Google Professional Machine Learning Engineer Practice Test 113
Question 1120
During model-training validation on a Google Cloud experimental project, the following condition occurs: A weeklong custom training job uses preemptible capacity but saves checkpoints only to a temporary local container filesystem. The change should preserve a meaningful independent model-quality comparison. Which configuration or procedure is the most appropriate correction?
- Use suitable text embeddings with vector similarity search and reranking rather than a binary spam decision.
- Configure early stopping with a justified patience window and restore the best validated checkpoint.
- Persist versioned checkpoints in durable Cloud Storage and make the training code resume optimizer and model state.
- Adapt the training program to the managed data input contract or stage files to an explicit configurable path.
Explanation
Correct Answer
C:Persist versioned checkpoints in durable Cloud Storage and make the training code resume optimizer and model state. This is the targeted corrective response for training checkpoint persistence across interruptions. Preemptible execution can end abruptly; local-only state is not a durable recovery mechanism. Verify the diagnosis using: Inspect shutdown events, checkpoint directory, durable artifact transfers and restoration logic after simulated interruptions. Confirm success using: A controlled interruption resumes near the saved step without restarting training from zero. This matters here because during model-training validation on a Google Cloud experimental project, the following condition occurs: A weeklong….
Incorrect Answers
A:Use suitable text embeddings with vector similarity search and reranking rather than a binary spam decision. This action addresses embedding retrieval versus text classifier, not the failure in the described training checkpoint persistence across interruptions case. The targeted correction is: Persist versioned checkpoints in durable Cloud Storage and make the training code resume optimizer and model state.
B:Configure early stopping with a justified patience window and restore the best validated checkpoint. This action addresses early stopping on validation divergence, not the failure in the described training checkpoint persistence across interruptions case. The targeted correction is: Persist versioned checkpoints in durable Cloud Storage and make the training code resume optimizer and model state.
D:Adapt the training program to the managed data input contract or stage files to an explicit configurable path. This action addresses custom training input schema mismatch, not the failure in the described training checkpoint persistence across interruptions case. The targeted correction is: Persist versioned checkpoints in durable Cloud Storage and make the training code resume optimizer and model state. This matters here because during model-training validation on a Google Cloud experimental project, the following condition occurs: A weeklong….
Question 1121
During a staged restoration test at a secondary serving deployment, the following condition occurs: A new image classifier has only offline test results, yet engineers plan to replace every customer prediction immediately after deployment. The team must prove an actual user-visible improvement without weakening access controls. What should the engineer inspect before selecting a correction?
- Inspect feature-view sync status, latest source timestamps, entity ID and online lookup freshness.
- Inspect traffic-splitting capabilities, release guardrails, rollback triggers and representative online metrics.
- Inspect trained class-to-index metadata, endpoint output tensors and gateway mapping version.
- Inspect experiment assignment IDs, routing hashes, variant weights and per-user exposure histories.
Explanation
Correct Answer
B:Inspect traffic-splitting capabilities, release guardrails, rollback triggers and representative online metrics. This inspects evidence relevant to canary gradual traffic rollout. Progressive traffic exposure reduces blast radius compared with instant replacement of all production serving. After identifying the cause, the matching correction is: Roll out a small canary percentage first, monitor error and quality guardrails and expand only after acceptance. This matters here because during a staged restoration test at a secondary serving deployment, the following condition occurs: A new image….
Incorrect Answers
A:Inspect feature-view sync status, latest source timestamps, entity ID and online lookup freshness. This examines feature store freshness and online sync rather than the specific mechanism reported here. It is not the decisive first check for canary gradual traffic rollout. The relevant evidence is: Inspect traffic-splitting capabilities, release guardrails, rollback triggers and representative online metrics.
C:Inspect trained class-to-index metadata, endpoint output tensors and gateway mapping version. This examines custom postprocessing label mapping drift rather than the specific mechanism reported here. It is not the decisive first check for canary gradual traffic rollout. The relevant evidence is: Inspect traffic-splitting capabilities, release guardrails, rollback triggers and representative online metrics.
D:Inspect experiment assignment IDs, routing hashes, variant weights and per-user exposure histories. This examines a/b assignment consistency for user journeys rather than the specific mechanism reported here. It is not the decisive first check for canary gradual traffic rollout. The relevant evidence is: Inspect traffic-splitting capabilities, release guardrails, rollback triggers and representative online metrics. This matters here because during a staged restoration test at a secondary serving deployment, the following condition occurs: A new image….
Question 1122
While preparing a production training workflow on Gemini Enterprise Agent Platform, the following condition occurs: A fraud team with 40 engineered numeric transaction features starts with a costly vision transformer while ignoring a strong tabular baseline. Engineers need an auditable, resource-aware correction to the issue. Which targeted engineering change most directly addresses the failure?
- Prefer suitable data parallel replicas with synchronized gradient aggregation, then benchmark scale efficiency.
- Choose a decision threshold using acceptable recall and alert burden with a validated independent cohort.
- Compare an interpretable model or validated feature attribution approach with the black box and document decision tradeoffs.
- Benchmark a suitable boosted-tree model against more complex candidates and retain complexity only when justified by held-out gains.
Explanation
Correct Answer
D:Benchmark a suitable boosted-tree model against more complex candidates and retain complexity only when justified by held-out gains. This is the targeted corrective response for deep model versus gradient boosted trees on tabular. Structured tabular datasets often have competitive simple baselines; architectural complexity is not inherently better. Verify the diagnosis using: Inspect feature modality, example counts, training cost, validation scores and latency requirements. Confirm success using: The selected candidate meets fraud recall, inference latency and cost limits on a representative untouched split. This matters here because while preparing a production training workflow on Gemini Enterprise Agent Platform, the following condition occurs:….
Incorrect Answers
A:Prefer suitable data parallel replicas with synchronized gradient aggregation, then benchmark scale efficiency. This action addresses distributed data parallel gradient synchronization, not the failure in the described deep model versus gradient boosted trees on tabular case. The targeted correction is: Benchmark a suitable boosted-tree model against more complex candidates and retain complexity only when justified by held-out gains.
B:Choose a decision threshold using acceptable recall and alert burden with a validated independent cohort. This action addresses cost-sensitive threshold for rare positives, not the failure in the described deep model versus gradient boosted trees on tabular case. The targeted correction is: Benchmark a suitable boosted-tree model against more complex candidates and retain complexity only when justified by held-out gains.
C:Compare an interpretable model or validated feature attribution approach with the black box and document decision tradeoffs. This action addresses interpretability-constrained model baseline, not the failure in the described deep model versus gradient boosted trees on tabular case. The targeted correction is: Benchmark a suitable boosted-tree model against more complex candidates and retain complexity only when justified by held-out gains. This matters here because while preparing a production training workflow on Gemini Enterprise Agent Platform, the following condition occurs:….
Question 1123
During model-training validation on a Google Cloud experimental project, the following condition occurs: A training job remains pending for hours because its requested high-end GPU count exceeds project quota in the chosen region. The change should preserve a meaningful independent model-quality comparison. What acceptance result verifies the intended behavior after remediation?
- The job schedules successfully with the approved hardware and remains inside budget and residency constraints.
- Relevant articles rank highly for unseen paraphrased queries and retrieval hit rate is measured against judgments.
- Scale efficiency improves and model quality remains stable without synchronization stalls dominating every step.
- The chosen accelerator meets throughput and convergence expectations without unsupported operations or unstable retracing.
Explanation
Correct Answer
A:The job schedules successfully with the approved hardware and remains inside budget and residency constraints. This is specific acceptance evidence for accelerator quota and regional availability. Cloud accelerators have regional quota and stock limits distinct from an application's logical training configuration. It confirms the controlled correction: Request justified quota or select an available compatible accelerator and region without compromising data residency controls. This matters here because during model-training validation on a Google Cloud experimental project, the following condition occurs: A training….
Incorrect Answers
B:Relevant articles rank highly for unseen paraphrased queries and retrieval hit rate is measured against judgments. This result could demonstrate success for embedding retrieval versus text classifier but not for the stated accelerator quota and regional availability issue. The direct test is: The job schedules successfully with the approved hardware and remains inside budget and residency constraints.
C:Scale efficiency improves and model quality remains stable without synchronization stalls dominating every step. This result could demonstrate success for distributed gradient communication bottleneck but not for the stated accelerator quota and regional availability issue. The direct test is: The job schedules successfully with the approved hardware and remains inside budget and residency constraints. This matters here because during model-training validation on a Google Cloud experimental project, the following condition occurs: A training….
D:The chosen accelerator meets throughput and convergence expectations without unsupported operations or unstable retracing. This result could demonstrate success for tpu versus gpu for tensorflow workload but not for the stated accelerator quota and regional availability issue. The direct test is: The job schedules successfully with the approved hardware and remains inside budget and residency constraints. This matters here because during model-training validation on a Google Cloud experimental project, the following condition occurs: A training….
Question 1124
While preparing a production training workflow on Gemini Enterprise Agent Platform, the following condition occurs: A retailer asks for next-week order value as a dollar amount, but the prototype returns yes/no purchase classes without estimated magnitude. Engineers need an auditable, resource-aware correction to the issue. Which technical explanation best accounts for the reported behavior?
- Reduced numeric precision narrows representable ranges and can underflow or overflow without proper scaling.
- Categorical classification is not an estimator of an unrestricted continuous order value.
- Interpretability is an engineering requirement that must be considered with predictive power, not appended as a label.
- Prompt-only requests for JSON may not guarantee type validity; interface contracts need machine-verifiable constraints.
Explanation
Correct Answer
B:Categorical classification is not an estimator of an unrestricted continuous order value. This explains the mechanism behind regression versus classification product choice and why the described failure is possible. A discriminating investigation is: Inspect required output type, business loss, sample labels, and the configured model objective and metric. The corresponding repair is: Select a regression architecture and matching numeric loss rather than using a binary classifier for amounts. This matters here because while preparing a production training workflow on Gemini Enterprise Agent Platform, the following condition occurs:….
Incorrect Answers
A:Reduced numeric precision narrows representable ranges and can underflow or overflow without proper scaling. This principle describes mixed precision overflow diagnosis, which is not the cause of the observed regression versus classification product choice failure. The controlling explanation is: Categorical classification is not an estimator of an unrestricted continuous order value.
C:Interpretability is an engineering requirement that must be considered with predictive power, not appended as a label. This principle describes interpretability-constrained model baseline, which is not the cause of the observed regression versus classification product choice failure. The controlling explanation is: Categorical classification is not an estimator of an unrestricted continuous order value. This matters here because while preparing a production training workflow on Gemini Enterprise Agent Platform, the following condition occurs:….
D:Prompt-only requests for JSON may not guarantee type validity; interface contracts need machine-verifiable constraints. This principle describes constrained generative json outputs, which is not the cause of the observed regression versus classification product choice failure. The controlling explanation is: Categorical classification is not an estimator of an unrestricted continuous order value. This matters here because while preparing a production training workflow on Gemini Enterprise Agent Platform, the following condition occurs:….
Question 1125
At a Google Cloud analytics team preparing a governed pilot, the following condition occurs: A financial regression model reports astonishing test accuracy because its validation query includes a post-settlement field derived from the target. The team must satisfy a reproducible low-code AI assessment. Which outcome most directly demonstrates a successful, measurable fix?
- A revised policy becomes answerable after document indexing without retraining, and retrieved citations match the active version.
- A trained candidate reaches the agreed quality threshold without maintaining unsupported bespoke training infrastructure.
- Held-out metrics reflect only features available before the prediction timestamp and remain credible on later periods.
- An unseen labeled cohort receives probabilities for the intended positive class with a calibrated operating threshold.
Explanation
Correct Answer
C:Held-out metrics reflect only features available before the prediction timestamp and remain credible on later periods. This is specific acceptance evidence for bigquery ml ml.evaluate leakage detection. A label-derived future field leaks the outcome into input features and makes offline metrics deceptively strong. It confirms the controlled correction: Exclude post-outcome fields, create point-in-time valid training and evaluation partitions, and rerun ML.EVALUATE. Here, the relevant situation is: the following condition occurs: A financial regression model reports astonishing test accuracy because its….
Incorrect Answers
A:A revised policy becomes answerable after document indexing without retraining, and retrieved citations match the active version. This result could demonstrate success for foundation model tuning versus retrieval update but not for the stated bigquery ml ml.evaluate leakage detection issue. The direct test is: Held-out metrics reflect only features available before the prediction timestamp and remain credible on later periods.
B:A trained candidate reaches the agreed quality threshold without maintaining unsupported bespoke training infrastructure. This result could demonstrate success for automl tabular supervised model selection but not for the stated bigquery ml ml.evaluate leakage detection issue. The direct test is: Held-out metrics reflect only features available before the prediction timestamp and remain credible on later periods.
D:An unseen labeled cohort receives probabilities for the intended positive class with a calibrated operating threshold. This result could demonstrate success for bigquery ml logistic model for conversion probability but not for the stated bigquery ml ml.evaluate leakage detection issue. The direct test is: Held-out metrics reflect only features available before the prediction timestamp and remain credible on later periods.
Question 1126
Within a jointly maintained Google Cloud data science workspace, the following condition occurs: A churn model joins historical feature rows to today's latest cancellation status, exposing a future outcome to prior prediction dates. Collaborators must avoid expanding data access or breaking experiment reproducibility. What implementation change resolves the specific issue without substituting an unrelated fix?
- Rebuild offline training data with point-in-time eligible joins and remove features available only after each prediction moment.
- Refactor hidden state into explicit inputs and functions and add a clean-kernel end-to-end execution gate.
- Rewrite the filter to permit effective partition pruning while preserving its requested date boundaries.
- Implement the batch aggregate in scheduled BigQuery SQL when the required transformation is expressible and fits its SLA.
Explanation
Correct Answer
A:Rebuild offline training data with point-in-time eligible joins and remove features available only after each prediction moment. This is the targeted corrective response for point-in-time feature leakage prevention. The globally latest value may contain future information that could not have informed the original prediction. Verify the diagnosis using: Inspect feature event timestamps, prediction cutoffs, feature joins and any post-cancellation administrative flags. Confirm success using: For every row the feature version predates its prediction cutoff and prospective holdout quality is realistic. This matters here because within a jointly maintained Google Cloud data science workspace, the following condition occurs: A churn model joins….
Incorrect Answers
B:Refactor hidden state into explicit inputs and functions and add a clean-kernel end-to-end execution gate. This action addresses notebook transient state and execution order, not the failure in the described point-in-time feature leakage prevention case. The targeted correction is: Rebuild offline training data with point-in-time eligible joins and remove features available only after each prediction moment. This matters here because within a jointly maintained Google Cloud data science workspace, the following condition occurs: A churn model joins….
C:Rewrite the filter to permit effective partition pruning while preserving its requested date boundaries. This action addresses bigquery partition filter cost optimization, not the failure in the described point-in-time feature leakage prevention case. The targeted correction is: Rebuild offline training data with point-in-time eligible joins and remove features available only after each prediction moment.
D:Implement the batch aggregate in scheduled BigQuery SQL when the required transformation is expressible and fits its SLA. This action addresses bigquery versus dataflow preparation selection, not the failure in the described point-in-time feature leakage prevention case. The targeted correction is: Rebuild offline training data with point-in-time eligible joins and remove features available only after each prediction moment.
Question 1127
While preparing a production training workflow on Gemini Enterprise Agent Platform, the following condition occurs: A neural network's training loss continues falling for fifty epochs while validation loss starts rising after epoch ten. Engineers need an auditable, resource-aware correction to the issue. Which TWO choices together provide the decisive diagnostic evidence and the most direct acceptance test? Select TWO.
- Compare training and validation curves, learning-rate schedule and the checkpoint corresponding to minimum validation loss.
- Auditors can reproduce feature influence evidence while holdout quality remains within the approved threshold.
- Training loss remains finite under representative batches while memory and throughput benefits are measured.
- Inspect task input configuration, container logs, expected file patterns and staged object paths.
- The deployed artifact corresponds to the best held-out epoch rather than the final overfit epoch.
Explanation
Correct Answers
A:Compare training and validation curves, learning-rate schedule and the checkpoint corresponding to minimum validation loss. This inspects evidence relevant to early stopping on validation divergence. More training steps can increase overfitting even when training-set loss keeps improving. After identifying the cause, the matching correction is: Configure early stopping with a justified patience window and restore the best validated checkpoint.
E:The deployed artifact corresponds to the best held-out epoch rather than the final overfit epoch. This is specific acceptance evidence for early stopping on validation divergence. More training steps can increase overfitting even when training-set loss keeps improving. It confirms the controlled correction: Configure early stopping with a justified patience window and restore the best validated checkpoint.
Incorrect Answers
B:Auditors can reproduce feature influence evidence while holdout quality remains within the approved threshold. This relates to interpretability-constrained model baseline but does not establish the requested before-and-after evidence. The actual diagnostic is: Compare training and validation curves, learning-rate schedule and the checkpoint corresponding to minimum validation loss. The specific acceptance result is: The deployed artifact corresponds to the best held-out epoch rather than the final overfit epoch.
C:Training loss remains finite under representative batches while memory and throughput benefits are measured. This relates to mixed precision overflow diagnosis but does not establish the requested before-and-after evidence. The actual diagnostic is: Compare training and validation curves, learning-rate schedule and the checkpoint corresponding to minimum validation loss. The specific acceptance result is: The deployed artifact corresponds to the best held-out epoch rather than the final overfit epoch.
D:Inspect task input configuration, container logs, expected file patterns and staged object paths. This relates to custom training input schema mismatch but does not establish the requested before-and-after evidence. The actual diagnostic is: Compare training and validation curves, learning-rate schedule and the checkpoint corresponding to minimum validation loss. The specific acceptance result is: The deployed artifact corresponds to the best held-out epoch rather than the final overfit epoch.
Question 1128
During a staged restoration test at a secondary serving deployment, the following condition occurs: A customer-facing inference API has high end-to-end latency in Asia while model execution is quick and the endpoint runs on another continent. The team must prove an actual user-visible improvement without weakening access controls. What acceptance result verifies the intended behavior after remediation?
- Regional clients experience lower p95 end-to-end latency without changes to model compute quality.
- Requests per second increase while tail latency stays inside the service objective.
- A known prediction request succeeds through the deployed endpoint and readiness reflects successful initialization.
- Authorized lookups return expected recent values and freshness alarms fire when synchronization falls behind.
Explanation
Correct Answer
A:Regional clients experience lower p95 end-to-end latency without changes to model compute quality. This is specific acceptance evidence for serving region proximity versus model compute. Network distance can dominate inference response times when the actual model execution is already short. It confirms the controlled correction: Select supported regional placement or routing close to users while respecting residency and model availability. This matters here because during a staged restoration test at a secondary serving deployment, the following condition occurs: A….
Incorrect Answers
B:Requests per second increase while tail latency stays inside the service objective. This result could demonstrate success for dynamic batch serving for gpu saturation but not for the stated serving region proximity versus model compute issue. The direct test is: Regional clients experience lower p95 end-to-end latency without changes to model compute quality. This matters here because during a staged restoration test at a secondary serving deployment, the following condition occurs: A….
C:A known prediction request succeeds through the deployed endpoint and readiness reflects successful initialization. This result could demonstrate success for inference container health interface mismatch but not for the stated serving region proximity versus model compute issue. The direct test is: Regional clients experience lower p95 end-to-end latency without changes to model compute quality.
D:Authorized lookups return expected recent values and freshness alarms fire when synchronization falls behind. This result could demonstrate success for feature store freshness and online sync but not for the stated serving region proximity versus model compute issue. The direct test is: Regional clients experience lower p95 end-to-end latency without changes to model compute quality.
Question 1129
During model-training validation on a Google Cloud experimental project, the following condition occurs: A custom PyTorch training job runs out of accelerator memory immediately after per-device batch size doubles while model architecture stays fixed. The change should preserve a meaningful independent model-quality comparison. Select TWO: the decisive diagnostic step and the targeted repair for this specific issue.
- Inspect GPU memory traces, activation tensor sizes, batch size, optimizer state and training failure logs.
- Compare pretraining benchmark scores, domain validation curves, trainable parameters and fine-tuning learning-rate choices.
- Split raw records first and resample only inside training folds, leaving validation cohorts untouched.
- Reduce per-device batch size or use validated gradient accumulation or mixed precision where compatible.
- Request justified quota or select an available compatible accelerator and region without compromising data residency controls.
Explanation
Correct Answers
A:Inspect GPU memory traces, activation tensor sizes, batch size, optimizer state and training failure logs. This inspects evidence relevant to gpu out-of-memory batch control. Training activation memory commonly increases with batch size, independent of available CPU cores. After identifying the cause, the matching correction is: Reduce per-device batch size or use validated gradient accumulation or mixed precision where compatible. This matters here because during model-training validation on a Google Cloud experimental project, the following condition occurs: A custom….
D:Reduce per-device batch size or use validated gradient accumulation or mixed precision where compatible. This is the targeted corrective response for gpu out-of-memory batch control. Training activation memory commonly increases with batch size, independent of available CPU cores. Verify the diagnosis using: Inspect GPU memory traces, activation tensor sizes, batch size, optimizer state and training failure logs. Confirm success using: The job completes representative epochs without OOM and convergence remains within the approved tolerance. This matters here because during model-training validation on a Google Cloud experimental project, the following condition occurs: A custom….
Incorrect Answers
B:Compare pretraining benchmark scores, domain validation curves, trainable parameters and fine-tuning learning-rate choices. This is an action or diagnostic relevant to fine-tuning catastrophic forgetting, not one of the required two steps. In this scenario verify: Inspect GPU memory traces, activation tensor sizes, batch size, optimizer state and training failure logs. Then correct: Reduce per-device batch size or use validated gradient accumulation or mixed precision where compatible.
C:Split raw records first and resample only inside training folds, leaving validation cohorts untouched. This is an action or diagnostic relevant to training class imbalance fold leakage, not one of the required two steps. In this scenario verify: Inspect GPU memory traces, activation tensor sizes, batch size, optimizer state and training failure logs. Then correct: Reduce per-device batch size or use validated gradient accumulation or mixed precision where compatible.
E:Request justified quota or select an available compatible accelerator and region without compromising data residency controls. This is an action or diagnostic relevant to accelerator quota and regional availability, not one of the required two steps. In this scenario verify: Inspect GPU memory traces, activation tensor sizes, batch size, optimizer state and training failure logs. Then correct: Reduce per-device batch size or use validated gradient accumulation or mixed precision where compatible.
Question 1130
During model-training validation on a Google Cloud experimental project, the following condition occurs: A weeklong custom training job uses preemptible capacity but saves checkpoints only to a temporary local container filesystem. The change should preserve a meaningful independent model-quality comparison. Which underlying engineering principle explains why the original approach fails?
- Preemptible execution can end abruptly; local-only state is not a durable recovery mechanism.
- Reduced numeric precision narrows representable ranges and can underflow or overflow without proper scaling.
- An accelerator can sit idle waiting for input even when its own memory and compute capacity are sufficient.
- Notebook filesystem locations are not portable contracts for managed container training jobs.
Explanation
Correct Answer
A:Preemptible execution can end abruptly; local-only state is not a durable recovery mechanism. This explains the mechanism behind training checkpoint persistence across interruptions and why the described failure is possible. A discriminating investigation is: Inspect shutdown events, checkpoint directory, durable artifact transfers and restoration logic after simulated interruptions. The corresponding repair is: Persist versioned checkpoints in durable Cloud Storage and make the training code resume optimizer and model state.
Incorrect Answers
B:Reduced numeric precision narrows representable ranges and can underflow or overflow without proper scaling. This principle describes mixed precision overflow diagnosis, which is not the cause of the observed training checkpoint persistence across interruptions failure. The controlling explanation is: Preemptible execution can end abruptly; local-only state is not a durable recovery mechanism.
C:An accelerator can sit idle waiting for input even when its own memory and compute capacity are sufficient. This principle describes accelerator starvation from slow input pipeline, which is not the cause of the observed training checkpoint persistence across interruptions failure. The controlling explanation is: Preemptible execution can end abruptly; local-only state is not a durable recovery mechanism.
D:Notebook filesystem locations are not portable contracts for managed container training jobs. This principle describes custom training input schema mismatch, which is not the cause of the observed training checkpoint persistence across interruptions failure. The controlling explanation is: Preemptible execution can end abruptly; local-only state is not a durable recovery mechanism.