Google Professional Machine Learning Engineer Practice Test 95
Question 940
During model-training validation on a Google Cloud experimental project, the following condition occurs: A product search team trains a binary spam classifier to find the most semantically relevant help article for a free-text question. 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.
- Evaluate ARIMA_PLUS or an appropriate time-series forecasting model before adopting an expensive generative approach.
- Compare an interpretable model or validated feature attribution approach with the black box and document decision tradeoffs.
- Use a supported automatic mixed-precision and loss-scaling strategy or return unstable operations to suitable precision.
Explanation
Correct Answer
A:Use suitable text embeddings with vector similarity search and reranking rather than a binary spam decision. This is the targeted corrective response for embedding retrieval versus text classifier. A binary classifier assigns predefined labels; semantic search needs query-to-document relevance ranking. Verify the diagnosis using: Inspect task requirements for retrieval ranking, available query-document pairs, embedding similarity and labeled relevance examples. Confirm success using: Relevant articles rank highly for unseen paraphrased queries and retrieval hit rate is measured against judgments. This matters here because during model-training validation on a Google Cloud experimental project, the following condition occurs: A product….
Incorrect Answers
B:Evaluate ARIMA_PLUS or an appropriate time-series forecasting model before adopting an expensive generative approach. This action addresses forecast trend and seasonality versus llm, not the failure in the described embedding retrieval versus text classifier case. The targeted correction is: Use suitable text embeddings with vector similarity search and reranking rather than a binary spam decision.
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 embedding retrieval versus text classifier case. The targeted correction is: Use suitable text embeddings with vector similarity search and reranking rather than a binary spam decision.
D:Use a supported automatic mixed-precision and loss-scaling strategy or return unstable operations to suitable precision. This action addresses mixed precision overflow diagnosis, not the failure in the described embedding retrieval versus text classifier case. The targeted correction is: Use suitable text embeddings with vector similarity search and reranking rather than a binary spam decision.
Question 941
During a postdeployment AI safety and reliability review, the following condition occurs: A model deployment's business conversion rate falls during canary traffic, but the pipeline automatically continues rolling out because only container CPU is monitored. The model owner must provide repeatable evidence rather than a cosmetic dashboard change. Which technical explanation best accounts for the reported behavior?
- Feature attribution drift can indicate changing model reliance even when overall score distributions remain stable.
- Concept drift changes conditional label relationships and may occur without obvious marginal feature drift.
- Resource health cannot substitute for model outcome quality; release decisions need explicit relevant monitoring signals.
- Supervised production quality requires reliable prediction-to-ground-truth linkage, not only unlabeled model outputs.
Explanation
Correct Answer
C:Resource health cannot substitute for model outcome quality; release decisions need explicit relevant monitoring signals. This explains the mechanism behind production incident rollback based on metrics and why the described failure is possible. A discriminating investigation is: Inspect cohort conversion metrics, assignment validity, minimum sample requirements and the release guardrail configuration. The corresponding repair is: Require validated business quality and technical reliability thresholds to continue rollout, with tested rollback on failure.
Incorrect Answers
A:Feature attribution drift can indicate changing model reliance even when overall score distributions remain stable. This principle describes feature attribution drift after deployment, which is not the cause of the observed production incident rollback based on metrics failure. The controlling explanation is: Resource health cannot substitute for model outcome quality; release decisions need explicit relevant monitoring signals.
B:Concept drift changes conditional label relationships and may occur without obvious marginal feature drift. This principle describes concept drift with stable feature statistics, which is not the cause of the observed production incident rollback based on metrics failure. The controlling explanation is: Resource health cannot substitute for model outcome quality; release decisions need explicit relevant monitoring signals.
D:Supervised production quality requires reliable prediction-to-ground-truth linkage, not only unlabeled model outputs. This principle describes ground truth association for quality monitoring, which is not the cause of the observed production incident rollback based on metrics failure. The controlling explanation is: Resource health cannot substitute for model outcome quality; release decisions need explicit relevant monitoring signals.
Question 942
During model-training validation on a Google Cloud experimental project, the following condition occurs: A demand model trains on moving averages computed after the prediction cutoff, so the held-out forecast benefits from future sales updates. The change should preserve a meaningful independent model-quality comparison. Which statement identifies the primary reason for this failure?
- A binary classifier assigns predefined labels; semantic search needs query-to-document relevance ranking.
- Search quality depends on comparable objectives and resource limits, not exhaustive combinations at any cost.
- Temporal features formed after prediction time leak future observations into model training and evaluation.
- Reduced numeric precision narrows representable ranges and can underflow or overflow without proper scaling.
Explanation
Correct Answer
C:Temporal features formed after prediction time leak future observations into model training and evaluation. This explains the mechanism behind temporal training label leakage in rolling windows and why the described failure is possible. A discriminating investigation is: Inspect feature computation cutoffs, window frame bounds, label timestamps and time-aware evaluation splits. The corresponding repair is: Recompute lagged windows using only prior observations and validate with rolling-origin splits. This matters here because during model-training validation on a Google Cloud experimental project, the following condition occurs: A demand….
Incorrect Answers
A:A binary classifier assigns predefined labels; semantic search needs query-to-document relevance ranking. This principle describes embedding retrieval versus text classifier, which is not the cause of the observed temporal training label leakage in rolling windows failure. The controlling explanation is: Temporal features formed after prediction time leak future observations into model training and evaluation.
B:Search quality depends on comparable objectives and resource limits, not exhaustive combinations at any cost. This principle describes hyperparameter search budget limit, which is not the cause of the observed temporal training label leakage in rolling windows failure. The controlling explanation is: Temporal features formed after prediction time leak future observations into model training and evaluation.
D: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 temporal training label leakage in rolling windows failure. The controlling explanation is: Temporal features formed after prediction time leak future observations into model training and evaluation.
Question 943
During model-training validation on a Google Cloud experimental project, the following condition occurs: A customer-service model correctly follows general instructions but lacks newly updated internal warranty details, so developers schedule daily weight fine-tuning. The change should preserve a meaningful independent model-quality comparison. Which diagnostic check should be performed FIRST to isolate the described cause?
- Determine whether failures arise from changing reference knowledge, stable behavior patterns or inadequate retrieval coverage.
- Examine model governance constraints, feature sensitivity, explanation fidelity and performance of simpler interpretable baselines.
- Measure model footprint, per-device utilization, communication cost and batch throughput on one and multiple accelerators.
- Inspect time stamps, autocorrelation, horizon, seasonal cycles and out-of-sample forecast metrics against traditional baselines.
Explanation
Correct Answer
A:Determine whether failures arise from changing reference knowledge, stable behavior patterns or inadequate retrieval coverage. This inspects evidence relevant to foundation model prompt versus fine-tuning. External reference updates need fresh context; changing foundation weights for every factual revision is inefficient. After identifying the cause, the matching correction is: Use grounded retrieval for frequently changing policy facts and reserve fine-tuning for measured durable behavior adaptation. This matters here because during model-training validation on a Google Cloud experimental project, the following condition occurs: A….
Incorrect Answers
B:Examine model governance constraints, feature sensitivity, explanation fidelity and performance of simpler interpretable baselines. This examines interpretability-constrained model baseline rather than the specific mechanism reported here. It is not the decisive first check for foundation model prompt versus fine-tuning. The relevant evidence is: Determine whether failures arise from changing reference knowledge, stable behavior patterns or inadequate retrieval coverage.
C:Measure model footprint, per-device utilization, communication cost and batch throughput on one and multiple accelerators. This examines distributed data parallel gradient synchronization rather than the specific mechanism reported here. It is not the decisive first check for foundation model prompt versus fine-tuning. The relevant evidence is: Determine whether failures arise from changing reference knowledge, stable behavior patterns or inadequate retrieval coverage.
D:Inspect time stamps, autocorrelation, horizon, seasonal cycles and out-of-sample forecast metrics against traditional baselines. This examines forecast trend and seasonality versus llm rather than the specific mechanism reported here. It is not the decisive first check for foundation model prompt versus fine-tuning. The relevant evidence is: Determine whether failures arise from changing reference knowledge, stable behavior patterns or inadequate retrieval coverage.
Question 944
During a postdeployment AI safety and reliability review, the following condition occurs: An approval model shows unchanged overall AUC but a recent feature change sharply increases false negatives for a protected subgroup. The model owner must provide repeatable evidence rather than a cosmetic dashboard change. Which configuration or procedure is the most appropriate correction?
- Version a single feature contract and add automated parity tests to the preprocessing pipeline and serving workflow.
- Evaluate and address documented subgroup disparities using appropriate fairness metrics and approved data/model changes.
- Present feature attributions as model-decision associations and use proper causal analysis before making intervention claims.
- Add appropriate Model Armor screening to untrusted inputs and outputs and reject or quarantine policy-violating content.
Explanation
Correct Answer
B:Evaluate and address documented subgroup disparities using appropriate fairness metrics and approved data/model changes. This is the targeted corrective response for model feature bias by subgroup. Aggregate model metrics can conceal uneven errors and harms across different population segments. Verify the diagnosis using: Inspect subgroup confusion matrices, sample sizes, training representation and evaluation slices over time. Confirm success using: The candidate meets subgroup harm thresholds on representative held-out data without hiding global tradeoffs. This matters here because an approval model shows unchanged overall AUC but a recent feature change sharply increases false negatives for a….
Incorrect Answers
A:Version a single feature contract and add automated parity tests to the preprocessing pipeline and serving workflow. This action addresses training-serving feature skew detection, not the failure in the described model feature bias by subgroup case. The targeted correction is: Evaluate and address documented subgroup disparities using appropriate fairness metrics and approved data/model changes. This matters here because an approval model shows unchanged overall AUC but a recent feature change sharply increases false negatives for a….
C:Present feature attributions as model-decision associations and use proper causal analysis before making intervention claims. This action addresses explainability versus causal claim, not the failure in the described model feature bias by subgroup case. The targeted correction is: Evaluate and address documented subgroup disparities using appropriate fairness metrics and approved data/model changes.
D:Add appropriate Model Armor screening to untrusted inputs and outputs and reject or quarantine policy-violating content. This action addresses model armor untrusted content screening, not the failure in the described model feature bias by subgroup case. The targeted correction is: Evaluate and address documented subgroup disparities using appropriate fairness metrics and approved data/model changes.
Question 945
At a Google Cloud analytics team preparing a governed pilot, the following condition occurs: A commerce analyst uses a binary classifier to predict exact revenue per order even though labels are continuous currency amounts. The team must satisfy a reproducible low-code AI assessment. Which configuration or procedure is the most appropriate correction?
- Use a suitable ARIMA_PLUS time-series model with timestamp and data columns and evaluate ML.FORECAST on rolling historical cutoffs.
- Use a BigQuery ML LINEAR_REG or suitable regression model with a numeric target and evaluate continuous-error metrics on held-out records.
- Train an appropriately scaled BigQuery ML KMEANS model and assess cluster cohesion, size and business interpretability.
- Select a permitted task-compatible Gemini or other supported foundation model and benchmark it against simpler candidates.
Explanation
Correct Answer
B:Use a BigQuery ML LINEAR_REG or suitable regression model with a numeric target and evaluate continuous-error metrics on held-out records. This is the targeted corrective response for bigquery ml linear regression for continuous value. Classification labels cannot directly represent unrestricted continuous outcomes; SQL regression models match numeric targets. Verify the diagnosis using: Inspect the target column type, business loss metric, and sample prediction schema alongside model_type in the BigQuery model metadata. Confirm success using: Predictions contain sensible numeric estimates and holdout MAE or RMSE is measured against actual monetary values.
Incorrect Answers
A:Use a suitable ARIMA_PLUS time-series model with timestamp and data columns and evaluate ML.FORECAST on rolling historical cutoffs. This action addresses bigquery arima_plus forecast horizon, not the failure in the described bigquery ml linear regression for continuous value case. The targeted correction is: Use a BigQuery ML LINEAR_REG or suitable regression model with a numeric target and evaluate continuous-error metrics on held-out records.
C:Train an appropriately scaled BigQuery ML KMEANS model and assess cluster cohesion, size and business interpretability. This action addresses bigquery ml kmeans customer segmentation, not the failure in the described bigquery ml linear regression for continuous value case. The targeted correction is: Use a BigQuery ML LINEAR_REG or suitable regression model with a numeric target and evaluate continuous-error metrics on held-out records.
D:Select a permitted task-compatible Gemini or other supported foundation model and benchmark it against simpler candidates. This action addresses model garden task and license selection, not the failure in the described bigquery ml linear regression for continuous value case. The targeted correction is: Use a BigQuery ML LINEAR_REG or suitable regression model with a numeric target and evaluate continuous-error metrics on held-out records.
Question 946
During a regulated MLOps pipeline deployment, the following condition occurs: A data ingestion pipeline changes a numeric feature into a string, but the model-serving path accepts it and silently substitutes missing values. The workflow must be repeatable and protect the last approved production model. Which TWO selections represent the correct investigation and remediation together? Select TWO.
- Enforce a shared typed schema contract and reject incompatible feature versions before training or inference.
- Train only on adequately matured labeled cohorts and monitor recent feature drift separately until outcomes arrive.
- Inspect model-metric artifacts, comparison rules, registry status and deployment dependencies.
- Make the load idempotent with keyed merge or overwrite-per-version semantics and safe retry handling.
- Inspect incoming schema versions, validation logs, casting rules and downstream prediction anomalies.
Explanation
Correct Answers
A:Enforce a shared typed schema contract and reject incompatible feature versions before training or inference. This is the targeted corrective response for serving schema validation component. Silent casting can hide producer errors and produce invalid model inputs across pipeline boundaries. Verify the diagnosis using: Inspect incoming schema versions, validation logs, casting rules and downstream prediction anomalies. Confirm success using: An incompatible payload fails visibly while compatible older versions follow documented conversion rules. This matters here because a data ingestion pipeline changes a numeric feature into a string, but the model-serving path accepts it and….
E:Inspect incoming schema versions, validation logs, casting rules and downstream prediction anomalies. This inspects evidence relevant to serving schema validation component. Silent casting can hide producer errors and produce invalid model inputs across pipeline boundaries. After identifying the cause, the matching correction is: Enforce a shared typed schema contract and reject incompatible feature versions before training or inference. This matters here because a data ingestion pipeline changes a numeric feature into a string, but the model-serving path accepts it and….
Incorrect Answers
B:Train only on adequately matured labeled cohorts and monitor recent feature drift separately until outcomes arrive. This is an action or diagnostic relevant to scheduled retraining on delayed ground truth, not one of the required two steps. In this scenario verify: Inspect incoming schema versions, validation logs, casting rules and downstream prediction anomalies. Then correct: Enforce a shared typed schema contract and reject incompatible feature versions before training or inference.
C:Inspect model-metric artifacts, comparison rules, registry status and deployment dependencies. This is an action or diagnostic relevant to model evaluation gate before deployment, not one of the required two steps. In this scenario verify: Inspect incoming schema versions, validation logs, casting rules and downstream prediction anomalies. Then correct: Enforce a shared typed schema contract and reject incompatible feature versions before training or inference.
D:Make the load idempotent with keyed merge or overwrite-per-version semantics and safe retry handling. This is an action or diagnostic relevant to airflow task idempotency on retry, not one of the required two steps. In this scenario verify: Inspect incoming schema versions, validation logs, casting rules and downstream prediction anomalies. Then correct: Enforce a shared typed schema contract and reject incompatible feature versions before training or inference.
Question 947
During a production ML inference release on Google Cloud, the following condition occurs: A CPU model exceeds its edge latency budget and engineers plan to delete layers without measuring whether quantization affects accuracy. The service must meet its agreed quality and latency requirements. What implementation change resolves the specific issue without substituting an unrelated fix?
- Apply bounded concurrency, idempotent retries with exponential backoff and explicit overload responses.
- Test a supported post-training quantization or optimization path and release only if target quality remains acceptable.
- Schedule and monitor approved Feature Store synchronization at the needed frequency, with stale-feature handling.
- Benchmark a Cloud Run custom container against a managed Agent Platform endpoint and choose for workload fit.
Explanation
Correct Answer
B:Test a supported post-training quantization or optimization path and release only if target quality remains acceptable. This is the targeted corrective response for inference optimization using quantization. Lower precision can improve serving efficiency but may harm numerical accuracy, so it must be validated. Verify the diagnosis using: Compare baseline model size, supported precision, target hardware instructions and accuracy on representative samples. Confirm success using: Latency and memory decline on target hardware while validation degradation stays within the approved threshold. This matters here because a CPU model exceeds its edge latency budget and engineers plan to delete layers without measuring whether….
Incorrect Answers
A:Apply bounded concurrency, idempotent retries with exponential backoff and explicit overload responses. This action addresses predictive serving rate limit and backpressure, not the failure in the described inference optimization using quantization case. The targeted correction is: Test a supported post-training quantization or optimization path and release only if target quality remains acceptable.
C:Schedule and monitor approved Feature Store synchronization at the needed frequency, with stale-feature handling. This action addresses feature store freshness and online sync, not the failure in the described inference optimization using quantization case. The targeted correction is: Test a supported post-training quantization or optimization path and release only if target quality remains acceptable.
D:Benchmark a Cloud Run custom container against a managed Agent Platform endpoint and choose for workload fit. This action addresses cloud run versus managed inference endpoint, not the failure in the described inference optimization using quantization case. The targeted correction is: Test a supported post-training quantization or optimization path and release only if target quality remains acceptable.
Question 948
While reproducing a training orchestration incident in isolated staging, the following condition occurs: A monthly CSV feed begins omitting account IDs, but the pipeline trains and publishes a model without examining null-rate changes. The team requires an idempotent pipeline behavior and traceable evidence. What implementation change resolves the specific issue without substituting an unrelated fix?
- Tune component resource requests/limits and stream or partition embeddings rather than materializing every vector at once.
- Add a data-validation component that quarantines nonconforming inputs and fails downstream training until reviewed.
- Use supported Ray retry policies and durable checkpoints at suitable stage boundaries instead of relying only on worker memory.
- Require inference contract and integration tests with representative requests before promoting each candidate artifact.
Explanation
Correct Answer
B:Add a data-validation component that quarantines nonconforming inputs and fails downstream training until reviewed. This is the targeted corrective response for data validation threshold gate. A schema existing does not imply data quality; pipeline gates must enforce business completeness rules. Verify the diagnosis using: Review schema tests, null-rate distributions, validation thresholds and which pipeline stages respond to failures. Confirm success using: A deliberately malformed dataset is rejected before training while healthy fixtures proceed successfully. This matters here because while reproducing a training orchestration incident in isolated staging, the following condition occurs: A monthly….
Incorrect Answers
A:Tune component resource requests/limits and stream or partition embeddings rather than materializing every vector at once. This action addresses custom kfp component resource limits, not the failure in the described data validation threshold gate case. The targeted correction is: Add a data-validation component that quarantines nonconforming inputs and fails downstream training until reviewed.
C:Use supported Ray retry policies and durable checkpoints at suitable stage boundaries instead of relying only on worker memory. This action addresses ray distributed task failure isolation, not the failure in the described data validation threshold gate case. The targeted correction is: Add a data-validation component that quarantines nonconforming inputs and fails downstream training until reviewed.
D:Require inference contract and integration tests with representative requests before promoting each candidate artifact. This action addresses continuous delivery integration contract tests, not the failure in the described data validation threshold gate case. The targeted correction is: Add a data-validation component that quarantines nonconforming inputs and fails downstream training until reviewed.
Question 949
During a production ML inference release on Google Cloud, the following condition occurs: A private inference service exists with proper invoker IAM, but a VPC client cannot connect because the private route and DNS mapping are missing. The service must meet its agreed quality and latency requirements. Which statement identifies the primary reason for this failure?
- Lower precision can improve serving efficiency but may harm numerical accuracy, so it must be validated.
- Synchronous request limits cannot be solved merely by expanding the model context or retrying indefinitely.
- Identity permission is not proof of network reachability; private serving also requires correct routing and name resolution.
- Endpoint invoker permissions should follow least privilege, rather than expanding access for convenience.
Explanation
Correct Answer
C:Identity permission is not proof of network reachability; private serving also requires correct routing and name resolution. This explains the mechanism behind private vpc routing failure for serving and why the described failure is possible. A discriminating investigation is: Inspect resolved endpoint address, authorized network routing, service access policy and connection logs. The corresponding repair is: Configure the supported private DNS, network route and endpoint access path without granting public ingress. This matters here because a private inference service exists with proper invoker IAM, but a VPC client cannot connect because the private….
Incorrect Answers
A:Lower precision can improve serving efficiency but may harm numerical accuracy, so it must be validated. This principle describes inference optimization using quantization, which is not the cause of the observed private vpc routing failure for serving failure. The controlling explanation is: Identity permission is not proof of network reachability; private serving also requires correct routing and name resolution.
B:Synchronous request limits cannot be solved merely by expanding the model context or retrying indefinitely. This principle describes online prediction timeout and batch fallback, which is not the cause of the observed private vpc routing failure for serving failure. The controlling explanation is: Identity permission is not proof of network reachability; private serving also requires correct routing and name resolution.
D:Endpoint invoker permissions should follow least privilege, rather than expanding access for convenience. This principle describes public endpoint iam oversharing, which is not the cause of the observed private vpc routing failure for serving failure. The controlling explanation is: Identity permission is not proof of network reachability; private serving also requires correct routing and name resolution.
Question 950
During a controlled replay of the data and model handoff to a second engineering team, the following condition occurs: A streaming feature pipeline aggregates transactions by processing arrival time, causing delayed events to land in the wrong customer-hour bucket. The handoff must preserve privacy, lineage and testable behavior. Which TWO selections represent the correct investigation and remediation together? Select TWO.
- Inspect the labeling payload schema, data classification, annotator roles and fields needed to complete the labeling task.
- Rebuild offline training data with point-in-time eligible joins and remove features available only after each prediction moment.
- Configure event-time windowing with an appropriate watermark and explicit late-data correction policy in the Dataflow pipeline.
- Compare source event timestamps, processing timestamps, watermark settings, allowed lateness and actual window outputs from a replay.
- Build a constrained judge rubric with blind repeated comparisons and human-adjudicated calibration on representative examples.
Explanation
Correct Answers
C:Configure event-time windowing with an appropriate watermark and explicit late-data correction policy in the Dataflow pipeline. This is the targeted corrective response for dataflow event-time window correctness. Arrival time and event time are different; event-time windows with lateness rules preserve business-time meaning. Verify the diagnosis using: Compare source event timestamps, processing timestamps, watermark settings, allowed lateness and actual window outputs from a replay. Confirm success using: Out-of-order sample events produce the intended hourly totals and late-data handling follows the documented contract. This matters here because during a controlled replay of the data and model handoff to a second engineering team, the following condition….
D:Compare source event timestamps, processing timestamps, watermark settings, allowed lateness and actual window outputs from a replay. This inspects evidence relevant to dataflow event-time window correctness. Arrival time and event time are different; event-time windows with lateness rules preserve business-time meaning. After identifying the cause, the matching correction is: Configure event-time windowing with an appropriate watermark and explicit late-data correction policy in the Dataflow pipeline. This matters here because during a controlled replay of the data and model handoff to a second engineering team, the following condition….
Incorrect Answers
A:Inspect the labeling payload schema, data classification, annotator roles and fields needed to complete the labeling task. This is an action or diagnostic relevant to data minimization for labeling inputs, not one of the required two steps. In this scenario verify: Compare source event timestamps, processing timestamps, watermark settings, allowed lateness and actual window outputs from a replay. Then correct: Configure event-time windowing with an appropriate watermark and explicit late-data correction policy in the Dataflow pipeline.
B:Rebuild offline training data with point-in-time eligible joins and remove features available only after each prediction moment. This is an action or diagnostic relevant to point-in-time feature leakage prevention, not one of the required two steps. In this scenario verify: Compare source event timestamps, processing timestamps, watermark settings, allowed lateness and actual window outputs from a replay. Then correct: Configure event-time windowing with an appropriate watermark and explicit late-data correction policy in the Dataflow pipeline.
E:Build a constrained judge rubric with blind repeated comparisons and human-adjudicated calibration on representative examples. This is an action or diagnostic relevant to llm-as-a-judge rubric reliability, not one of the required two steps. In this scenario verify: Compare source event timestamps, processing timestamps, watermark settings, allowed lateness and actual window outputs from a replay. Then correct: Configure event-time windowing with an appropriate watermark and explicit late-data correction policy in the Dataflow pipeline.