Self-service ML platform architecture in MLOps - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

Key components of a self-service ML platform

Which component is essential in a self-service ML platform to allow data scientists to train models without deep infrastructure knowledge?

A Direct access to raw hardware without abstraction

B Manual server configuration by data scientists

C A user-friendly interface with automated resource provisioning

D A command-line tool requiring complex scripting

💡 Hint

Think about what helps non-technical users easily use the platform.

A self-service ML platform must provide a simple interface that automates infrastructure tasks, so data scientists can focus on modeling.

🔀 Workflow

intermediate

Order of steps in a self-service ML platform workflow

What is the correct order of these steps in a typical self-service ML platform workflow?

A 2,1,4,3

B 1,2,3,4

C 4,2,1,3

D 2,4,1,3

💡 Hint

Think about what must happen before training and deployment.

Data must be ingested and preprocessed first, then the model is trained, evaluated, and finally deployed.

❓ Troubleshoot

advanced

Troubleshooting model deployment failures

In a self-service ML platform, a model deployment fails with an error indicating insufficient compute resources. What is the most likely cause?

A The model code has syntax errors

B The platform's resource quota for the user is exceeded

C The data preprocessing step was skipped

D The model evaluation metrics are too low

💡 Hint

Consider platform resource limits rather than code or data issues.

Deployment failures due to resource errors usually mean the user exceeded their allowed compute quota on the platform.

✅ Best Practice

advanced

Best practice for managing model versions in self-service ML platforms

Which practice is best for managing multiple model versions in a self-service ML platform?

A Deploy models without tracking versions

B Overwrite the existing model file on deployment

C Keep models only on local machines of data scientists

D Use a centralized model registry with version control

💡 Hint

Think about how to keep track of models safely and clearly.

A centralized model registry helps track, compare, and manage different model versions systematically.

💻 Command Output

expert

Output of Kubernetes command for ML platform pods

What is the output of the command kubectl get pods -l app=ml-platform -o jsonpath='{.items[*].metadata.name}' if there are three pods named ml-platform-1, ml-platform-2, and ml-platform-3 running?

A ml-platform-1 ml-platform-2 ml-platform-3

B ["ml-platform-1", "ml-platform-2", "ml-platform-3"]

C ml-platform-1,ml-platform-2,ml-platform-3

D Error: label selector not found

💡 Hint

Consider how jsonpath outputs multiple items separated by spaces.

The jsonpath expression outputs pod names separated by spaces without commas or brackets.