Data Science Fundamentals on AWS

Please provide your name *

1. Which phase of the CRISP-DM process typically consumes the largest share of the effort on a real data science project? *

Modelling Data Understanding and Data Preparation Deployment Business Understanding

2. Which AWS service lets you run standard SQL queries directly against data stored in S3, without provisioning a database or cluster? *

AWS Glue Amazon Athena Amazon SageMaker Autopilot AWS Lake Formation

3. What is “target leakage” in a machine learning model? *

When a feature contains information that would not actually be available at the time of prediction When the training dataset is too small to represent the wider population When the target variable contains missing values that were imputed When a model is deployed to an endpoint before it has been approved

4. A credit scoring model scores 98% accuracy on the data it was trained on, but only 62% on the held-out test set. What is the most likely explanation? *

The model has overfitted the training data and does not generalise to unseen data The test set is too large relative to the training set The features were incorrectly scaled before training The model needs to be trained for more iterations

5. What should a Data Scientist reasonably expect a Data Engineer to deliver as part of the handoff? *

Model-ready engineered features built specifically for each model Clean, documented source tables with agreed schemas, quality and refresh SLAs A trained baseline model to compare against Hyperparameter tuning results for the production model

6. What is the primary purpose of SageMaker Experiments? *

To automatically select the best algorithm for a dataset without writing any code To provision and auto-scale the compute instances used for training jobs To monitor deployed endpoints for data drift and raise CloudWatch alarms To record the parameters, metrics and artefacts of each training run so that runs can be compared

7. In which situation is SageMaker Autopilot the most appropriate choice? *

When you need a custom loss function that reflects a specific business cost When the problem depends on deep, domain-specific feature engineering When you need a quick, reliable baseline model with minimal manual effort When you need full manual control over every step of the training pipeline

8. When building a custom model with scikit-learn, which method is used to train the model on your prepared training data? *

predict() transform() fit() score()

9. In the SageMaker Model Registry, what does an approval status of “PendingManualApproval” indicate? *

The training job has not yet finished running Autopilot is still evaluating its candidate models The endpoint has been provisioned but is not yet receiving traffic The model version is registered but has not yet been signed off for deployment

10. What is the difference between data drift and concept drift? *

Data drift is a change in the distribution of the input features; concept drift is a change in the relationship between the features and the target Data drift affects batch inference only; concept drift affects real-time endpoints only Data drift is caused by infrastructure failure; concept drift is caused by code changes They are two names for the same phenomenon