Databricks Certified Machine Learning Professional Practice Test 44
Question 430
The engineering team encounters the following issue: A team publishes an online feature table but neglects its refresh schedule while the predictive model assumes values no older than 15 minutes. What direct consequence should the owner prioritize if no fix is made?
- An apparently healthy endpoint serves outdated feature values and produces obsolete scores.
- Scores after failover revert to incomplete or uninitialized feature state.
- Training-serving skew silently changes the meaning of risk thresholds.
- A backtest gains information that production inference could not have used.
Explanation
Correct Answer
A:This consequence follows from the specific unresolved condition. An apparently healthy endpoint serves outdated feature values and produces obsolete scores. The causal constraint is: An online store's existence does not guarantee its feature freshness; synchronization must be monitored.
Incorrect Answers
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 The more immediate risk here is an apparently healthy endpoint serves outdated feature values and produces obsolete scores.
C:That direct consequence is directed at a different advanced feature store concepts situation: A customer-risk feature needs a ratio of current debt to income, but online serving recomputes it differently from the training pipeline. The more immediate risk here is an apparently healthy endpoint serves outdated feature values and produces obsolete scores.
D: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 The more immediate risk here is an apparently healthy endpoint serves outdated feature values and produces obsolete scores.
Question 431
Review this production model incident: An ML project has a single sequential external feature-service dependency and expects data-parallel workers to increase throughput 20-fold. What evidence would best distinguish the actual fault from neighboring ML concerns?
- Compare sample group outputs and null/type handling against the declared pandas function schema.
- Compare dataset version, random seed, fold definition and metric contract attached to each MLflow run.
- Compare persistent Ray tuning result storage and restoration metadata with the fresh-launch configuration.
- Profile each step's critical path and shared upstream request quotas.
Explanation
Correct Answer
D:This is the most discriminating check for the observed symptoms. Profile each step's critical path and shared upstream request quotas. The underlying technical issue is that parallelizable fraction and upstream service limits bound the benefits of adding model-training workers.
Incorrect Answers
A:That diagnostic evidence is directed at a different scaling and tuning situation: A per-city pandas UDF returns different prediction schemas for cities with absent categories, causing a distributed aggregation failure. For the stated incident, the decisive observation is profile each step's critical path and shared upstream request quotas.
B:That diagnostic evidence is directed at a different scaling and tuning situation: A hyperparameter team compares two Optuna studies with different train/validation splits and chooses the lower loss as though both metrics were directly comparable. For the stated incident, the decisive observation is profile each step's critical path and shared upstream request quotas.
C:That diagnostic evidence is directed at a different scaling and tuning situation: An interrupted Ray hyperparameter sweep has an expensive completed-trial history but a new launch discards its experiment checkpoint. For the stated incident, the decisive observation is profile each step's critical path and shared upstream request quotas.
Question 432
Consider the following fictional Databricks production case: A feature stream loses half its source events and a retraining policy treats the resulting input shift as proof that the learned concept changed. Which observation most directly tests the suspected failure boundary?
- Compare candidate predictive metrics with model footprint, serving latency and expected concurrency.
- Compare source completeness, freshness and drift metrics across the same incident window.
- Compare champion and candidate per-segment false-negative rates on equivalent untouched evaluation cohorts.
- Compare holdout snapshot IDs, label maturity and metric computation versions in candidate MLflow runs.
Explanation
Correct Answer
B:This is the most discriminating check for the observed symptoms. Compare source completeness, freshness and drift metrics across the same incident window. The underlying technical issue is that data pipeline outages must be distinguished from genuine population shift before retraining.
Incorrect Answers
A:That diagnostic evidence 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. For the stated incident, the decisive observation is compare source completeness, freshness and drift metrics across the same incident window.
C:That diagnostic evidence is directed at a different automated retraining situation: A new fraud classifier improves overall F1 but doubles false negatives for high-value payment accounts, and an automated policy chooses it on F1 alone. For the stated incident, the decisive observation is compare source completeness, freshness and drift metrics across the same incident window.
D:That diagnostic evidence 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. For the stated incident, the decisive observation is compare source completeness, freshness and drift metrics across the same incident window.
Question 433
A production ML review contains this situation: A preprocessing function divides by a standard deviation and returns infinity for a constant numeric column. What is the most material immediate risk in this situation?
- A library compatibility defect escapes despite green tests against an older environment.
- Different releases disagree on whether a model passes the same quality gate.
- A model that looks accurate offline cannot be called successfully by production clients.
- Invalid standardized values contaminate model input vectors or crash scoring.
Explanation
Correct Answer
D:This consequence follows from the specific unresolved condition. Invalid standardized values contaminate model input vectors or crash scoring. The causal constraint is: Deterministic function tests should include degenerate data distributions that violate assumed mathematical denominators.
Incorrect Answers
A:That direct consequence is directed at a different validation testing situation: A release pipeline runs development unit tests against a different dependency image from the one deployed to production. The more immediate risk here is invalid standardized values contaminate model input vectors or crash scoring.
B:That direct consequence is directed at a different validation testing situation: An engineering group duplicates a validation helper across six notebooks, and two copies enforce outdated threshold values. The more immediate risk here is invalid standardized values contaminate model input vectors or crash scoring.
C:That direct consequence is directed at a different validation testing situation: A churn feature job succeeds and a classifier achieves acceptable F1, but the deployed endpoint returns HTTP 400 because request column names differ. The more immediate risk here is invalid standardized values contaminate model input vectors or crash scoring.
Question 434
Consider the following fictional Databricks production case: A new merchant group contributes only 12 observations, yet a monitor reports a large categorical divergence from a baseline of 50,000 and the team assumes the result is definitive. Which implementation is most appropriate for the stated incident?
- Repair the profile computation and alert on stale metric output separately from the model's predictions.
- Apply documented sample-sufficiency checks and collect more representative data before strong conclusions.
- Create a dedicated approved critical-region alert with suitable minimum volume and failure-rate conditions.
- Configure a supported time-series profile with the appropriate timestamp and daily granularity.
Explanation
Correct Answer
B:This corrects the relevant failure boundary instead of merely hiding the symptom. Apply documented sample-sufficiency checks and collect more representative data before strong conclusions. The issue matters because random variation in very small samples causes unstable alarms and false retraining triggers.
Incorrect Answers
A:That corrective action is directed at a different drift detection and lakehouse monitoring situation: A team has accurate inference logs but its monitoring job never refreshes computed drift metrics after a table schema change. The relevant corrective action here is apply documented sample-sufficiency checks and collect more representative data before strong conclusions.
C:That corrective action is directed at a different drift detection and lakehouse monitoring situation: An alert query uses one global transaction-failure count and masks a complete outage in a small high-value region. The relevant corrective action here is apply documented sample-sufficiency checks and collect more representative data before strong conclusions.
D: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 apply documented sample-sufficiency checks and collect more representative data before strong conclusions.
Question 435
Consider the following fictional Databricks production case: A recommendation team wraps a large pandas UDF around entire customer histories, then discovers some partitions exceed worker memory while others finish quickly. Which result is the strongest acceptance criterion after remediation?
- Worst-case synthetic groups complete under the documented worker memory budget.
- Every completed trial has a linked run, parameter map, metric and deterministic winner selection.
- End-to-end throughput improves after removing the true serialized bottleneck.
- A measured end-to-end trial batch reuses data without redundant full-table copies.
Explanation
Correct Answer
A:This verification measures the required result rather than just process liveness. Worst-case synthetic groups complete under the documented worker memory budget. It confirms whether the targeted action succeeded: repartition or redesign oversized groups, set appropriate batch boundaries and validate memory headroom.
Incorrect Answers
B:That acceptance test 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. Here the actual success criterion is worst-case synthetic groups complete under the documented worker memory budget.
C:That acceptance test is directed at a different scaling and tuning situation: An ML project has a single sequential external feature-service dependency and expects data-parallel workers to increase throughput 20-fold. Here the actual success criterion is worst-case synthetic groups complete under the documented worker memory budget.
D:That acceptance test 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. Here the actual success criterion is worst-case synthetic groups complete under the documented worker memory budget.
Question 436
The engineering team encounters the following issue: 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. Which technical principle most directly explains the root of the issue?
- Small workloads may run more efficiently on one node because Spark startup, scheduling and serialization have fixed costs.
- StringIndexer is an Estimator that must fit its category mapping before its resulting model can transform records.
- Any fitted preprocessing stage can learn from holdout records and must be fitted within the training boundary.
- Training-set fit does not establish out-of-sample predictive quality or cold-start behavior.
Explanation
Correct Answer
A:The technical constraint is specific to the reported failure. Small workloads may run more efficiently on one node because Spark startup, scheduling and serialization have fixed costs. To test it, measure the local estimator runtime against cluster initialization and task scheduling for the same sample.
Incorrect Answers
B:That underlying principle 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. The current incident instead turns on small workloads may run more efficiently on one node because Spark startup, scheduling and serialization have fixed costs.
C:That underlying principle is directed at a different using spark ml situation: A credit-risk pipeline fits StringIndexer on the entire dataset before splitting train and holdout, and rare holdout labels influence its learned category mapping. The current incident instead turns on small workloads may run more efficiently on one node because Spark startup, scheduling and serialization have fixed costs.
D: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 small workloads may run more efficiently on one node because Spark startup, scheduling and serialization have fixed costs.
Question 437
A production ML review contains this situation: A fraud PyFunc uses a global mutable feature cache initialized by the first request and mixes data between subsequent users. Which check shows that this particular problem has been resolved?
- Concurrent cross-user fixtures return isolated correct outputs with no cache-key collision.
- Reviewers can open a versioned chart artifact from the same run as its calibration metrics.
- Each worker's fit results appear under its own child run without overwritten artifacts.
- A scan of new run artifacts finds only approved fields while reproducing the required diagnostic behavior.
Explanation
Correct Answer
A:This verification measures the required result rather than just process liveness. Concurrent cross-user fixtures return isolated correct outputs with no cache-key collision. It confirms whether the targeted action succeeded: remove unsafe global request-specific state and use explicitly keyed, governed feature resolution.
Incorrect Answers
B:That acceptance test is directed at a different advanced mlflow usage situation: An ML engineer logs a calibration plot as a string metric, then downstream reviewers cannot inspect the visualization or its input version. Here the actual success criterion is concurrent cross-user fixtures return isolated correct outputs with no cache-key collision.
C:That acceptance test is directed at a different advanced mlflow usage situation: A training orchestrator starts parallel child fits but all workers accidentally log into one active MLflow run and overwrite each other's metrics. Here the actual success criterion is concurrent cross-user fixtures return isolated correct outputs with no cache-key collision.
D:That acceptance test is directed at a different advanced mlflow usage situation: A regulated analytics team uploads full customer identifiers inside MLflow model artifacts during troubleshooting. Here the actual success criterion is concurrent cross-user fixtures return isolated correct outputs with no cache-key collision.
Question 438
Consider the following fictional Databricks production case: An ML project has a single sequential external feature-service dependency and expects data-parallel workers to increase throughput 20-fold. Which implementation is most appropriate for the stated incident?
- Use a supported grouped pandas function to fit a separate model for each store with stable group keys.
- Eliminate or batch the serialized dependency before purchasing more training replicas.
- Pin shared validation protocol and data snapshot for the compared Optuna trials.
- Scale worker capacity and partition layout according to measured distributed CPU demand.
Explanation
Correct Answer
B:This corrects the relevant failure boundary instead of merely hiding the symptom. Eliminate or batch the serialized dependency before purchasing more training replicas. The issue matters because more workers only saturate the service, increasing retries and overall latency.
Incorrect Answers
A:That corrective action is directed at a different scaling and tuning situation: A retailer needs one independent forecasting model per store but concatenates all store histories before fitting one global model. The relevant corrective action here is eliminate or batch the serialized dependency before purchasing more training replicas.
C:That corrective action is directed at a different scaling and tuning situation: A hyperparameter team compares two Optuna studies with different train/validation splits and chooses the lower loss as though both metrics were directly comparable. The relevant corrective action here is eliminate or batch the serialized dependency before purchasing more training replicas.
D:That corrective action is directed at a different scaling and tuning situation: A feature-engineering job is CPU-bound on hundreds of independent daily partitions and a team upgrades only the driver to a much larger machine. The relevant corrective action here is eliminate or batch the serialized dependency before purchasing more training replicas.
Question 439
The engineering team encounters the following issue: A feature function maps a null purchase amount to zero without distinguishing missing observations, and a notebook unit test covers only positive amounts. Which technical principle most directly explains the root of the issue?
- Integration fixtures must check semantic prediction values and class labels, not only response existence.
- Separating pure functions from orchestration can make unit tests faster while retaining notebook-driven workflows.
- Unit tests belong on controlled synthetic inputs before integrations, and production smoke tests do not replace them.
- Unit tests should exercise edge cases and expected output contracts, not just the happy-path fixture.
Explanation
Correct Answer
D:The technical constraint is specific to the reported failure. Unit tests should exercise edge cases and expected output contracts, not just the happy-path fixture. To test it, inspect function outputs for null, zero and positive synthetic inputs against the documented business rule.
Incorrect Answers
A:That underlying principle is directed at a different validation testing situation: An end-to-end validation uses only a synthetic positive label and fails to detect a swapped positive/negative class mapping in inference. The current incident instead turns on unit tests should exercise edge cases and expected output contracts, not just the happy-path fixture.
B:That underlying principle is directed at a different validation testing situation: A team hides all preprocessing logic inside one long notebook, making it hard to test independently without starting a Spark cluster. The current incident instead turns on unit tests should exercise edge cases and expected output contracts, not just the happy-path fixture.
C:That underlying principle is directed at a different validation testing situation: A developer changes a feature transform in a notebook and directly uses production data as the only test case. The current incident instead turns on unit tests should exercise edge cases and expected output contracts, not just the happy-path fixture.
Question 440
In an enterprise ML workflow, the reported issue is: A new model version uses twice the memory per replica, but capacity monitoring tracks only CPU and crash counts. What likely operational or model-quality harm does this issue create?
- Observability becomes a channel for customer-level data leakage.
- Instances restart unpredictably from memory pressure and disrupt scoring.
- The live population drifts far from model training even though daily changes remain modest.
- On-call teams are paged repeatedly for harmless short-lived noise.
Explanation
Correct Answer
B:This consequence follows from the specific unresolved condition. Instances restart unpredictably from memory pressure and disrupt scoring. The causal constraint is: Serving memory headroom and resource trends are necessary to detect OOM risk in large-model releases.
Incorrect Answers
A:That direct consequence is directed at a different drift detection and lakehouse monitoring situation: A monitor owner writes a custom metric that logs unaggregated account-level health data in a broadly shared results table. The more immediate risk here is instances restart unpredictably from memory pressure and disrupt scoring.
C:That direct consequence 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 more immediate risk here is instances restart unpredictably from memory pressure and disrupt scoring.
D:That direct consequence is directed at a different drift detection and lakehouse monitoring situation: A monitor alerts whenever a ratio exceeds 0.3 for a single minute, although normal traffic produces brief spikes and operations require ten-minute persistence. The more immediate risk here is instances restart unpredictably from memory pressure and disrupt scoring.