Oracle Cloud Infrastructure Data Science Professional (1Z0-1110-26) - Practice Exam 1

1. A business application needs low-latency online predictions from a cataloged model. Which OCI resource is intended for this?
2. A categorical encoder fails when production receives an unseen product category. Which design should be tested?
3. A classifier predicts the majority class for 99% of examples and reports 99% accuracy on a dataset with 1% fraud. Which evaluation is most useful next?
4. A deep-learning job needs GPUs, but provisioning is rejected before code starts. Which operational prerequisite should be checked?
5. A deployment's feature distributions change significantly after a new upstream data release. What should happen before automatically retraining and promoting?
6. A deployment's p95 latency rises while each replica is highly utilized. Which change should be evaluated with load testing?
7. A distance-based classifier is dominated by a feature measured in millions while other features range from zero to one. Which preparation step should be evaluated?
8. A feature contains the date a loan was finally written off, but predictions are made at loan origination. What should happen?
9. A forecasting model is evaluated with random folds that mix later records into training for earlier predictions. What alternative should be evaluated?
10. A fraud project scales all rows before splitting training and test data. Why should the pipeline be changed?
11. A job run is interrupted after saving an expensive intermediate result. What design can reduce wasted work on retry?
12. A model deployment must reach a private database under customer network governance. Which setting requires review?
13. A model must be recalled after a data defect is found. Which metadata most directly helps identify affected artifacts?
14. A model performs poorly after timestamps are parsed in a different timezone during inference. What should be standardized?
15. A new model has passed offline tests, but its serving artifact has never handled production-shaped requests. What is necessary before full rollout?
16. A nightly model retraining task should run independently of an interactive notebook. Which OCI Data Science resource fits?
17. A nominal category has no meaningful order. Which encoding is generally safer than assigning arbitrary increasing integers to a linear model?
18. A notebook must read a private Object Storage dataset without embedding a user's API key in code. Which approach should be evaluated?
19. A pipeline evaluation step fails its quality threshold. What should the promotion logic do?
20. A probability model has good ranking but overconfident probabilities. Which assessment should be added?
21. A regression model must be especially sensitive to large prediction errors. Which metric should be evaluated alongside business requirements?
22. A repeatable ML workflow must prepare features, train, evaluate, and register only acceptable models. Which resource should orchestrate it?
23. A retraining pipeline may resume after partial failure. Which property is important for registration and output writes?
24. A reviewer must reproduce a registered model's result. Which records should be preserved? Select TWO.
25. A scientist deactivates a notebook session and later reactivates it. Where should important working files be placed for supported persistence?
26. A security policy requires only approved signed serving containers. Which capability should be evaluated?
27. A small positive class is absent from some ordinary random validation folds. Which split may improve class representation when temporal and grouping constraints permit?
28. A standard managed model deployment loads an artifact but cannot find its inference implementation. Which file should be reviewed?
29. A standard model artifact requires a published conda environment for serving. Where is that inference environment documented for deployment?
30. A team deactivates unused notebook sessions to save cost. Which charge can remain for retained sessions?
31. A team wants Oracle's Python tools for common lifecycle tasks such as training, evaluation, explanations, and model-catalog integration. Which library fits?
32. A training job runs correctly locally but fails because a library version differs in OCI. What should be recorded and deployed?
33. A training workload needs read access to one bucket and write access to a model compartment. What is the preferred permission scope?
34. A workload's Object Storage call fails after a resource-principal configuration change. Which pair should be examined? Select TWO.
35. An endpoint's error rate rises after an artifact update. Which records are most useful for diagnosis?
36. An engineer may create notebooks but cannot create model deployments. What should an administrator review?
37. An explanation tool identifies features contributing to a particular prediction. Which claim should be avoided?
38. An online prediction request is denied although the endpoint is active. What should be checked first?
39. Both training and validation error are high and similar. Which hypothesis should be investigated?
40. Hyperparameters are selected using the same test set repeatedly. Which revised process is appropriate?
41. Missing a positive case is much more costly than reviewing a false alarm. Which operating-point decision should be evaluated?
42. Missing income values require imputation. Which process avoids evaluation leakage?
43. Of 100 flagged transactions, 80 are truly fraudulent. What is the precision of the flagged set?
44. Overall performance is acceptable, but a small customer segment has poor recall. What should be reported?
45. Permuting one feature barely changes validation performance, but that feature is strongly correlated with another predictor. What should be concluded?
46. Records from the same customer appear in both training and validation sets, inflating performance on customer-specific signals. What split should be evaluated?
47. Sensitive training rows accidentally appear in debug logs. What correction is appropriate?
48. The model loads successfully, but predictions fail on valid inputs because preprocessing was left in the training notebook. What should be packaged?
49. Training error is low, but validation error remains high as model complexity grows. Which direction should be evaluated?
50. Two experiments use different validation partitions, and one scores slightly higher. What is the best comparison approach?

Comments — 0