Oracle Cloud Infrastructure Data Science Professional Exam Practice Test
The Oracle Cloud Infrastructure Data Science Professional Exam focuses on developing and operating machine-learning solutions in OCI. Practice topics include preparing data without leakage, authenticating notebooks to cloud services, comparing model configurations, and evaluating models against the data conditions they will face in production.
Exam overview
About the Oracle Cloud Infrastructure Data Science Professional Exam
Data science work involves more than choosing an estimator: practitioners must build transformations that learn only from training data, select validation methods that support fair model comparisons, and choose metrics suited to the target distribution and cost of errors. Other relevant tasks include representing free-form text for conventional classifiers, measuring feature importance through score changes, and checking whether predictions vary across sensitive groups. OCI workflows also call for using resource-assigned identity rather than embedding long-lived credentials and maintaining consistent code and package environments. After deployment, monitoring for changes in incoming data and responding with investigation or retraining helps address declining performance. These concepts connect model development decisions to secure, reproducible, and responsible operation in the cloud.
Sample Practice Questions
Review a selection from this exam's published US Practice Tests practice bank.
A training pipeline applies a transformation that estimates parameters from data, such as imputing missing values with column means. To prevent validation information from influencing training, what should the pipeline do?
Show answer
Correct answer: AFitting preprocessing steps on training data only prevents information from the validation split leaking into the training process. The fitted transformation is then applied unchanged to validation data.
A notebook needs to call an OCI service using the identity and permissions assigned to its OCI Data Science resource. Which authentication approach avoids placing a long-lived secret key in notebook code?
Show answer
Correct answer: AWhere supported and properly authorized, resource principals or instance principals let OCI resources authenticate as an assigned identity. This avoids embedding long-lived user credentials in source code; the identity still needs appropriate policies.
A data scientist wants to compare several model configurations while reducing dependence on one train-validation split. Which approach is most appropriate?
Show answer
Correct answer: ACross-validation evaluates configurations across multiple folds of the training data, providing a more stable basis for model selection. The final test set should be reserved for an unbiased final evaluation.
A trained model's performance declines after it has been deployed, while the incoming data distribution has changed. What is a useful operational response?
Show answer
Correct answer: AProduction monitoring can reveal data or performance drift. Investigating the cause and retraining or revising the system when supported by evidence is more reliable than assuming performance remains constant.
A team needs to keep notebook code and its required Python packages consistent across runs. Which practice best supports reproducibility?
Show answer
Correct answer: APinning or otherwise recording dependencies and using a controlled environment makes it easier to reproduce notebook behavior. Installing unpinned latest versions can introduce changes between runs.
A team is training a model with a target that has a long-tailed numeric distribution. Which metric is generally less sensitive to a small number of very large prediction errors than mean squared error?
Show answer
Correct answer: AMean absolute error averages absolute residuals, so errors are not squared and extreme residuals have less influence than they do under mean squared error. Metric choice should still reflect the application's objectives and error costs.
A team wants to compare the relative importance of input features in a fitted model without relying on a model-specific coefficient scale. Which approach measures the effect of disrupting each feature's values on a chosen evaluation score?
Show answer
Correct answer: APermutation importance evaluates a model after shuffling one feature and measures the resulting score change. A larger degradation suggests the model relied more on that feature, though correlated features can complicate interpretation.
How to use this practice test
The free US Practice Tests bank contains 50 questions; each practice session presents up to 20 random questions and shows results afterward. Use a session to identify gaps in areas such as validation design, metrics, OCI notebook authentication, reproducibility, and production monitoring, then revisit missed concepts before another attempt.
Preparation tips
- For preprocessing, fit imputers, encoders, and other learned transformations on training folds only; apply the fitted transformations to validation data without recalculating parameters.
- Compare cross-validation with a single holdout split, and consider how class proportions in evaluation data affect estimates of production performance.
- Match regression metrics to error behavior: compare squared-error measures with absolute-error measures when targets contain extreme values.
- Practice explaining OCI resource principals as an alternative to placing long-lived API keys in notebook code, and identify how environment specifications improve repeatability.
- For deployed models, distinguish data drift from a change in model quality; consider monitoring input distributions, evaluating subgroup performance, and deciding when investigation or retraining is warranted.