ML OPs

Question 2

What does MLOps stand for?

Answer 2

The way teams build, deploy, monitor and update machine-learning models in real life

MLOps encompasses everything around the training of machine-learning models.

Question 3

What is the real ML life cycle in MLOps?

Answer 3

• Define business goal
• Get data
• Train models
• Test models
• Deploy model
• Monitor model
• Retrain / fix
• Repeat

MLOps is about running this loop properly.

Question 4

Why do we need MLOps?

Answer 4

To manage:
* Lots of data
* Lots of models
* Lots of users
* Lots of updates

Without MLOps, things can break very fast.

Question 5

What are some problems MLOps tries to stop?

Answer 5

• Model suddenly becomes inaccurate
• Data changes silently
• Wrong model version gets deployed
• No one knows which model is in production
• Retraining is manual and slow

These issues highlight the importance of MLOps in maintaining model reliability.

Question 6

What must you have before training in MLOps?

Answer 6

• A clear business objective
• Success metrics
• Acceptance criteria

No MLOps magic can happen without these foundational elements.

Question 7

What does model training & experiments include?

Answer 7

• Running experiments
• Tuning hyperparameters
• Comparing models
• Tracking results
• Storing models and versions

This phase is crucial for developing effective machine-learning models.

Question 8

What is an important point regarding a trained model?

Answer 8

A trained model is not a product

You still need code around it, APIs, pipelines, scalability, and automation.

Question 9

What does deployment in MLOps refer to?

Answer 9

Putting the model into the real system (web app, backend, mobile app, etc.)

Deployment is a critical step in making the model usable.

Question 10

What should you monitor after release of a model?

Answer 10

• Prediction quality
• Error rates
• Data changes
• Performance (speed, cost, failures)

Monitoring is a huge aspect of MLOps.

Question 11

What is a very important idea regarding retraining and updating?

Answer 11

Releasing a model is NOT the end

If monitoring shows data changed or accuracy dropped, you must go back to training.

Question 12

What is the most important idea in MLOps?

Answer 12

It is a continuous cycle

The cycle includes training, deploying, monitoring, improving, and redeploying.

Question 13

Who does MLOps involve?

Answer 13

• Data scientists
• Software engineers
• Platform / cloud engineers
• Product people

MLOps is a cross-functional discipline connecting ML work, software engineering, and operations.

Question 14

How does MLOps compare to DevOps?

Answer 14

MLOps = build and operate ML systems

DevOps focuses on building and operating software.

Question 15

What is the ultra-short summary of MLOps?

Answer 15

MLOps is the discipline that makes machine-learning models usable and reliable in production

It covers training, testing, packaging, deployment, monitoring, retraining, teamwork, and tooling.

Question 16

What is an ML pipeline?

Answer 16

The full chain that turns raw data into a running prediction system

Not just a model.

Question 17

What are the first two steps in an ML pipeline?

Answer 17

• Business understanding
• Data collection

Business understanding involves defining the problem and success metrics.

Question 18

What is the purpose of data cleaning in an ML pipeline?

Answer 18

Fix missing values, errors, broken rows, weird formats

Ensures the data is usable for modeling.

Question 19

What does feature engineering involve?

Answer 19

Turn raw data into useful inputs

Essential for improving model performance.

Question 20

What is data augmentation used for?

Answer 20

Create extra data (mainly for images / sound / text)

Helps improve model robustness.

Question 21

What is the role of data annotation in supervised learning?

Answer 21

Add labels

Necessary for training supervised models.

Question 22

What is the purpose of feature selection?

Answer 22

Keep only the useful inputs

Reduces dimensionality and improves model efficiency.

Question 23

What does feature scaling achieve?

Answer 23

Put numbers on comparable ranges

Important for algorithms sensitive to the scale of data.

Question 24

What is involved in the modeling step?

Answer 24

Train the model

This is where the actual learning occurs.

Question 25

What is the purpose of **evaluation** in an ML pipeline?

Answer 25

Check how good it is (metrics, validation, tests) ## Footnote Ensures the model meets performance criteria.

Question 26

What does **deployment** refer to in an ML pipeline?

Answer 26

Put the system into real use ## Footnote Involves making the model accessible for predictions.

Question 27

What is the focus of the **monitoring** step?

Answer 27

Watch it in production: is it still accurate? ## Footnote Ensures ongoing performance and reliability.

Question 28

What is the final output of an ML pipeline?

Answer 28

A real prediction system that the business actually uses ## Footnote This is the ultimate goal of the pipeline.

Question 29

True or false: You deploy a model in an ML pipeline.

Answer 29

FALSE ## Footnote You deploy a pipeline, which includes data cleaning, feature processing, scaling, transformations, the model, and prediction logic.

Question 30

Who primarily builds **ML pipelines**?

Answer 30

* MLOps engineers * Software engineers * Data scientists ## Footnote It’s system architecture work.

Question 31

What is the difference between a **Jupyter notebook** and an **ML pipeline**?

Answer 31

A Jupyter notebook = experiment; An ML pipeline = production system ## Footnote This highlights the transition from experimentation to deployment.

Question 32

What are the key components of an **ML pipeline**?

Answer 32

* Business goal * Data → cleaning → features * Train model * Evaluate * Deploy * Monitor * Improve ## Footnote This summarizes the entire process from start to finish.

Question 33

What are the **four main steps** in the process described?

Answer 33

* Modelling * Evaluation * Deployment * Monitoring ## Footnote These steps outline the workflow for developing and maintaining a machine learning model.

Question 34

In the **Modelling** step, what must you document?

Answer 34

* Algorithm used * Features used * Hyperparameters tried * Error limits accepted ## Footnote Documentation is crucial for reproducibility in model development.

Question 35

What is the goal of the **Evaluation** step?

Answer 35

To determine if the model can predict unknown data well ## Footnote This is assessed by running the model on a test set and computing performance metrics.

Question 36

True or false: A trained model is enough for deployment.

Answer 36

FALSE ## Footnote Deployment requires the model, preprocessing, feature steps, scaling, and prediction code.

Question 37

What are the main goals of the **Deployment** step?

Answer 37

* Scalability * Stability * Automation ## Footnote This involves turning the model into a real product and often includes continuous delivery.

Question 38

What two aspects should be monitored after deployment?

Answer 38

* Software performance * Model performance ## Footnote Monitoring ensures that both the software and the model continue to function effectively in real-world conditions.

Question 39

What should you do if model performance drops?

Answer 39

* Go back to earlier pipeline steps * Retrain with new data * Adjust model/hyperparameters * Evaluate again * Redeploy ## Footnote This process ensures that the model remains valid and effective over time.

Question 40

Fill in the blank: You deploy the ________, not just the model.

Answer 40

pipeline ## Footnote This emphasizes the importance of deploying the entire workflow for machine learning.

Question 41

What is the **output** of the Modelling step?

Answer 41

* Model type * Features * Hyperparameters * Learned weights * Training data info (metadata) ## Footnote This output is essential for understanding the model's structure and performance.

Question 42

What does monitoring model performance involve?

Answer 42

Comparing predicted values vs real outcomes ## Footnote This helps determine if the model is still valid in changing conditions.

Question 43

What does **MLOps maturity** refer to?

Answer 43

How automated, reliable, and scalable your ML system is ## Footnote MLOps maturity indicates the effectiveness of managing machine learning operations.

Question 44

What is **Level 0** in MLOps maturity?

Answer 44

* No MLOps * Everything is manual * Train manually * Test manually * Deploy manually * No monitoring ## Footnote Models are treated with uncertainty: 'We trained something… we hope it still works.'

Question 45

What characterizes **Level 1** in MLOps maturity?

Answer 45

* DevOps only (no real MLOps) * Software has automation (builds, tests, releases) * Models treated like files * Little or no feedback on model performance ## Footnote Code is professional, but ML processes remain messy.

Question 46

What advancements are present in **Level 2** of MLOps maturity?

Answer 46

* Automated model training * Model performance tracking * Better model management ## Footnote Deployment is still mostly manual, but you can determine which model is better and which experiment worked.

Question 47

What features define **Level 3** in MLOps maturity?

Answer 47

* Automated deployment * Proper train/test/production environments * A/B testing between models * Full traceability ## Footnote The system can safely decide which model to deploy.

Question 48

What does **Level 4** in MLOps maturity entail?

Answer 48

* Continuous monitoring * Production performance feeds back automatically * Automatic retraining triggers * Models compared against live metrics * Zero-downtime releases ## Footnote The system improves itself autonomously.

Question 49

What does the **tiny table** summarize about MLOps maturity levels?

Answer 49

* Level 0: everything manual * Level 1: software automated, ML not * Level 2: training & tracking automated * Level 3: deployment automated * Level 4: monitoring + retraining automated ## Footnote This table provides an easy reference for understanding MLOps maturity.

Question 50

What roles are necessary to move up the MLOps maturity levels?

Answer 50

* Data scientists * Data engineers * Software engineers * MLOps/platform people ## Footnote Teams must collaborate and stop being siloed to improve MLOps maturity.

Question 51

What changes occur as you mature in MLOps?

Answer 51

* Manual runs → automated pipelines * Manual releases → CI/CD * No tracking → full version control * No monitoring → performance feedback * Manual retraining → triggered retraining ## Footnote These changes enhance the efficiency and reliability of ML operations.

Question 52

True or false: A successful ML project is only about technology.

Answer 52

FALSE ## Footnote Management support, access to data, easy deployment paths, and time for maintenance are also crucial.

Question 53

What should you always ask regarding MLOps projects?

Answer 53

'What MLOps level is this project actually at?' ## Footnote This question helps identify the current maturity level and plan for improvements.

Question 54

What is the **ultra-short summary** of MLOps maturity?

Answer 54

MLOps maturity = how automated and reliable your ML lifecycle is ## Footnote It emphasizes the transition from manual experiments to self-monitoring, self-retraining systems.

Question 55

What does **CRISP-DM** stand for?

Answer 55

Cross-Industry Standard Process for Data Mining ## Footnote CRISP-DM is a checklist for data projects that prevents jumping straight to modeling.

Question 56

List the **six steps** of the CRISP-DM workflow.

Answer 56

* Business understanding * Data understanding * Data preparation * Modelling * Evaluation * Deployment ## Footnote These steps guide the execution of a data/ML project.

Question 57

What is the first step in the **CRISP-DM** workflow?

Answer 57

Business understanding ## Footnote This step involves defining the problem and success criteria.

Question 58

What does the **Data understanding** step involve?

Answer 58

* Assessing available data * Evaluating data quality * Exploring data with visualizations ## Footnote Engaging with business stakeholders to clarify data meanings is also important.

Question 59

What is typically the **longest step** in the CRISP-DM process?

Answer 59

Data preparation ## Footnote This step includes cleaning data, merging files, fixing errors, and feature engineering.

Question 60

What are the key tasks in the **Modelling** step?

Answer 60

* Splitting data into train/validation/test * Trying different models * Tuning hyperparameters ## Footnote This step focuses on developing the predictive model.

Question 61

What is evaluated in the **Evaluation** step?

Answer 61

* Model performance on unseen data * Achievement of business targets ## Footnote This step determines if the model meets the defined success criteria.

Question 62

What does the **Deployment** step entail?

Answer 62

* Deploying the entire pipeline * Documenting the process * Planning for monitoring and maintenance ## Footnote This step ensures the model is operational and maintained.

Question 63

True or false: In the CRISP-DM workflow, you move in a straight line from one step to the next.

Answer 63

FALSE ## Footnote The process involves going back and forth between steps as needed.

Question 64

What is the main advantage of **cloud computing** for ML projects?

Answer 64

Allows renting computers (including GPUs) instead of owning them ## Footnote This makes machine learning more affordable.

Question 65

List the **three service types** in cloud computing.

Answer 65

* IaaS – Infrastructure * PaaS – Platform * SaaS – Software ## Footnote These services provide varying levels of control and tools for ML projects.

Question 66

Name three major **cloud platforms** mentioned.

Answer 66

* Google Cloud Platform (GCP) * Amazon Web Services (AWS) * Microsoft Azure ## Footnote These platforms offer various tools and services for machine learning.

Question 67

What is a **container** in the context of cloud ML systems?

Answer 67

An environment where the app, model, and libraries run together ## Footnote This ensures reliable deployment.

Question 68

What is a **reality check** regarding cloud platforms?

Answer 68

* They are powerful * Often complex * May be overkill for small projects ## Footnote Simpler platforms are often better for learning or demos.

Question 69

Summarize the **CRISP-DM** workflow in one line.

Answer 69

A simple project workflow that helps plan, organize, and iterate properly ## Footnote It is essential for managing data and ML projects effectively.

Question 70

Summarize the role of **cloud computing** in ML.

Answer 70

Enables running ML on rented machines and GPUs, making it accessible and scalable ## Footnote This has transformed the affordability of machine learning.