1 of 64

MLOps 101

What and Why

2 of 64

แนะนำตัว

Boyd Sorratat�AI Engineer�@CJ Express Group Co., Ltd.

2

3 of 64

อะไรคือ Machine Learning

3

4 of 64

อะไรคือ Machine Learning

4

Problem

Machine

Result

5 of 64

อะไรคือ Machine Learning

5

Problem

Machine

Result

6 of 64

อะไรคือ Machine Learning

7 of 64

อะไรคือ Machine Learning

8 of 64

Machine Learning Projects

9 of 64

Machine Learning Projects

Data Scientist

Focus ที่ Improve Model

ทำไงก็ได้ให้ Model แม่นๆ

10 of 64

Machine Learning Projects

Data Scientist

Focus ที่ Improve Model

ทำไงก็ได้ให้ Model แม่นๆ

ML Engineer

เตรียม ML Projects ให้พร้อม Deploy ทำให้ Code สามารถ Maintenance และ Scalability

11 of 64

Machine Learning Projects

Data Scientist

Focus ที่ Improve Model

ทำไงก็ได้ให้ Model แม่นๆ

ML Engineer

เตรียม ML Projects ให้พร้อม Deploy ทำให้ Code สามารถ Maintenance และ Scalability

ML Ops

Deploy ดูแลระบบ ML Projects พร้อมใช้งานตลอดเวลา น่าเชื่อถือ และตรวจสอบได้

12 of 64

Machine Learning Projects

Data Scientist

Timeline

Deadline

13 of 64

Machine Learning Projects

Data Scientist

Timeline

Explore Data

Deadline

14 of 64

Machine Learning Projects

Data Scientist

Timeline

Explore Data

Deadline

15 of 64

Machine Learning Projects

Data Scientist

Timeline

Preprocessing Data

Ideal

Deadline

16 of 64

Machine Learning Projects

Data Scientist

Timeline

Preprocessing Data

Reality

Deadline

17 of 64

Machine Learning Projects

Data Scientist

Timeline

Modeling

Deadline

18 of 64

Machine Learning Projects

Data Scientist

Timeline

Modeling

Hypertuning

Post processing

Slide Presentation

Blaa bla bla

Deadline

19 of 64

Machine Learning Projects

Management or Business Team

Data Scientist

20 of 64

Machine Learning Projects

Management or Business Team

Data Scientist

ฝาก Projects อื่นต่อด้วยนะ

อีกประมาณ 100 Projects

21 of 64

Machine Learning Projects

Management or Business Team

ได้ครับพี่

ดีครับผม

เหมาะสมครับนาย

สบายครับท่าน

Data Scientist

22 of 64

Machine Learning Projects

Data Scientist

ส่งต่องาน

ML Engineer

ML Ops

23 of 64

Machine Learning Projects

Data Scientist

ส่งต่องาน

ML Engineer

ML Ops

ชิบหาx ละพวกตรู

24 of 64

อะไรคือ MLOps

24

MLOps (machine learning operations)

25 of 64

อะไรคือ MLOps

MLOps ไม่ใช่สถานที่ แต่คือผู้คน

25

MLOps (machine learning operations)

26 of 64

อะไรคือ MLOps

26

MLOps (machine learning operations)

Ref: Machine Learning Operations (MLOps): Overview, Definition, and Architecture

27 of 64

อะไรคือ MLOps

27

The goal of MLOps is to reduce technical friction to get the model from an idea into production in the shortest possible time to market with as little risk as possible.

MLOps (machine learning operations)

Ref: https://valohai.com/mlops/

28 of 64

Data Scientist สู่ MLOps

Data Scientist

ML Engineer

ML Ops

AI Engineer

29 of 64

MLOps Principle

  1. Iterative-Incremental Process in MLOps
  2. Automation
  3. Continuous X
  4. Versioning
  5. Experiments Tracking
  6. Testing
  7. Monitoring
  8. Reproducibility

Ref: https://ml-ops.org/content/mlops-principles

30 of 64

Iterative-Incremental Process in MLOp

​

31 of 64

Automation

3 levels of automation

  1. Manual process.
  2. ML pipeline automation.
  3. CI/CD pipeline automation.

32 of 64

Automation

33 of 64

Continuous X

Model Deployment

  • Continuous Integration (CI)
  • Continuous Delivery (CD)
  • Continuous Training (CT)
  • Continuous Monitoring (CM)

34 of 64

Versioning

​

35 of 64

Versioning

​

36 of 64

Versioning

​

37 of 64

Versioning

​

Preparing Data

38 of 64

Versioning

​

Preparing Data

Modeling

39 of 64

Versioning

​

Preparing Data

Modeling

40 of 64

Versioning

​

​

Code

Data

Model

41 of 64

Versioning

​

​

Code

Data

Model

  • Keep code version
  • Size limitation

42 of 64

Versioning

​

​

Code

Data

Model

  • Solve Git’s limitation
  • Easy to learn
  • Dataset tracking
  • Keep code version
  • Size limitation

43 of 64

Versioning

​

​

Code

Data

Model

  • Solve Git’s limitation
  • Easy to learn
  • Dataset tracking
  • Keep code version
  • Size limitation

44 of 64

Versioning

​

​

45 of 64

Experiment Tracking

In contrast to the traditional software development process, in ML development, multiple experiments on model training can be executed in parallel before making the decision what model will be promoted to production.

46 of 64

Testing

Ref: The ML Test Score: A Rubric for ML Production Readiness and Technical Debt Reduction

47 of 64

Testing

Ref: The ML Test Score: A Rubric for ML Production Readiness and Technical Debt Reduction

48 of 64

Testing

Which one is Dog!

49 of 64

Monitoring

50 of 64

Reproducibility

  • Collecting data
  • Feature Engineering
  • Model Training
  • Model Deployment

51 of 64

MLE MLOps

MLOps

Platform Automation

Data Automation

DevOps

SageMaker, VertexAI, Valohai

Clean, Curated, Check

Produce and Reliable

MLE

Ref: https://github.com/noahgift

52 of 64

MLOps

DS

SWE

DE

BU

Rule of 25%

DevOps

Data

Model

Business

Ref: https://github.com/noahgift

53 of 64

MLOps

Ref: Machine Learning Operations (MLOps): Overview, Definition, and Architecture

54 of 64

MLOps

Ref: https://neptune.ai/wp-content/uploads/MLOps_cycle.jpg

55 of 64

MLOps Challenge Topics

MLOps

56 of 64

MLOps Challenge Topics

MLOps

MLSecOps

57 of 64

MLOps Challenge Topics

Practical Machine Learning Security: Major Security Flaws in ML

and How to Avoid Them with MLSecOps

58 of 64

Resources Learning

Bootcamp

59 of 64

Resources Learning

Video walkthrough

60 of 64

Resources Learning

Blog and Products

https://ml-ops.org/

https://valohai.com/mlops/

https://neptune.ai/blog/

61 of 64

Q&A

62 of 64

Q&A

63 of 64

Q&A

64 of 64

MLOps 101

What and Why