Google Professional Machine Learning Engineer Practice Test 55
Question 540
During a staged restoration test at a secondary serving deployment, the following condition occurs: An online recommender test randomly selects a model for every click, exposing one person to both variants and confusing conversion attribution. The team must prove an actual user-visible improvement without weakening access controls. Which TWO complementary choices jointly establish the evidence and appropriate correction? Select TWO.
- Inspect experiment assignment IDs, routing hashes, variant weights and per-user exposure histories.
- Add guarded traffic shifts and automatic rollback to the retained healthy revision when threshold alarms trip.
- Set a justified warm minimum or scheduled capacity and balance initialization overhead against idle cost.
- Compare tokenizer hash, normalization rules, token IDs and inference schema against the trained model artifacts.
- Use stable cohort-level assignment and record model identity on requests and measured outcomes.
Explanation
Correct Answers
A:Inspect experiment assignment IDs, routing hashes, variant weights and per-user exposure histories. This inspects evidence relevant to a/b assignment consistency for user journeys. Per-request routing can mix treatments within a user and contaminate a longitudinal A/B experiment. After identifying the cause, the matching correction is: Use stable cohort-level assignment and record model identity on requests and measured outcomes. This matters here because during a staged restoration test at a secondary serving deployment, the following condition occurs: An online….
E:Use stable cohort-level assignment and record model identity on requests and measured outcomes. This is the targeted corrective response for a/b assignment consistency for user journeys. Per-request routing can mix treatments within a user and contaminate a longitudinal A/B experiment. Verify the diagnosis using: Inspect experiment assignment IDs, routing hashes, variant weights and per-user exposure histories. Confirm success using: Repeated requests from the same test user stay on one variant and conversion comparisons preserve randomized balance. This matters here because during a staged restoration test at a secondary serving deployment, the following condition occurs: An online….
Incorrect Answers
B:Add guarded traffic shifts and automatic rollback to the retained healthy revision when threshold alarms trip. This is an action or diagnostic relevant to model traffic rollback on high error rate, not one of the required two steps. In this scenario verify: Inspect experiment assignment IDs, routing hashes, variant weights and per-user exposure histories. Then correct: Use stable cohort-level assignment and record model identity on requests and measured outcomes.
C:Set a justified warm minimum or scheduled capacity and balance initialization overhead against idle cost. This is an action or diagnostic relevant to autoscaling minimum capacity versus cold load, not one of the required two steps. In this scenario verify: Inspect experiment assignment IDs, routing hashes, variant weights and per-user exposure histories. Then correct: Use stable cohort-level assignment and record model identity on requests and measured outcomes.
D:Compare tokenizer hash, normalization rules, token IDs and inference schema against the trained model artifacts. This is an action or diagnostic relevant to inference preprocessing tokenization mismatch, not one of the required two steps. In this scenario verify: Inspect experiment assignment IDs, routing hashes, variant weights and per-user exposure histories. Then correct: Use stable cohort-level assignment and record model identity on requests and measured outcomes.
Question 541
During a staged restoration test at a secondary serving deployment, the following condition occurs: A deployment grants every project member broad model prediction permissions merely to make a limited partner integration work. The team must prove an actual user-visible improvement without weakening access controls. Which TWO selections form an appropriate before-and-after verification pair? Select TWO.
- Approved in-network clients can invoke predictions while unapproved public network paths cannot reach the service.
- Previously missed customer IDs retrieve the intended feature values without collapsing distinct leading-zero identifiers.
- Compare baseline model size, supported precision, target hardware instructions and accuracy on representative samples.
- Examine effective IAM bindings, invocation audit logs and the intended partner service account identity.
- The approved integration continues to work while unrelated accounts receive permission denied on inference calls.
Explanation
Correct Answers
D:Examine effective IAM bindings, invocation audit logs and the intended partner service account identity. This inspects evidence relevant to public endpoint iam oversharing. Endpoint invoker permissions should follow least privilege, rather than expanding access for convenience. After identifying the cause, the matching correction is: Restrict invoke access to the necessary principal and endpoint, using appropriate short-lived authentication.
E:The approved integration continues to work while unrelated accounts receive permission denied on inference calls. This is specific acceptance evidence for public endpoint iam oversharing. Endpoint invoker permissions should follow least privilege, rather than expanding access for convenience. It confirms the controlled correction: Restrict invoke access to the necessary principal and endpoint, using appropriate short-lived authentication.
Incorrect Answers
A:Approved in-network clients can invoke predictions while unapproved public network paths cannot reach the service. This relates to private endpoint for regulated inference traffic but does not establish the requested before-and-after evidence. The actual diagnostic is: Examine effective IAM bindings, invocation audit logs and the intended partner service account identity. The specific acceptance result is: The approved integration continues to work while unrelated accounts receive permission denied on inference calls.
B:Previously missed customer IDs retrieve the intended feature values without collapsing distinct leading-zero identifiers. This relates to feature online entity key mismatch but does not establish the requested before-and-after evidence. The actual diagnostic is: Examine effective IAM bindings, invocation audit logs and the intended partner service account identity. The specific acceptance result is: The approved integration continues to work while unrelated accounts receive permission denied on inference calls.
C:Compare baseline model size, supported precision, target hardware instructions and accuracy on representative samples. This relates to inference optimization using quantization but does not establish the requested before-and-after evidence. The actual diagnostic is: Examine effective IAM bindings, invocation audit logs and the intended partner service account identity. The specific acceptance result is: The approved integration continues to work while unrelated accounts receive permission denied on inference calls.
Question 542
During a regulated MLOps pipeline deployment, the following condition occurs: An operations team retrains its classifier every hour based on a clock even when data have not changed, wasting compute and risking unreviewed model swaps. The workflow must be repeatable and protect the last approved production model. Which targeted engineering change most directly addresses the failure?
- Retain the last healthy model and require canary validation before switching all prediction traffic.
- Enforce a shared typed schema contract and reject incompatible feature versions before training or inference.
- Promote explicit immutable model and image versions with versioned release manifests and review gates.
- Trigger candidate retraining only when justified evidence and sufficient labeled data are available, with evaluation gates.
Explanation
Correct Answer
D:Trigger candidate retraining only when justified evidence and sufficient labeled data are available, with evaluation gates. This is the targeted corrective response for data drift triggered retraining policy. Retraining on a timer alone does not imply better models and may use incomplete or unrepresentative labels. Verify the diagnosis using: Inspect training data freshness, drift signals, label availability, business cadence and automated promotion rules. Confirm success using: No-op periods create no unnecessary models and validated drift events produce candidates requiring approval.
Incorrect Answers
A:Retain the last healthy model and require canary validation before switching all prediction traffic. This action addresses continuous training rollback protection, not the failure in the described data drift triggered retraining policy case. The targeted correction is: Trigger candidate retraining only when justified evidence and sufficient labeled data are available, with evaluation gates.
B:Enforce a shared typed schema contract and reject incompatible feature versions before training or inference. This action addresses serving schema validation component, not the failure in the described data drift triggered retraining policy case. The targeted correction is: Trigger candidate retraining only when justified evidence and sufficient labeled data are available, with evaluation gates.
C:Promote explicit immutable model and image versions with versioned release manifests and review gates. This action addresses cloud build release version promotion, not the failure in the described data drift triggered retraining policy case. The targeted correction is: Trigger candidate retraining only when justified evidence and sufficient labeled data are available, with evaluation gates.
Question 543
While reproducing a training orchestration incident in isolated staging, the following condition occurs: A build pipeline deploys an arbitrary model tag without tying it to the source revision and the approved evaluation report. The team requires an idempotent pipeline behavior and traceable evidence. Which outcome most directly demonstrates a successful, measurable fix?
- The deployed model and container can be traced to a particular approved commit and evaluation result.
- No-op periods create no unnecessary models and validated drift events produce candidates requiring approval.
- An audit can resolve model version to exact source data and processing code without guessing by timestamps.
- Training manifests exclude unlabeled recent transactions and measured metrics use verified ground truth.
Explanation
Correct Answer
A:The deployed model and container can be traced to a particular approved commit and evaluation result. This is specific acceptance evidence for cloud build release version promotion. Unpinned tags do not establish a trustworthy relationship between code, model artifacts and production rollout. It confirms the controlled correction: Promote explicit immutable model and image versions with versioned release manifests and review gates. This matters here because while reproducing a training orchestration incident in isolated staging, the following condition occurs: A build….
Incorrect Answers
B:No-op periods create no unnecessary models and validated drift events produce candidates requiring approval. This result could demonstrate success for data drift triggered retraining policy but not for the stated cloud build release version promotion issue. The direct test is: The deployed model and container can be traced to a particular approved commit and evaluation result. This matters here because while reproducing a training orchestration incident in isolated staging, the following condition occurs: A build….
C:An audit can resolve model version to exact source data and processing code without guessing by timestamps. This result could demonstrate success for model feature lineage for audit but not for the stated cloud build release version promotion issue. The direct test is: The deployed model and container can be traced to a particular approved commit and evaluation result.
D:Training manifests exclude unlabeled recent transactions and measured metrics use verified ground truth. This result could demonstrate success for scheduled retraining on delayed ground truth but not for the stated cloud build release version promotion issue. The direct test is: The deployed model and container can be traced to a particular approved commit and evaluation result.
Question 544
During an independent acceptance rehearsal of a customer-facing AI capability, the following condition occurs: A retailer has no trustworthy segment labels but attempts a supervised classifier with synthetic classes supplied by the sales team. The owners require measurable quality and controlled operating cost. Which result is the strongest postchange evidence of success?
- Stable clusters reveal coherent customer behavior without pretending an unsupervised assignment is known supervised ground truth.
- Twenty-one future daily estimates are produced with reported prediction intervals and error is scored on forward dates.
- Predictions contain sensible numeric estimates and holdout MAE or RMSE is measured against actual monetary values.
- A revised policy becomes answerable after document indexing without retraining, and retrieved citations match the active version.
Explanation
Correct Answer
A:Stable clusters reveal coherent customer behavior without pretending an unsupervised assignment is known supervised ground truth. This is specific acceptance evidence for bigquery ml kmeans customer segmentation. K-means discovers unlabeled similarity groups; fabricating labels would make supervised accuracy meaningless. It confirms the controlled correction: Train an appropriately scaled BigQuery ML KMEANS model and assess cluster cohesion, size and business interpretability.
Incorrect Answers
B:Twenty-one future daily estimates are produced with reported prediction intervals and error is scored on forward dates. This result could demonstrate success for bigquery arima_plus forecast horizon but not for the stated bigquery ml kmeans customer segmentation issue. The direct test is: Stable clusters reveal coherent customer behavior without pretending an unsupervised assignment is known supervised ground truth.
C:Predictions contain sensible numeric estimates and holdout MAE or RMSE is measured against actual monetary values. This result could demonstrate success for bigquery ml linear regression for continuous value but not for the stated bigquery ml kmeans customer segmentation issue. The direct test is: Stable clusters reveal coherent customer behavior without pretending an unsupervised assignment is known supervised ground truth.
D: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 kmeans customer segmentation issue. The direct test is: Stable clusters reveal coherent customer behavior without pretending an unsupervised assignment is known supervised ground truth.
Question 545
During model-training validation on a Google Cloud experimental project, the following condition occurs: An eight-GPU job scales only slightly beyond two GPUs because each small batch requires excessive cross-worker gradient synchronization. The change should preserve a meaningful independent model-quality comparison. Which TWO complementary choices jointly establish the evidence and appropriate correction? Select TWO.
- Inspect all-reduce time, network bandwidth, effective batch size, worker stragglers and samples per second.
- Use grounded retrieval for frequently changing policy facts and reserve fine-tuning for measured durable behavior adaptation.
- Profile data-loader wait time, GCS read latency, prefetch queues and device utilization on a representative epoch.
- Tune supported distributed batch sizing, communication strategy or placement after measuring the bottleneck.
- Request justified quota or select an available compatible accelerator and region without compromising data residency controls.
Explanation
Correct Answers
A:Inspect all-reduce time, network bandwidth, effective batch size, worker stragglers and samples per second. This inspects evidence relevant to distributed gradient communication bottleneck. Distributed speedups depend on communication-to-compute ratio, not merely the number of accelerators allocated. After identifying the cause, the matching correction is: Tune supported distributed batch sizing, communication strategy or placement after measuring the bottleneck.
D:Tune supported distributed batch sizing, communication strategy or placement after measuring the bottleneck. This is the targeted corrective response for distributed gradient communication bottleneck. Distributed speedups depend on communication-to-compute ratio, not merely the number of accelerators allocated. Verify the diagnosis using: Inspect all-reduce time, network bandwidth, effective batch size, worker stragglers and samples per second. Confirm success using: Scale efficiency improves and model quality remains stable without synchronization stalls dominating every step.
Incorrect Answers
B:Use grounded retrieval for frequently changing policy facts and reserve fine-tuning for measured durable behavior adaptation. This is an action or diagnostic relevant to foundation model prompt versus fine-tuning, not one of the required two steps. In this scenario verify: Inspect all-reduce time, network bandwidth, effective batch size, worker stragglers and samples per second. Then correct: Tune supported distributed batch sizing, communication strategy or placement after measuring the bottleneck.
C:Profile data-loader wait time, GCS read latency, prefetch queues and device utilization on a representative epoch. This is an action or diagnostic relevant to accelerator starvation from slow input pipeline, not one of the required two steps. In this scenario verify: Inspect all-reduce time, network bandwidth, effective batch size, worker stragglers and samples per second. Then correct: Tune supported distributed batch sizing, communication strategy or placement after measuring the bottleneck.
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 all-reduce time, network bandwidth, effective batch size, worker stragglers and samples per second. Then correct: Tune supported distributed batch sizing, communication strategy or placement after measuring the bottleneck.
Question 546
During model-training validation on a Google Cloud experimental project, the following condition occurs: An accelerator training job using half-precision arithmetic produces NaNs after a learning-rate increase although its full-precision baseline converges. The change should preserve a meaningful independent model-quality comparison. Which result is the strongest postchange evidence of success?
- A controlled interruption resumes near the saved step without restarting training from zero.
- Scale efficiency improves and model quality remains stable without synchronization stalls dominating every step.
- Training loss remains finite under representative batches while memory and throughput benefits are measured.
- Every evaluation feature is available at its simulated prediction time and later-period scores remain stable.
Explanation
Correct Answer
C:Training loss remains finite under representative batches while memory and throughput benefits are measured. This is specific acceptance evidence for mixed precision overflow diagnosis. Reduced numeric precision narrows representable ranges and can underflow or overflow without proper scaling. It confirms the controlled correction: Use a supported automatic mixed-precision and loss-scaling strategy or return unstable operations to suitable precision. This matters here because during model-training validation on a Google Cloud experimental project, the following condition occurs: An….
Incorrect Answers
A:A controlled interruption resumes near the saved step without restarting training from zero. This result could demonstrate success for training checkpoint persistence across interruptions but not for the stated mixed precision overflow diagnosis issue. The direct test is: Training loss remains finite under representative batches while memory and throughput benefits are measured.
B: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 mixed precision overflow diagnosis issue. The direct test is: Training loss remains finite under representative batches while memory and throughput benefits are measured.
D:Every evaluation feature is available at its simulated prediction time and later-period scores remain stable. This result could demonstrate success for temporal training label leakage in rolling windows but not for the stated mixed precision overflow diagnosis issue. The direct test is: Training loss remains finite under representative batches while memory and throughput benefits are measured.
Question 547
During a postdeployment AI safety and reliability review, the following condition occurs: A deployed risk model receives income values far outside the training distribution but the dashboard checks only endpoint availability. The model owner must provide repeatable evidence rather than a cosmetic dashboard change. Which technical explanation best accounts for the reported behavior?
- HTTP health is not evidence that incoming feature distributions still resemble training data.
- Normal recurring seasonality can resemble drift against an unsegmented static baseline.
- Fluent text or semantic similarity is not proof an answer is supported by the retrieved source material.
- Screening helps detect malicious instructions or sensitive content crossing the model trust boundary.
Explanation
Correct Answer
A:HTTP health is not evidence that incoming feature distributions still resemble training data. This explains the mechanism behind input data distribution drift alarm and why the described failure is possible. A discriminating investigation is: Inspect live feature histograms, training baselines, data-capture completeness and drift metric thresholds. The corresponding repair is: Configure Agent Platform Model Monitoring on relevant features with representative baseline and actionable drift alerts.
Incorrect Answers
B:Normal recurring seasonality can resemble drift against an unsegmented static baseline. This principle describes model monitoring false-alert seasonality, which is not the cause of the observed input data distribution drift alarm failure. The controlling explanation is: HTTP health is not evidence that incoming feature distributions still resemble training data.
C:Fluent text or semantic similarity is not proof an answer is supported by the retrieved source material. This principle describes generative ai factual grounding evaluation, which is not the cause of the observed input data distribution drift alarm failure. The controlling explanation is: HTTP health is not evidence that incoming feature distributions still resemble training data.
D:Screening helps detect malicious instructions or sensitive content crossing the model trust boundary. This principle describes model armor untrusted content screening, which is not the cause of the observed input data distribution drift alarm failure. The controlling explanation is: HTTP health is not evidence that incoming feature distributions still resemble training data.
Question 548
During a controlled replay of the data and model handoff to a second engineering team, the following condition occurs: A prototype notebook commits a service-account JSON private key beside its feature code so colleagues can open Cloud Storage directly. The handoff must preserve privacy, lineage and testable behavior. Which configuration or procedure is the most appropriate correction?
- Configure event-time windowing with an appropriate watermark and explicit late-data correction policy in the Dataflow pipeline.
- Revoke and rotate the exposed key and use managed workload identity or approved short-lived credentials with least privilege.
- Refactor hidden state into explicit inputs and functions and add a clean-kernel end-to-end execution gate.
- Push aggregations into BigQuery or stream bounded samples to the notebook rather than materializing the whole table.
Explanation
Correct Answer
B:Revoke and rotate the exposed key and use managed workload identity or approved short-lived credentials with least privilege. This is the targeted corrective response for workbench credentials embedded in repository. Long-lived secrets copied into notebooks or Git can be reused outside intended workloads and must be rotated. Verify the diagnosis using: Inspect repository history, mounted credentials, authorized identities and any logs of secret access from the shared project. Confirm success using: The notebook can access allowed buckets without a stored private key and the old key no longer authenticates. 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:Configure event-time windowing with an appropriate watermark and explicit late-data correction policy in the Dataflow pipeline. This action addresses dataflow event-time window correctness, not the failure in the described workbench credentials embedded in repository case. The targeted correction is: Revoke and rotate the exposed key and use managed workload identity or approved short-lived credentials with least privilege.
C: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 workbench credentials embedded in repository case. The targeted correction is: Revoke and rotate the exposed key and use managed workload identity or approved short-lived credentials with least privilege. This matters here because during a controlled replay of the data and model handoff to a second engineering team, the following condition….
D:Push aggregations into BigQuery or stream bounded samples to the notebook rather than materializing the whole table. This action addresses notebook memory exhaustion on full data extraction, not the failure in the described workbench credentials embedded in repository case. The targeted correction is: Revoke and rotate the exposed key and use managed workload identity or approved short-lived credentials with least privilege. This matters here because during a controlled replay of the data and model handoff to a second engineering team, the following condition….
Question 549
During a controlled replay of the data and model handoff to a second engineering team, the following condition occurs: A credit service looks up a feature by account number online but the training table keys records by customer identifier, causing inconsistent histories. The handoff must preserve privacy, lineage and testable behavior. What should the engineer inspect before selecting a correction?
- Inspect join-key frequency distributions, per-task runtime, shuffle spill and the output cardinality before optimization.
- Inspect feature quantiles, physical validity bounds, missingness and training-only fit statistics against independent validation data.
- Compare entity IDs, key types, online feature-view definitions, offline joins and records for the same verified customer.
- Inspect query result sizes, runtime memory, pandas allocations and which aggregations can run in BigQuery SQL.
Explanation
Correct Answer
C:Compare entity IDs, key types, online feature-view definitions, offline joins and records for the same verified customer. This inspects evidence relevant to feature store online and offline entity alignment. Feature stores rely on consistent entity keys; incompatible identifiers lead to silent training-serving misalignment. After identifying the cause, the matching correction is: Standardize stable entity identifiers and validated feature definitions across training and online serving pathways.
Incorrect Answers
A:Inspect join-key frequency distributions, per-task runtime, shuffle spill and the output cardinality before optimization. This examines managed spark join-key skew repair rather than the specific mechanism reported here. It is not the decisive first check for feature store online and offline entity alignment. The relevant evidence is: Compare entity IDs, key types, online feature-view definitions, offline joins and records for the same verified customer.
B:Inspect feature quantiles, physical validity bounds, missingness and training-only fit statistics against independent validation data. This examines outlier handling before feature standardization rather than the specific mechanism reported here. It is not the decisive first check for feature store online and offline entity alignment. The relevant evidence is: Compare entity IDs, key types, online feature-view definitions, offline joins and records for the same verified customer.
D:Inspect query result sizes, runtime memory, pandas allocations and which aggregations can run in BigQuery SQL. This examines notebook memory exhaustion on full data extraction rather than the specific mechanism reported here. It is not the decisive first check for feature store online and offline entity alignment. The relevant evidence is: Compare entity IDs, key types, online feature-view definitions, offline joins and records for the same verified customer.
Question 550
While reproducing a training orchestration incident in isolated staging, the following condition occurs: A CI/CD workflow always deploys the most recent checkpoint even when its validation F1 is below the previous approved baseline. The team requires an idempotent pipeline behavior and traceable evidence. Which outcome most directly demonstrates a successful, measurable fix?
- A deliberately malformed dataset is rejected before training while healthy fixtures proceed successfully.
- An intentionally weak candidate cannot deploy, whereas an approved passing model can proceed with traceable evidence.
- A deliberately incompatible payload prevents the release and a compatible fixture completes an end-to-end prediction.
- An audit can resolve model version to exact source data and processing code without guessing by timestamps.
Explanation
Correct Answer
B:An intentionally weak candidate cannot deploy, whereas an approved passing model can proceed with traceable evidence. This is specific acceptance evidence for model evaluation gate before deployment. Evaluation without a blocking release decision cannot protect production from quality regressions. It confirms the controlled correction: Require candidate evaluation against a governed baseline with an enforced acceptance threshold before promotion. This matters here because while reproducing a training orchestration incident in isolated staging, the following condition occurs: A CI/CD….
Incorrect Answers
A:A deliberately malformed dataset is rejected before training while healthy fixtures proceed successfully. This result could demonstrate success for data validation threshold gate but not for the stated model evaluation gate before deployment issue. The direct test is: An intentionally weak candidate cannot deploy, whereas an approved passing model can proceed with traceable evidence.
C:A deliberately incompatible payload prevents the release and a compatible fixture completes an end-to-end prediction. This result could demonstrate success for continuous delivery integration contract tests but not for the stated model evaluation gate before deployment issue. The direct test is: An intentionally weak candidate cannot deploy, whereas an approved passing model can proceed with traceable evidence.
D:An audit can resolve model version to exact source data and processing code without guessing by timestamps. This result could demonstrate success for model feature lineage for audit but not for the stated model evaluation gate before deployment issue. The direct test is: An intentionally weak candidate cannot deploy, whereas an approved passing model can proceed with traceable evidence.