ML MLOp Ops CI CI/CD CD for or Ma Machine Le Learn rning - - PowerPoint PPT Presentation

▶
ml mlop ops
SMART_READER_LITE
LIVE PREVIEW

ML MLOp Ops CI CI/CD CD for or Ma Machine Le Learn rning - - PowerPoint PPT Presentation

Jan 09, 2024 119 likes •804 views

ML MLOp Ops CI CI/CD CD for or Ma Machine Le Learn rning SASHA ROSENBAUM Sasha Rosenbaum Sr. Program Manager @DivineOps https://www.deliveryconf.com/ Agenda Machine Learning 101 ML CI/CD Pipeline Overview Potential

slide-1
SLIDE 1
slide-2
SLIDE 2

ML MLOp Ops

CI CI/CD CD for

  • r Ma

Machine Le Learn rning

SASHA ROSENBAUM

slide-3
SLIDE 3

Sasha Rosenbaum

  • Sr. Program Manager

@DivineOps

slide-4
SLIDE 4

https://www.deliveryconf.com/

slide-5
SLIDE 5

Agenda

§Machine Learning 101 §ML CI/CD Pipeline Overview §Potential implementation §Demo

slide-6
SLIDE 6

Trigger the pipeline!

slide-7
SLIDE 7

What is MLOps and WHY should you care?

slide-8
SLIDE 8

Machine Learning (ML)

Is the science of getting computers to act Without being explicitly programmed

slide-9
SLIDE 9
slide-10
SLIDE 10

Machine Learning vs DevOps Google searches

slide-11
SLIDE 11

Python questions on Stack Overflow

slide-12
SLIDE 12

OK, but why should YOU care?

slide-13
SLIDE 13

13

slide-14
SLIDE 14

Data Scientists just want to Data Science

slide-15
SLIDE 15

Deep Learning

Some ML training algorithms are complex

slide-16
SLIDE 16

Deep Learning - Backpropagation

Some ML training algorithms are complex

slide-17
SLIDE 17

Typical data scientist work environment

slide-18
SLIDE 18

We’ve got the notebook into source control!

slide-19
SLIDE 19

Programming

Algorithm Data Answers

slide-20
SLIDE 20

Machine Learning

Algorithm Data Answers

slide-21
SLIDE 21

Model Data Answers

Machine Learning

slide-22
SLIDE 22

Model Data Answers

Machine Learning

slide-23
SLIDE 23

Predictions Data Model Data Answers

Machine Learning

slide-24
SLIDE 24

How do we put the model in production?

slide-25
SLIDE 25

What is an ML model?

slide-26
SLIDE 26

Linear Regression – Housing Prices

The training finds a and b such that

Y = a+bX+ϵ

slide-27
SLIDE 27

Deep Learning

The input and

  • utput may be

vectors ! 𝑌, ! 𝑍

slide-28
SLIDE 28

Image Classification

slide-29
SLIDE 29

ML Model

A definition of the mathematical formula with a number of parameters that are learned from the data

slide-30
SLIDE 30

Isn’t this just an API endpoint?!

slide-31
SLIDE 31

Do models really change that often?

slide-32
SLIDE 32

Models must be improved continuously

slide-33
SLIDE 33

The dataset matters!

slide-34
SLIDE 34
slide-35
SLIDE 35
slide-36
SLIDE 36
slide-37
SLIDE 37
slide-38
SLIDE 38
slide-39
SLIDE 39
slide-40
SLIDE 40
slide-41
SLIDE 41
slide-42
SLIDE 42

The model predictions depend

  • n what it has “seen”
slide-43
SLIDE 43

=> Dataset is part of the model version!

slide-44
SLIDE 44

FB FBLear arne ner Fl Flow Te TensorFlow Extended Ube Uber’s Michelangelo Mic Microso soft Ae Aether

slide-45
SLIDE 45

But I don’t work at a big company with thousands of ML engineers!

slide-46
SLIDE 46

How do we iterate?

slide-47
SLIDE 47

Machine Learning Lifecycle

slide-48
SLIDE 48

DevOps/SRE Data Scientist

  • Quick iteration
  • Versioning
  • Reuse
  • Great tools
  • Ease of

management

  • Unlimited scale
  • Eliminating drift
  • Quick iteration
  • Versioning
  • Reuse
  • Compliance
  • Observability
  • Uptime
  • Updates

Friends?

slide-49
SLIDE 49

App developer

MLOps Workflow

Build app Collaborate Test app Release app Monitor app Model reproducibility Model retraining Model deployment Model validation

Data scientist

slide-50
SLIDE 50

Code, dataset, and environment versioning

Model reproducibility Model retraining Model deployment Model validation Build app Collaborate Test app Release app Monitor app

MLOps Workflow

App developer Data scientist

slide-51
SLIDE 51

MLOps Workflow

Model reproducibility Model retraining Model deployment Model validation

Automated ML ML Pipelines Hyperparameter tuning

Train model Build app Collaborate Test app Release app Monitor app

App developer Data scientist

slide-52
SLIDE 52

MLOps Workflow

Model validation & certification

Model reproducibility Model retraining Model deployment Model validation Train model Validate model Build app Collaborate Test app Release app Monitor app

App developer Data scientist

slide-53
SLIDE 53

MLOps Workflow

Model packaging Simple deployment

Model reproducibility Model retraining Model deployment Model validation Train model Validate model Deploy model Build app Collaborate Test app Release app Monitor app

App developer Data scientist

slide-54
SLIDE 54

MLOps Workflow

Model management & monitoring Model performance analysis

Model reproducibility Model retraining Model deployment Model validation Train model Validate model Deploy model Monitor model Retrain model Build app Collaborate Test app Release app Monitor app

App developer Data scientist

slide-55
SLIDE 55

Build Your Own MLOps Platform

+ +

slide-56
SLIDE 56

ML Pipeline

A reusable, scaleable ML workflow template

Kubeflow pipeline

A reusable, scalable ML workflow template that runs on containers

slide-57
SLIDE 57

Azure ML

  • Prep data
  • Train
  • Test
  • Deploy
  • Manage
slide-58
SLIDE 58

Demo

slide-59
SLIDE 59

Even a simple CI/CD pipeline is better than none!

slide-60
SLIDE 60

DevOps Because change is the

  • nly constant

in life

slide-61
SLIDE 61

AI Ethics

slide-62
SLIDE 62

Bias is a property of information

slide-63
SLIDE 63

We must build AI responsibly

slide-64
SLIDE 64

Build AI responsibly!

slide-65
SLIDE 65

Thank You!

@DivineOps

slide-66
SLIDE 66

Questions?

slide-67
SLIDE 67

Resources

GitHub repo https://www.kubeflow.org/docs/azure/azureendtoend/ Deploy Kubeflow on Azure https://www.kubeflow.org/docs/azure/deploy/install-kubeflow/ Example Kubeflow Azure Pipeline https://www.kubeflow.org/docs/azure/azureendtoend/ Release pipeline https://dev.azure.com/sasrose/kubeflow/_release