Databricks Certified Machine Learning Professional Practice Test 9
Question 80
Review this production model incident: A feature refresh writes duplicate records for each entity and logical timestamp after retries, breaking unique lookups. Which failure risk follows most closely from the unresolved behavior?
- Invalid numeric features produce serving errors or hidden downstream score distortions.
- An ancient measurement can influence a current score as if it were fresh evidence.
- A repeated refresh doubles feature rows and multiplies downstream training joins.
- Scores after failover revert to incomplete or uninitialized feature state.
Explanation
Correct Answer
C:This consequence follows from the specific unresolved condition. A repeated refresh doubles feature rows and multiplies downstream training joins. The causal constraint is: Idempotent feature pipeline writes need stable entity/time keys and safe retry semantics.
Incorrect Answers
A:That direct consequence is directed at a different advanced feature store concepts situation: A payment classifier's on-demand feature division returns infinity for zero denominators and fails serialization in the serving endpoint. The more immediate risk here is a repeated refresh doubles feature rows and multiplies downstream training joins.
B:That direct consequence is directed at a different advanced feature store concepts situation: A sensor classifier silently reuses features from a year-old device after months without readings, despite a maximum allowed lookback of seven days. The more immediate risk here is a repeated refresh doubles feature rows and multiplies downstream training joins.
D:That direct consequence is directed at a different advanced feature store concepts situation: A business deploys a model whose features are computed from a stream in memory only and disappear when its compute cluster restarts. The more immediate risk here is a repeated refresh doubles feature rows and multiplies downstream training joins.
Question 81
A production ML review contains this situation: A low-latency feature table uses external customer names as lookup keys; names change and are not unique. What is the most material immediate risk in this situation?
- Training consumes stale or differently computed features depending on who ran the notebook.
- Invalid numeric features produce serving errors or hidden downstream score distortions.
- One purchase increases the spending feature twice, affecting downstream risk scores.
- Scores pull the wrong customer's features or fail after name changes.
Explanation
Correct Answer
D:This consequence follows from the specific unresolved condition. Scores pull the wrong customer's features or fail after name changes. The causal constraint is: Online feature lookup keys must be stable and uniquely identify the intended entity.
Incorrect Answers
A:That direct consequence is directed at a different advanced feature store concepts situation: A team manually reruns a feature notebook each week before training; feature definitions change without an execution history or stable destination. The more immediate risk here is scores pull the wrong customer's features or fail after name changes.
B:That direct consequence is directed at a different advanced feature store concepts situation: A payment classifier's on-demand feature division returns infinity for zero denominators and fails serialization in the serving endpoint. The more immediate risk here is scores pull the wrong customer's features or fail after name changes.
C:That direct consequence is directed at a different advanced feature store concepts situation: A real-time feature job updates user purchase totals without checking whether upstream transaction events are delivered twice. The more immediate risk here is scores pull the wrong customer's features or fail after name changes.
Question 82
A production ML review contains this situation: A fraud model predicts instantly but labels arrive 21 days later; an inference dashboard treats unlabelled transactions as negative examples. Which check shows that this particular problem has been resolved?
- Weekend fixtures match appropriate seasonality while atypical weekend changes still alert.
- A safe synthetic threshold crossing reaches the correct team with the monitored metric and window.
- A deliberate new-customer regression is detected with sufficient labeled sample support.
- Metrics change only when genuine label observations are joined to the correct predictions.
Explanation
Correct Answer
D:This verification measures the required result rather than just process liveness. Metrics change only when genuine label observations are joined to the correct predictions. It confirms whether the targeted action succeeded: evaluate performance only on sufficiently matured matched predictions and report pending labels separately.
Incorrect Answers
A:That acceptance test is directed at a different drift detection and lakehouse monitoring situation: An online feature monitor compares weekend shopping to weekday reference data and treats expected weekend behavior as a model failure. Here the actual success criterion is metrics change only when genuine label observations are joined to the correct predictions.
B:That acceptance test is directed at a different drift detection and lakehouse monitoring situation: A fraud-feature drift monitor crosses its threshold but the alert has no assigned recipient and no incident ownership. Here the actual success criterion is metrics change only when genuine label observations are joined to the correct predictions.
C:That acceptance test is directed at a different drift detection and lakehouse monitoring situation: An account-risk model's fraud recall is stable overall but unexpectedly falls for new customers in their first 30 days. Here the actual success criterion is metrics change only when genuine label observations are joined to the correct predictions.
Question 83
In an enterprise ML workflow, the reported issue is: A logistics model must score 60 million daily routes already in Delta but the scoring script collects each route into Python before calling predict. Which change addresses the real cause without undermining other requirements?
- Use a compatible streaming scoring pipeline with durable progress and bounded output writes.
- Load the fitted PipelineModel and transform the full batch DataFrame through supported Spark execution.
- Hold out a final test set before tuning and score the selected pipeline exactly once after selection.
- Add a justified tail-sensitive error measure such as RMSE and inspect failure slices in model selection.
Explanation
Correct Answer
B:This corrects the relevant failure boundary instead of merely hiding the symptom. Load the fitted PipelineModel and transform the full batch DataFrame through supported Spark execution. The issue matters because driver memory exhaustion halts scoring and leaves most routes unprocessed.
Incorrect Answers
A:That corrective action is directed at a different using spark ml situation: An energy plant needs one model prediction for every new sensor micro-batch but proposes a synchronous REST call per row from a driver collect loop. The relevant corrective action here is load the fitted PipelineModel and transform the full batch DataFrame through supported Spark execution.
C:That corrective action is directed at a different using spark ml situation: After choosing the best parameter map, a data scientist reports the same cross-validation score as an independent final test result without any untouched data. The relevant corrective action here is load the fitted PipelineModel and transform the full batch DataFrame through supported Spark execution.
D:That corrective action is directed at a different using spark ml situation: A forecasting model has a few catastrophic numeric errors hidden by a low median absolute error and operations care about large misses. The relevant corrective action here is load the fitted PipelineModel and transform the full batch DataFrame through supported Spark execution.
Question 84
In an enterprise ML workflow, the reported issue is: A bundle references a model registry name without qualifying its environment, and a development deployment accidentally updates production registered-model aliases. What technical distinction is essential to this investigation?
- Bundle target parameters must resolve to explicitly authorized environment-specific resources.
- Scalable environment design must account for data locality, residency and endpoint boundaries.
- Multi-team ML environments need enforceable resource policies and observability to prevent noisy-neighbor failure.
- Environment architecture should isolate access, data destinations and ownership for different stages.
Explanation
Correct Answer
A:The technical constraint is specific to the reported failure. Bundle target parameters must resolve to explicitly authorized environment-specific resources. To test it, inspect resolved resource names and deployment plan output for each bundle target.
Incorrect Answers
B:That underlying principle is directed at a different environment architectures situation: A multilingual prediction service must handle regional data residency, but the team copies a shared feature cache across regions without governance. The current incident instead turns on bundle target parameters must resolve to explicitly authorized environment-specific resources.
C:That underlying principle is directed at a different environment architectures situation: A production deployment scales to 30 teams but all jobs share one compute pool with no capacity limits or cost attribution. The current incident instead turns on bundle target parameters must resolve to explicitly authorized environment-specific resources.
D:That underlying principle is directed at a different environment architectures situation: A company operates development and production models in one workspace under the same broad service account and frequently overwrites production tables during experiments. The current incident instead turns on bundle target parameters must resolve to explicitly authorized environment-specific resources.
Question 85
Consider the following fictional Databricks production case: An alert from a feature monitor automatically replaces the live model with a newly trained artifact without testing or approval. Which implementation is most appropriate for the stated incident?
- Require a meaningful quality improvement supported by adequate evidence before changing the champion.
- Evaluate all eligible candidates against one controlled frozen benchmark and rank only comparable results.
- Rank only eligible candidates that pass accuracy and operational SLA gates before promotion.
- Run candidate training after the trigger but require explicit quality/security gates before any live promotion.
Explanation
Correct Answer
D:This corrects the relevant failure boundary instead of merely hiding the symptom. Run candidate training after the trigger but require explicit quality/security gates before any live promotion. The issue matters because a poorer or unsafe model reaches customers just because monitoring fired.
Incorrect Answers
A:That corrective action is directed at a different automated retraining situation: A new model beats the champion by 0.1 percentage points on a tiny evaluation set, and the team automatically promotes it despite large uncertainty. The relevant corrective action here is run candidate training after the trigger but require explicit quality/security gates before any live promotion.
B:That corrective action is directed at a different automated retraining situation: A retraining pipeline ranks candidate models using metrics calculated on different dataset versions, then selects the numerically highest result. The relevant corrective action here is run candidate training after the trigger but require explicit quality/security gates before any live promotion.
C:That corrective action is directed at a different automated retraining situation: An AutoML training job always chooses the model with the lowest training loss, even when the deployed service has a strict memory and P95 latency budget. The relevant corrective action here is run candidate training after the trigger but require explicit quality/security gates before any live promotion.
Question 86
Consider the following fictional Databricks production case: A pricing endpoint queries a large offline Delta feature table for every user request, and P95 response time reaches several seconds. What consequence most specifically follows from leaving the issue unchanged?
- Restarted jobs double count behavior or lose events, corrupting online feature values.
- A repeated refresh doubles feature rows and multiplies downstream training joins.
- A backtest gains information that production inference could not have used.
- Live predictions violate the latency budget even though offline training is fast.
Explanation
Correct Answer
D:This consequence follows from the specific unresolved condition. Live predictions violate the latency budget even though offline training is fast. The causal constraint is: Low-latency serving requires online feature access with a documented freshness and lookup contract.
Incorrect Answers
A:That direct consequence is directed at a different advanced feature store concepts situation: An online behavioral feature counter consumes a streaming source but does not retain checkpoint state after worker restarts. The more immediate risk here is live predictions violate the latency budget even though offline training is fast.
B:That direct consequence is directed at a different advanced feature store concepts situation: A feature refresh writes duplicate records for each entity and logical timestamp after retries, breaking unique lookups. The more immediate risk here is live predictions violate the latency budget even though offline training is fast.
C:That direct consequence is directed at a different advanced feature store concepts situation: A retailer's features arrive two hours late, and offline training uses event time alone even though the online system could not have seen those late arrivals at prediction time. The more immediate risk here is live predictions violate the latency budget even though offline training is fast.
Question 87
Consider the following fictional Databricks production case: An online pricing service requires a personalized response within 70 milliseconds per request but its team schedules a full overnight batch job as the only inference route. What underlying principle determines the most appropriate interpretation?
- VectorAssembler is a Transformer that creates a vector column, not a trainable predictor.
- RMSE squares residuals and highlights large errors, while median error alone may obscure expensive outliers.
- Per-request low-latency inference calls for an appropriately sized online serving path, not batch-only scheduling.
- Distributed batch inference can apply a saved SparkML PipelineModel directly to a DataFrame without collecting every row.
Explanation
Correct Answer
C:The technical constraint is specific to the reported failure. Per-request low-latency inference calls for an appropriately sized online serving path, not batch-only scheduling. To test it, compare request latency and concurrency service-level objectives with the batch schedule and observed turnaround.
Incorrect Answers
A:That underlying principle is directed at a different using spark ml situation: A team wants to convert several numeric inputs into one features vector but treats VectorAssembler as a clustering Estimator that should output predictions. The current incident instead turns on per-request low-latency inference calls for an appropriately sized online serving path, not batch-only scheduling.
B:That underlying principle is directed at a different using spark ml situation: A forecasting model has a few catastrophic numeric errors hidden by a low median absolute error and operations care about large misses. The current incident instead turns on per-request low-latency inference calls for an appropriately sized online serving path, not batch-only scheduling.
D:That underlying principle is directed at a different using spark ml situation: A logistics model must score 60 million daily routes already in Delta but the scoring script collects each route into Python before calling predict. The current incident instead turns on per-request low-latency inference calls for an appropriately sized online serving path, not batch-only scheduling.
Question 88
In an enterprise ML workflow, the reported issue is: A binary classifier ranks risky cases well but sends too many alerts at a default 0.5 decision threshold. How should the engineer validate the correction?
- A stored raw fixture produces matching transformed feature vectors and predictions in training and deployment.
- The selected candidate has the minimum validated RMSE among comparable parameter configurations.
- A held-out review confirms alert volume and true-positive capture at the frozen threshold.
- A controlled example maps known training categories to stable numeric indices using the fitted model.
Explanation
Correct Answer
C:This verification measures the required result rather than just process liveness. A held-out review confirms alert volume and true-positive capture at the frozen threshold. It confirms whether the targeted action succeeded: calibrate or tune the decision threshold using validation data and the stated response budget.
Incorrect Answers
A:That acceptance test is directed at a different using spark ml situation: A churn experiment separately scripts feature scaling in training but omits that scaling when the deployed prediction job loads the classifier. Here the actual success criterion is a held-out review confirms alert volume and true-positive capture at the frozen threshold.
B:That acceptance test is directed at a different using spark ml situation: A SparkML grid-search experiment selects the model with the largest error metric even though the evaluator reports RMSE to minimize. Here the actual success criterion is a held-out review confirms alert volume and true-positive capture at the frozen threshold.
D:That acceptance test is directed at a different using spark ml situation: A new Spark developer calls transform on an unfitted StringIndexer instance and expects learned category labels to appear automatically. Here the actual success criterion is a held-out review confirms alert volume and true-positive capture at the frozen threshold.
Question 89
The engineering team encounters the following issue: A production deployment scales to 30 teams but all jobs share one compute pool with no capacity limits or cost attribution. What investigation would most effectively isolate this behavior?
- Examine bundle validation results, actual resource permissions and downstream job execution traces.
- Review per-team compute usage, job queuing and capacity constraints during overlapping workloads.
- Inspect resolved resource names and deployment plan output for each bundle target.
- Compare current official supported bundle resource fields with the pinned CLI version and existing bundle schema.
Explanation
Correct Answer
B:This is the most discriminating check for the observed symptoms. Review per-team compute usage, job queuing and capacity constraints during overlapping workloads. The underlying technical issue is that multi-team ML environments need enforceable resource policies and observability to prevent noisy-neighbor failure.
Incorrect Answers
A:That diagnostic evidence is directed at a different environment architectures situation: A continuous-delivery job deploys a notebook change through a bundle but skips validation of referenced experiment paths and job permissions. For the stated incident, the decisive observation is review per-team compute usage, job queuing and capacity constraints during overlapping workloads.
C:That diagnostic evidence is directed at a different environment architectures situation: A bundle references a model registry name without qualifying its environment, and a development deployment accidentally updates production registered-model aliases. For the stated incident, the decisive observation is review per-team compute usage, job queuing and capacity constraints during overlapping workloads.
D:That diagnostic evidence is directed at a different environment architectures situation: An ML platform moves its configuration from older Databricks Asset Bundles terminology to newer Declarative Automation Bundles references and assumes every resource schema stayed identical. For the stated incident, the decisive observation is review per-team compute usage, job queuing and capacity constraints during overlapping workloads.
Question 90
Consider the following fictional Databricks production case: A pipeline engineer fits a vectorizer after an estimator in the stage list and then tries to use the estimator's prediction as the vectorizer input. Which implementation is most appropriate for the stated incident?
- Retain the correctly validated single-node training path unless a measured scale requirement justifies SparkML.
- Configure selection against the intended minimization metric and rerun model comparison.
- Deploy the fitted preprocessing and estimator together as one versioned pipeline artifact.
- Topologically order vectorizers, assemblers and estimators so required feature columns exist before model fitting.
Explanation
Correct Answer
D:This corrects the relevant failure boundary instead of merely hiding the symptom. Topologically order vectorizers, assemblers and estimators so required feature columns exist before model fitting. The issue matters because pipeline fitting fails with missing-column or mismatched-schema errors.
Incorrect Answers
A:That corrective action is directed at a different using spark ml situation: A laboratory runs a linear regression every quarter on only 280 clean observations and proposes a 48-worker Spark cluster to reduce a two-second local fit. The relevant corrective action here is topologically order vectorizers, assemblers and estimators so required feature columns exist before model fitting.
B:That corrective action is directed at a different using spark ml situation: A SparkML grid-search experiment selects the model with the largest error metric even though the evaluator reports RMSE to minimize. The relevant corrective action here is topologically order vectorizers, assemblers and estimators so required feature columns exist before model fitting.
C:That corrective action is directed at a different using spark ml situation: A batch scoring service loads a classifier version from last month while using today's new feature indexer, silently changing categorical positions. The relevant corrective action here is topologically order vectorizers, assemblers and estimators so required feature columns exist before model fitting.