corporation 2

Page 3: The Data Science andAI lifecycle - Deutsche Messe AGfiles.messe.de/abstracts/88812_GB_CeBit_Meetup_IBM_AI... · 2018. 6. 18. · 1 Prepare& understandthe data & 2 Buildthemachinelearning

1

Prepare & understand thedata

&

2

Build the machine learningmodel

3IBM Cloud / DEVELOPERS @ CEBIT - Data Science and AI Lifecycle / June 2018 / © 2018 IBM Corporation

Page 4: The Data Science andAI lifecycle - Deutsche Messe AGfiles.messe.de/abstracts/88812_GB_CeBit_Meetup_IBM_AI... · 2018. 6. 18. · 1 Prepare& understandthe data & 2 Buildthemachinelearning

Step 1: Build the model

4

Data Science can be done in many ways:

Result:A machine learning model that describes theinsights learned from the data.

IBM Cloud / DEVELOPERS @ CEBIT - Data Science and AI Lifecycle / June 2018 / © 2018 IBM Corporation

Page 5: The Data Science andAI lifecycle - Deutsche Messe AGfiles.messe.de/abstracts/88812_GB_CeBit_Meetup_IBM_AI... · 2018. 6. 18. · 1 Prepare& understandthe data & 2 Buildthemachinelearning

3

Deploy the machine learningmodel

5IBM Cloud / DEVELOPERS @ CEBIT - Data Science and AI Lifecycle / June 2018 / © 2018 IBM Corporation

Page 6: The Data Science andAI lifecycle - Deutsche Messe AGfiles.messe.de/abstracts/88812_GB_CeBit_Meetup_IBM_AI... · 2018. 6. 18. · 1 Prepare& understandthe data & 2 Buildthemachinelearning

Deployment overview

6

Different ways:Rest API, Batch, Streaming

Target: • Apply the learnings to new data = scoring• Integrate machine learning into processes and

applications

Different ways to deploy:

Service with automated model management ...

Build your own API, e.g. using Flask or Function as a Service

IBM Cloud / DEVELOPERS @ CEBIT - Data Science and AI Lifecycle / June 2018 / © 2018 IBM Corporation

Page 7: The Data Science andAI lifecycle - Deutsche Messe AGfiles.messe.de/abstracts/88812_GB_CeBit_Meetup_IBM_AI... · 2018. 6. 18. · 1 Prepare& understandthe data & 2 Buildthemachinelearning

A practical example

7

Code: https://datascience.ibm.com/exchange/public/entry/view/db078b55b82aee7146210a087cb22f89

Tutorial:https://datascience.ibm.com/docs/content/analyze-data/ml-bluemix-app.html?cm_sp=dw-dwtv-_-data-science-_-putting-face-machine-learning

IBM Cloud / DEVELOPERS @ CEBIT - Data Science and AI Lifecycle / June 2018 / © 2018 IBM Corporation

Page 8: The Data Science andAI lifecycle - Deutsche Messe AGfiles.messe.de/abstracts/88812_GB_CeBit_Meetup_IBM_AI... · 2018. 6. 18. · 1 Prepare& understandthe data & 2 Buildthemachinelearning

Deployment aspects

8

• Infrastructure

• Availability

• Automation

• Catalog / Maintain Model information

• Versioning

You just have gotten into DevOpstopics

IBM Cloud / DEVELOPERS @ CEBIT - Data Science and AI Lifecycle / June 2018 / © 2018 IBM Corporation

Page 9: The Data Science andAI lifecycle - Deutsche Messe AGfiles.messe.de/abstracts/88812_GB_CeBit_Meetup_IBM_AI... · 2018. 6. 18. · 1 Prepare& understandthe data & 2 Buildthemachinelearning

4

Learning never stops –continuous learning

9IBM Cloud / DEVELOPERS @ CEBIT - Data Science and AI Lifecycle / June 2018 / © 2018 IBM Corporation

Page 10: The Data Science andAI lifecycle - Deutsche Messe AGfiles.messe.de/abstracts/88812_GB_CeBit_Meetup_IBM_AI... · 2018. 6. 18. · 1 Prepare& understandthe data & 2 Buildthemachinelearning

Motivation

10

Companies are realizing that in many settings machine learning models start degrading soon after they get deployed to production.

(https://www.oreilly.com/ideas/why-continuous-learning-is-key-to-ai )

https://www.slideshare.net/DavidTalby/when-models-go-rogue-hard-earned-lessons-about-using-machine-learning-in-production

IBM Cloud / DEVELOPERS @ CEBIT - Data Science and AI Lifecycle / June 2018 / © 2018 IBM Corporation

Page 11: The Data Science andAI lifecycle - Deutsche Messe AGfiles.messe.de/abstracts/88812_GB_CeBit_Meetup_IBM_AI... · 2018. 6. 18. · 1 Prepare& understandthe data & 2 Buildthemachinelearning

A few words about methods

11

https://www.slideshare.net/DavidTalby/when-models-go-rogue-hard-earned-lessons-about-using-machine-learning-in-production

IBM Cloud / DEVELOPERS @ CEBIT - Data Science and AI Lifecycle / June 2018 / © 2018 IBM Corporation

Page 12: The Data Science andAI lifecycle - Deutsche Messe AGfiles.messe.de/abstracts/88812_GB_CeBit_Meetup_IBM_AI... · 2018. 6. 18. · 1 Prepare& understandthe data & 2 Buildthemachinelearning

Step 3: Continuous Learning

12

Target: Ensure accurate predictionsAs humans, we are continuously learning, machine learning models should do the same to cope with an ever changing world

https://medium.com/ibm-data-science-experience/continuous-learning-on-watson-data-platform-cc39f3fd5042

IBM Cloud / DEVELOPERS @ CEBIT - Data Science and AI Lifecycle / June 2018 / © 2018 IBM Corporation

Page 13: The Data Science andAI lifecycle - Deutsche Messe AGfiles.messe.de/abstracts/88812_GB_CeBit_Meetup_IBM_AI... · 2018. 6. 18. · 1 Prepare& understandthe data & 2 Buildthemachinelearning

Key take away

13

Never forget to update yourmachine learning models overtime!&Automate!!!!

It is easier than you think!IBM Cloud / DEVELOPERS @ CEBIT - Data Science and AI Lifecycle / June 2018 / © 2018 IBM Corporation