Databricks Certified Machine Learning Professional Practice Test 85
Question 840
Consider the following fictional Databricks production case: A credit-risk pipeline fits StringIndexer on the entire dataset before splitting train and holdout, and rare holdout labels influence its learned category mapping. What underlying principle determines the most appropriate interpretation?
- Training-set fit does not establish out-of-sample predictive quality or cold-start behavior.
- Scoring should use preprocessing components trained and versioned with the exact estimator artifact.
- Hyperparameter selection uses validation evidence; the final unbiased estimate requires an independent untouched test split.
- Any fitted preprocessing stage can learn from holdout records and must be fitted within the training boundary.
Explanation
Correct Answer
D:The technical constraint is specific to the reported failure. Any fitted preprocessing stage can learn from holdout records and must be fitted within the training boundary. To test it, trace when the indexer fit was called relative to the train/validation split and compare observed label inventories.
Incorrect Answers
A:That underlying principle is directed at a different using spark ml situation: A recommender experiment computes all evaluation metrics on training interactions and reports the result as generalization to new customers. The current incident instead turns on any fitted preprocessing stage can learn from holdout records and must be fitted within the training boundary.
B:That underlying principle 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 current incident instead turns on any fitted preprocessing stage can learn from holdout records and must be fitted within the training boundary.
C:That underlying principle 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 current incident instead turns on any fitted preprocessing stage can learn from holdout records and must be fitted within the training boundary.
Question 841
A production ML review contains this situation: An insurer already stores its risk features as a large Spark DataFrame but exports them into Python lists before every vector transformation and repeatedly hits out-of-memory errors. What production-safe change best matches the evidence?
- Replace unnecessary local conversions with supported distributed SparkML transforms and retain an end-to-end Spark pipeline.
- 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
A:This corrects the relevant failure boundary instead of merely hiding the symptom. Replace unnecessary local conversions with supported distributed SparkML transforms and retain an end-to-end Spark pipeline. The issue matters because the job can fail during preprocessing before its distributed estimator receives any input.
Incorrect Answers
B:That corrective action 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 relevant corrective action here is replace unnecessary local conversions with supported distributed SparkML transforms and retain an end-to-end Spark pipeline.
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 replace unnecessary local conversions with supported distributed SparkML transforms and retain an end-to-end Spark pipeline.
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 replace unnecessary local conversions with supported distributed SparkML transforms and retain an end-to-end Spark pipeline.
Question 842
In an enterprise ML workflow, the reported issue is: A team logs a score labeled accuracy with values between 0 and 100 in some runs and between 0 and 1 in others, then ranks runs automatically. How should the engineer validate the correction?
- All candidate scores use the same documented calculation and scale in a fresh comparison.
- An experiment query retrieves one parent with all expected child trials and their distinct metrics.
- A model registry record references the exact child run and parameters used to create the artifact.
- A clean serving image loads the artifact and returns the same output for a saved fixture.
Explanation
Correct Answer
A:This verification measures the required result rather than just process liveness. All candidate scores use the same documented calculation and scale in a fresh comparison. It confirms whether the targeted action succeeded: standardize metric naming and units, version the computation and recalculate a comparable ranking.
Incorrect Answers
B:That acceptance test is directed at a different advanced mlflow usage situation: An experiment tunes 40 candidates but logs each trial as an unrelated root MLflow run, so reviewers cannot reconstruct which search produced a winner. Here the actual success criterion is all candidate scores use the same documented calculation and scale in a fresh comparison.
C:That acceptance test is directed at a different advanced mlflow usage situation: A data scientist selects the lowest validation loss from nested runs but reports only the parent run ID as the model-producing run. Here the actual success criterion is all candidate scores use the same documented calculation and scale in a fresh comparison.
D:That acceptance test is directed at a different advanced mlflow usage situation: A team logs a custom Python model that imports a package installed only in the author's notebook, and the serving endpoint fails when loading the artifact. Here the actual success criterion is all candidate scores use the same documented calculation and scale in a fresh comparison.
Question 843
Review this production model incident: A fraud model predicts instantly but labels arrive 21 days later; an inference dashboard treats unlabelled transactions as negative examples. Which mitigation resolves the defect rather than concealing it?
- Normalize timestamp semantics before applying approved local-hour performance slices.
- Evaluate performance only on sufficiently matured matched predictions and report pending labels separately.
- Maintain a model-version-linked training baseline alongside rolling comparisons.
- Log permitted request/prediction metadata to a governed inference table with retention and redaction controls.
Explanation
Correct Answer
B:This corrects the relevant failure boundary instead of merely hiding the symptom. Evaluate performance only on sufficiently matured matched predictions and report pending labels separately. The issue matters because false optimism or pessimism distorts the model's apparent accuracy and promotion decisions.
Incorrect Answers
A:That corrective action is directed at a different drift detection and lakehouse monitoring situation: A model's prediction error appears to spike at 2 a.m. local time but the monitoring table groups timestamps in UTC for a globally distributed region. The relevant corrective action here is evaluate performance only on sufficiently matured matched predictions and report pending labels separately.
C:That corrective action is directed at a different drift detection and lakehouse monitoring situation: A model trained on last year's customer mix is monitored only against yesterday's already shifted cohort, so its original training-population mismatch is hidden. The relevant corrective action here is evaluate performance only on sufficiently matured matched predictions and report pending labels separately.
D:That corrective action is directed at a different drift detection and lakehouse monitoring situation: A serving endpoint appears healthy but nobody records model version, request identifier or prediction value, so a drift incident cannot be reconstructed. The relevant corrective action here is evaluate performance only on sufficiently matured matched predictions and report pending labels separately.
Question 844
In an enterprise ML workflow, the reported issue is: A language model's weights cannot fit on one accelerator yet a team launches pure data-parallel replicas, each loading the entire model. What likely operational or model-quality harm does this issue create?
- The winning configuration cannot be reproduced or independently audited.
- Repeated cross-engine transfers dominate tuning time and may exhaust network capacity.
- Available accelerators sit idle while the tuning window expires.
- Each replica fails device memory allocation regardless of the number of added replicas.
Explanation
Correct Answer
D:This consequence follows from the specific unresolved condition. Each replica fails device memory allocation regardless of the number of added replicas. The causal constraint is: Data parallelism duplicates model parameters, while model parallelism or sharding distributes parameters when required.
Incorrect Answers
A:That direct consequence is directed at a different scaling and tuning situation: An Optuna search runs 80 hyperparameter trials but records only the last winning score in MLflow, making comparisons impossible. The more immediate risk here is each replica fails device memory allocation regardless of the number of added replicas.
B:That direct consequence is directed at a different scaling and tuning situation: A hyperparameter search operates on large Spark-managed feature tables but each candidate model is small and independent, and workers repeatedly convert the entire table for each Ray task. The more immediate risk here is each replica fails device memory allocation regardless of the number of added replicas.
C:That direct consequence is directed at a different scaling and tuning situation: A team runs independent GPU-heavy neural trials sequentially, despite having several isolated GPU workers that can fit each model. The more immediate risk here is each replica fails device memory allocation regardless of the number of added replicas.
Question 845
Review this production model incident: A new recommendation model consumes three times as much memory and the team plans a canary based solely on prediction accuracy, ignoring serving capacity. What is the central engineering constraint behind this behavior?
- An actionable rollback plan needs a retained deployable champion artifact and a tested restoration procedure.
- Rollout safety needs operational latency and memory gates alongside predictive quality metrics.
- Model Serving configuration acceptance is not equivalent to endpoint/model readiness for user traffic.
- Model Serving deployments must validate backward compatibility of request schema across existing clients.
Explanation
Correct Answer
B:The technical constraint is specific to the reported failure. Rollout safety needs operational latency and memory gates alongside predictive quality metrics. To test it, compare replica memory, P95 latency and model quality for both versions under realistic canary load.
Incorrect Answers
A:That underlying principle is directed at a different deployment strategies situation: An ML operator promotes a new model version but its automated rollback points to a model that was deleted from the endpoint configuration. The current incident instead turns on rollout safety needs operational latency and memory gates alongside predictive quality metrics.
C:That underlying principle is directed at a different deployment strategies situation: An endpoint update request succeeds but the newly configured model version is still loading; monitoring marks the release complete immediately. The current incident instead turns on rollout safety needs operational latency and memory gates alongside predictive quality metrics.
D:That underlying principle is directed at a different deployment strategies situation: A rollout updates the serving model but a renamed input column breaks requests from an older mobile client. The current incident instead turns on rollout safety needs operational latency and memory gates alongside predictive quality metrics.
Question 846
A production ML review contains this situation: A team updates a shared on-demand feature function without version control and unexpectedly changes both batch retraining and online serving behavior. What is the most material immediate risk in this situation?
- Live predictions violate the latency budget even though offline training is fast.
- Scores after failover revert to incomplete or uninitialized feature state.
- Training consumes stale or differently computed features depending on who ran the notebook.
- Several models change their effective input distributions without deliberate retraining or review.
Explanation
Correct Answer
D:This consequence follows from the specific unresolved condition. Several models change their effective input distributions without deliberate retraining or review. The causal constraint is: Shared feature logic changes should be versioned and regression-tested across all consuming models.
Incorrect Answers
A:That direct consequence is directed at a different advanced feature store concepts situation: A pricing endpoint queries a large offline Delta feature table for every user request, and P95 response time reaches several seconds. The more immediate risk here is several models change their effective input distributions without deliberate retraining or review.
B: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 several models change their effective input distributions without deliberate retraining or review.
C: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 several models change their effective input distributions without deliberate retraining or review.
Question 847
Consider the following fictional Databricks production case: An operations model uses a SparkML estimator that itself distributes data, yet the team launches dozens of concurrent full-cluster fits on the same fixed small cluster. What underlying principle determines the most appropriate interpretation?
- Reliable tuning provenance requires parameter, metric and trial identity logging for every evaluated trial.
- Data parallelism is often more natural than model-parallel networking for a small model with large independent data.
- Horizontal replica count alone does not combine GPU memory for one model without a supported parallelization strategy.
- Parallel outer trials of already distributed estimators can contend for the same workers and increase overall elapsed time.
Explanation
Correct Answer
D:The technical constraint is specific to the reported failure. Parallel outer trials of already distributed estimators can contend for the same workers and increase overall elapsed time. To test it, compare executor saturation, shuffle backlog and trial overlap for individual versus concurrent fits.
Incorrect Answers
A:That underlying principle is directed at a different scaling and tuning situation: An Optuna search runs 80 hyperparameter trials but records only the last winning score in MLflow, making comparisons impossible. The current incident instead turns on parallel outer trials of already distributed estimators can contend for the same workers and increase overall elapsed time.
B:That underlying principle is directed at a different scaling and tuning situation: A tabular classifier fits comfortably on one GPU and 100 million independent examples are available, but a team proposes splitting the small model's layers across hosts. The current incident instead turns on parallel outer trials of already distributed estimators can contend for the same workers and increase overall elapsed time.
C:That underlying principle is directed at a different scaling and tuning situation: An image model needs 90 GB accelerator memory for one training replica while each available GPU has 24 GB and engineers propose only adding more identical nodes. The current incident instead turns on parallel outer trials of already distributed estimators can contend for the same workers and increase overall elapsed time.
Question 848
The engineering team encounters the following issue: An inference profile groups two active model versions into one aggregate and hides a regression introduced by the new release. What corrective action should the engineer prefer?
- Use an appropriate categorical drift metric with handling for new labels and stable population definition.
- Configure a supported time-series profile with the appropriate timestamp and daily granularity.
- Separate versioned baseline storage from current data and block unintended updates.
- Configure model-identity-aware inference monitoring with version-specific comparisons.
Explanation
Correct Answer
D:This corrects the relevant failure boundary instead of merely hiding the symptom. Configure model-identity-aware inference monitoring with version-specific comparisons. The issue matters because a deteriorating new model is concealed by still-good traffic on the old model.
Incorrect Answers
A:That corrective action is directed at a different drift detection and lakehouse monitoring situation: A customer product-category feature acquires many new category values and analysts try to detect the change using only the overall row count. The relevant corrective action here is configure model-identity-aware inference monitoring with version-specific comparisons.
B:That corrective action is directed at a different drift detection and lakehouse monitoring situation: A fleet sensor table records measurement timestamps and analysts need daily drift trends; a snapshot-only profile erases the time dimension. The relevant corrective action here is configure model-identity-aware inference monitoring with version-specific comparisons.
C:That corrective action is directed at a different drift detection and lakehouse monitoring situation: A baseline table is overwritten every night with the current day's data before a drift calculation, making the comparison always look unchanged. The relevant corrective action here is configure model-identity-aware inference monitoring with version-specific comparisons.
Question 849
The engineering team encounters the following issue: An online feature monitor compares weekend shopping to weekday reference data and treats expected weekend behavior as a model failure. Which experiment most directly demonstrates a successful fix?
- A deteriorating new-user fixture raises the slice metric without falsely altering established-user results.
- Weekend fixtures match appropriate seasonality while atypical weekend changes still alert.
- Two fixtures with equal risk proportions but different volumes produce the same custom rate.
- A safe synthetic threshold crossing reaches the correct team with the monitored metric and window.
Explanation
Correct Answer
B:This verification measures the required result rather than just process liveness. Weekend fixtures match appropriate seasonality while atypical weekend changes still alert. It confirms whether the targeted action succeeded: compare to a justified weekend reference or adjust baseline design with documented change control.
Incorrect Answers
A:That acceptance test is directed at a different drift detection and lakehouse monitoring situation: A recommender reports stable overall accuracy while click-through prediction quality for first-time users deteriorates sharply. Here the actual success criterion is weekend fixtures match appropriate seasonality while atypical weekend changes still alert.
C:That acceptance test is directed at a different drift detection and lakehouse monitoring situation: A monitoring team wants a business metric for the share of transactions exceeding a risk threshold but records only the raw count of high-risk transactions. Here the actual success criterion is weekend fixtures match appropriate seasonality while atypical weekend changes still alert.
D: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 weekend fixtures match appropriate seasonality while atypical weekend changes still alert.
Question 850
Review this production model incident: A multilingual prediction service must handle regional data residency, but the team copies a shared feature cache across regions without governance. What is the central engineering constraint behind this behavior?
- Bundle target parameters must resolve to explicitly authorized environment-specific resources.
- Product naming and resource schema changes should be checked against the active Databricks CLI and documentation for the deployed version.
- Scalable environment design must account for data locality, residency and endpoint boundaries.
- Declarative automation or asset-bundle definitions can version environment-specific ML assets and reproduce deployments.
Explanation
Correct Answer
C:The technical constraint is specific to the reported failure. Scalable environment design must account for data locality, residency and endpoint boundaries. To test it, map data classes, approved processing regions and cross-region feature access paths.
Incorrect Answers
A:That underlying principle 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. The current incident instead turns on scalable environment design must account for data locality, residency and endpoint boundaries.
B:That underlying principle 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. The current incident instead turns on scalable environment design must account for data locality, residency and endpoint boundaries.
D:That underlying principle is directed at a different environment architectures situation: A model serving endpoint is created manually in the UI and every staging deployment differs subtly from production because no deployment definition is saved. The current incident instead turns on scalable environment design must account for data locality, residency and endpoint boundaries.