DL-04 - MLOps

Question 1

DL-04 - MLOps

What are the steps of the DevOps cycle? (7, PCVPRCM)

Answer 1

1) Plan
2) Create
3) Verify
4) Package
5) Release
6) Configure
7) Monitor

Question 2

DL-04 - MLOps

What are the goals of MLOps? (4, EMDQ)

Answer 2

• Faster experimentation
• Faster model development
• Faster deployment of updated models
• Quality assurance

Question 3

DL-04 - MLOps

Name the steps of the MLOps lifecycle? (8, EDDTRDIM)

Answer 3

1) EDA (Exploratory data analysis)
2) Data Prep
3) Develop
4) (Re-)Train
5) Review
6) Deploy
7) Inference
8) Monitor

Question 4

DL-04 - MLOps

What happens in CI in MLOps? (2)

Answer 4

• When you update your code/data, the ML pipeline reruns.
• Integration is run before release, e.g. performance/error tests

Question 5

DL-04 - MLOps

What are some things you should monitor in MLOps? (2)

Answer 5

• Performance over time
• Data drift

Question 6

DL-04 - MLOps

What happens in CD in MLOps? (2 examples)

Answer 6

• Test model in staging environment.
• Soft release to some portion of customers, canary release.

Question 7

DL-04 - MLOps

What is data drift?

Answer 7

Data drift is the change in data distribution over time, affecting the model’s performance and accuracy.

Question 8

DL-04 - MLOps

What is a canary release?

Answer 8

A canary release is a deployment strategy where a new model or application version is gradually rolled out to a small percentage of users to test its performance and identify potential issues before a full-scale release.

Question 9

DL-04 - MLOps

What are the most important things about DVC? (5, VRCPS)

Answer 9

• Version control for data
• Reproducibility
• collaboration
• pipeline management
• remote storage integration.

Question 10

DL-04 - MLOps

What are the most important things about CML? (4, ACSR)

Answer 10

• automation
• collaboration
• scalability
• reproducibility

Question 11

DL-04 - MLOps

How does CML mitigate the problem of people working on the same code?

Answer 11

By enabling collaboration through version control and automated machine learning pipelines.

Question 12

DL-04 - MLOps

How can you ensure reproducibility in deep learning projects?

Answer 12

By tracking what logic, dataset, and pipeline were used to get a certain model/performance.

Question 13

DL-04 - MLOps

In Github, what is a Self-hosted runner?

Answer 13

A Self-hosted runner is an installable application that you can host on your own hardware to run GitHub Actions and Workflows.

Question 14

DL-04 - MLOps

What does a self-hosted runner for Github allow you to do?

Answer 14

Run GitHub Actions workflows on your own infrastructure.