AIF-C01 Machine Learning Operations (MLOps) 1 — Questions and Answers
Question 1: What is the primary goal of MLOps?
- To maximize the number of ML models deployed
- To streamline the end-to-end lifecycle of ML models from development to production (Correct answer)
- To replace data scientists with automated tools
- To minimize the cost of cloud computing for AI workloads
Correct answer: To streamline the end-to-end lifecycle of ML models from development to production
MLOps combines ML, DevOps, and data engineering practices to standardize and automate the full ML lifecycle including building, deploying, and monitoring models.
Question 2: Which type of drift occurs when the statistical properties of the input features change over time?
- Concept drift
- Model drift
- Data drift (Correct answer)
- Label drift
Correct answer: Data drift
Data drift (also called feature drift or covariate shift) refers to changes in the statistical properties of model input features compared to the training data.
Question 3: What AWS service provides a central repository to catalog, version, and manage ML models?
- Amazon S3
- AWS CodeArtifact
- Amazon SageMaker Model Registry (Correct answer)
- AWS Glue Data Catalog
Correct answer: Amazon SageMaker Model Registry
Amazon SageMaker Model Registry allows teams to catalog models, manage model versions, associate metadata, and control model approval status for deployment.
Question 4: In a CI/CD pipeline for ML, what does 'CT' (Continuous Training) specifically refer to?
- Continuously testing model endpoints for latency
- Automatically retraining models when new data or triggers are detected (Correct answer)
- Continuously monitoring training job costs
- Running hyperparameter tuning on a scheduled basis
Correct answer: Automatically retraining models when new data or triggers are detected
Continuous Training (CT) automatically re-triggers the model training pipeline when new data arrives, performance degrades, or scheduled intervals occur.
Question 5: What is the purpose of a baseline in Amazon SageMaker Model Monitor?
- To set the minimum acceptable model accuracy threshold
- To define the expected statistical properties of input data and model outputs for comparison (Correct answer)
- To establish the initial training dataset for a model
- To configure the default instance type for inference endpoints
Correct answer: To define the expected statistical properties of input data and model outputs for comparison
A baseline captures statistics and constraints from the training data so Model Monitor can compare live inference data against it to detect drift and violations.
Question 6: Which deployment strategy sends a small percentage of live traffic to a new model version while the majority goes to the existing version?
- Blue/green deployment
- Rolling deployment
- Canary deployment (Correct answer)
- Shadow deployment
Correct answer: Canary deployment
Canary deployment routes a small slice of production traffic (e.g., 5%) to the new model version, allowing real-world validation before a full rollout.
Question 7: What does Amazon SageMaker Pipelines provide?
- A managed ETL service for transforming raw datasets
- An orchestration tool for automating and reproducing end-to-end ML workflows (Correct answer)
- A deployment service for containerized web applications
- A monitoring dashboard for AWS billing and cost allocation
Correct answer: An orchestration tool for automating and reproducing end-to-end ML workflows
SageMaker Pipelines is a purpose-built CI/CD service for ML that lets you define, automate, and track each step of the ML workflow as a directed acyclic graph (DAG).
AIF-C01 Machine Learning Operations (MLOps) 1
What is the primary goal of MLOps?