Google Professional Machine Learning Engineer Practice Test 25
Question 240
Within a jointly maintained Google Cloud data science workspace, the following condition occurs: An analyst runs a long-lived custom Apache Beam streaming service to compute a simple daily SQL join and aggregate over tables already in BigQuery. Collaborators must avoid expanding data access or breaking experiment reproducibility. Select TWO: the decisive diagnostic step and the targeted repair for this specific issue.
- Rebuild offline training data with point-in-time eligible joins and remove features available only after each prediction moment.
- Inspect feature quantiles, physical validity bounds, missingness and training-only fit statistics against independent validation data.
- Implement the batch aggregate in scheduled BigQuery SQL when the required transformation is expressible and fits its SLA.
- Examine source locations, batch cadence, SQL transform complexity, streaming latency requirements and operational burden.
- Pin compatible dependency versions in a reproducible environment specification and validate fresh builds on clean runtimes.
Explanation
Correct Answers
C:Implement the batch aggregate in scheduled BigQuery SQL when the required transformation is expressible and fits its SLA. This is the targeted corrective response for bigquery versus dataflow preparation selection. Choosing a managed SQL engine for warehouse-resident batch joins avoids unnecessary always-on stream processing. Verify the diagnosis using: Examine source locations, batch cadence, SQL transform complexity, streaming latency requirements and operational burden. Confirm success using: The scheduled query produces equivalent daily features with less service orchestration and within the run window.
D:Examine source locations, batch cadence, SQL transform complexity, streaming latency requirements and operational burden. This inspects evidence relevant to bigquery versus dataflow preparation selection. Choosing a managed SQL engine for warehouse-resident batch joins avoids unnecessary always-on stream processing. After identifying the cause, the matching correction is: Implement the batch aggregate in scheduled BigQuery SQL when the required transformation is expressible and fits its SLA.
Incorrect Answers
A: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: Examine source locations, batch cadence, SQL transform complexity, streaming latency requirements and operational burden. Then correct: Implement the batch aggregate in scheduled BigQuery SQL when the required transformation is expressible and fits its SLA.
B:Inspect feature quantiles, physical validity bounds, missingness and training-only fit statistics against independent validation data. This is an action or diagnostic relevant to outlier handling before feature standardization, not one of the required two steps. In this scenario verify: Examine source locations, batch cadence, SQL transform complexity, streaming latency requirements and operational burden. Then correct: Implement the batch aggregate in scheduled BigQuery SQL when the required transformation is expressible and fits its SLA.
E:Pin compatible dependency versions in a reproducible environment specification and validate fresh builds on clean runtimes. This is an action or diagnostic relevant to notebook dependency lock reproducibility, not one of the required two steps. In this scenario verify: Examine source locations, batch cadence, SQL transform complexity, streaming latency requirements and operational burden. Then correct: Implement the batch aggregate in scheduled BigQuery SQL when the required transformation is expressible and fits its SLA.
Question 241
While preparing a production training workflow on Gemini Enterprise Agent Platform, 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. Engineers need an auditable, resource-aware correction to the issue. Which TWO selections represent the correct investigation and remediation together? Select TWO.
- Adapt the training program to the managed data input contract or stage files to an explicit configurable path.
- Inspect model framework, accelerator kernel support, batch sizes, input pipeline speed and per-step cost.
- Inspect region-zone capacity, quota limits, accelerator requirements and scheduling error events.
- Request justified quota or select an available compatible accelerator and region without compromising data residency controls.
- Use appropriately sharded data, parallel decode, caching or prefetching within memory limits to feed the accelerator.
Explanation
Correct Answers
C:Inspect region-zone capacity, quota limits, accelerator requirements and scheduling error events. This inspects evidence relevant to accelerator quota and regional availability. Cloud accelerators have regional quota and stock limits distinct from an application's logical training configuration. After identifying the cause, the matching correction is: Request justified quota or select an available compatible accelerator and region without compromising data residency controls.
D:Request justified quota or select an available compatible accelerator and region without compromising data residency controls. This is the targeted corrective response for accelerator quota and regional availability. Cloud accelerators have regional quota and stock limits distinct from an application's logical training configuration. Verify the diagnosis using: Inspect region-zone capacity, quota limits, accelerator requirements and scheduling error events. Confirm success using: The job schedules successfully with the approved hardware and remains inside budget and residency constraints.
Incorrect Answers
A:Adapt the training program to the managed data input contract or stage files to an explicit configurable path. This is an action or diagnostic relevant to custom training input schema mismatch, not one of the required two steps. In this scenario verify: Inspect region-zone capacity, quota limits, accelerator requirements and scheduling error events. Then correct: Request justified quota or select an available compatible accelerator and region without compromising data residency controls.
B:Inspect model framework, accelerator kernel support, batch sizes, input pipeline speed and per-step cost. This is an action or diagnostic relevant to tpu versus gpu for tensorflow workload, not one of the required two steps. In this scenario verify: Inspect region-zone capacity, quota limits, accelerator requirements and scheduling error events. Then correct: Request justified quota or select an available compatible accelerator and region without compromising data residency controls.
E:Use appropriately sharded data, parallel decode, caching or prefetching within memory limits to feed the accelerator. 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 region-zone capacity, quota limits, accelerator requirements and scheduling error events. Then correct: Request justified quota or select an available compatible accelerator and region without compromising data residency controls.
Question 242
While reproducing a training orchestration incident in isolated staging, the following condition occurs: A freshly trained PyTorch model deploys but rejects the production client's JSON keys because only training unit tests ran in CI. The team requires an idempotent pipeline behavior and traceable evidence. What implementation change resolves the specific issue without substituting an unrelated fix?
- Use a retraining policy with significant trigger thresholds, independent evaluation and staged deployment approval.
- Create immutable versioned input snapshots or partition manifests and record them in training metadata.
- Require inference contract and integration tests with representative requests before promoting each candidate artifact.
- Enforce a shared typed schema contract and reject incompatible feature versions before training or inference.
Explanation
Correct Answer
C:Require inference contract and integration tests with representative requests before promoting each candidate artifact. This is the targeted corrective response for continuous delivery integration contract tests. Training success does not guarantee compatibility with the deployed model's consumer API. Verify the diagnosis using: Inspect request schema, container route behavior, test fixtures and CI stage dependency graph. Confirm success using: A deliberately incompatible payload prevents the release and a compatible fixture completes an end-to-end prediction. This matters here because while reproducing a training orchestration incident in isolated staging, the following condition occurs: A freshly….
Incorrect Answers
A:Use a retraining policy with significant trigger thresholds, independent evaluation and staged deployment approval. This action addresses canary failure feedback into ct policy, not the failure in the described continuous delivery integration contract tests case. The targeted correction is: Require inference contract and integration tests with representative requests before promoting each candidate artifact.
B:Create immutable versioned input snapshots or partition manifests and record them in training metadata. This action addresses data version reproducibility in ct jobs, not the failure in the described continuous delivery integration contract tests case. The targeted correction is: Require inference contract and integration tests with representative requests before promoting each candidate artifact.
D: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 continuous delivery integration contract tests case. The targeted correction is: Require inference contract and integration tests with representative requests before promoting each candidate artifact.
Question 243
Within a jointly maintained Google Cloud data science workspace, the following condition occurs: A model evaluation cell appears to pass only because an old variable remains in kernel memory after cells were executed out of order. Collaborators must avoid expanding data access or breaking experiment reproducibility. What implementation change resolves the specific issue without substituting an unrelated fix?
- Choose a compatible Model Garden model/runtime or perform a validated checkpoint conversion before training.
- Push aggregations into BigQuery or stream bounded samples to the notebook rather than materializing the whole table.
- Refactor hidden state into explicit inputs and functions and add a clean-kernel end-to-end execution gate.
- Build a constrained judge rubric with blind repeated comparisons and human-adjudicated calibration on representative examples.
Explanation
Correct Answer
C:Refactor hidden state into explicit inputs and functions and add a clean-kernel end-to-end execution gate. This is the targeted corrective response for notebook transient state and execution order. Notebook execution can retain stale objects; a clean deterministic run detects hidden state dependencies. Verify the diagnosis using: Restart the runtime, run the notebook top to bottom and compare variable initialization with committed source revision. Confirm success using: A newly started Workbench or Colab Enterprise runtime reproduces the complete experiment without undocumented manual cell order.
Incorrect Answers
A:Choose a compatible Model Garden model/runtime or perform a validated checkpoint conversion before training. This action addresses model garden jax accelerator compatibility, not the failure in the described notebook transient state and execution order case. The targeted correction is: Refactor hidden state into explicit inputs and functions and add a clean-kernel end-to-end execution gate.
B: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 notebook transient state and execution order case. The targeted correction is: Refactor hidden state into explicit inputs and functions and add a clean-kernel end-to-end execution gate.
D:Build a constrained judge rubric with blind repeated comparisons and human-adjudicated calibration on representative examples. This action addresses llm-as-a-judge rubric reliability, not the failure in the described notebook transient state and execution order case. The targeted correction is: Refactor hidden state into explicit inputs and functions and add a clean-kernel end-to-end execution gate.
Question 244
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 evidence most directly establishes what is wrong with this AI or ML workflow?
- Inspect subgroup confusion matrices, sample sizes, training representation and evaluation slices over time.
- Compare live and offline feature SQL, window definitions, sample entity values and prediction-time cutoffs.
- Inspect retrieved document boundaries, instruction hierarchy, malicious test prompts and the application's output actions.
- Review IAM resource bindings, bucket access policies, egress paths and audit records for unauthorized data transfers.
Explanation
Correct Answer
A:Inspect subgroup confusion matrices, sample sizes, training representation and evaluation slices over time. This inspects evidence relevant to model feature bias by subgroup. Aggregate model metrics can conceal uneven errors and harms across different population segments. After identifying the cause, the matching 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….
Incorrect Answers
B:Compare live and offline feature SQL, window definitions, sample entity values and prediction-time cutoffs. This examines training-serving feature skew detection rather than the specific mechanism reported here. It is not the decisive first check for model feature bias by subgroup. The relevant evidence is: Inspect subgroup confusion matrices, sample sizes, training representation and evaluation slices over time.
C:Inspect retrieved document boundaries, instruction hierarchy, malicious test prompts and the application's output actions. This examines prompt injection through retrieved documents rather than the specific mechanism reported here. It is not the decisive first check for model feature bias by subgroup. The relevant evidence is: Inspect subgroup confusion matrices, sample sizes, training representation and evaluation slices over time.
D:Review IAM resource bindings, bucket access policies, egress paths and audit records for unauthorized data transfers. This examines training data exfiltration through broad access rather than the specific mechanism reported here. It is not the decisive first check for model feature bias by subgroup. The relevant evidence is: Inspect subgroup confusion matrices, sample sizes, training representation and evaluation slices over time.
Question 245
During a staged restoration test at a secondary serving deployment, the following condition occurs: A PyTorch model container builds successfully but the serving platform marks its deployed revision unhealthy due to a missing health-check route and incorrect listener port. The team must prove an actual user-visible improvement without weakening access controls. Which TWO choices together provide the decisive diagnostic evidence and the most direct acceptance test? Select TWO.
- The selected service meets p95 latency and payload requirements with lower idle cost and reliable scaling.
- Measure bandwidth, network availability, latency deadline, edge hardware limits and model compression compatibility.
- Inspect startup logs, HTTP listening port, health-check expectations and request and response schema.
- 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 Answers
C:Inspect startup logs, HTTP listening port, health-check expectations and request and response schema. This inspects evidence relevant to inference container health interface mismatch. A valid container image is not necessarily compatible with an inference service's HTTP startup and health contract. After identifying the cause, the matching correction is: Implement the required serving contract and make the container report readiness only after loading model assets.
D:A known prediction request succeeds through the deployed endpoint and readiness reflects successful initialization. This is specific acceptance evidence for inference container health interface mismatch. A valid container image is not necessarily compatible with an inference service's HTTP startup and health contract. It confirms the controlled correction: Implement the required serving contract and make the container report readiness only after loading model assets.
Incorrect Answers
A:The selected service meets p95 latency and payload requirements with lower idle cost and reliable scaling. This relates to cloud run versus managed inference endpoint but does not establish the requested before-and-after evidence. The actual diagnostic is: Inspect startup logs, HTTP listening port, health-check expectations and request and response schema. The specific acceptance result is: A known prediction request succeeds through the deployed endpoint and readiness reflects successful initialization.
B:Measure bandwidth, network availability, latency deadline, edge hardware limits and model compression compatibility. This relates to edge versus cloud inference network constraints but does not establish the requested before-and-after evidence. The actual diagnostic is: Inspect startup logs, HTTP listening port, health-check expectations and request and response schema. The specific acceptance result is: A known prediction request succeeds through the deployed endpoint and readiness reflects successful initialization.
E:Authorized lookups return expected recent values and freshness alarms fire when synchronization falls behind. This relates to feature store freshness and online sync but does not establish the requested before-and-after evidence. The actual diagnostic is: Inspect startup logs, HTTP listening port, health-check expectations and request and response schema. The specific acceptance result is: A known prediction request succeeds through the deployed endpoint and readiness reflects successful initialization.
Question 246
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 implementation change resolves the specific issue without substituting an unrelated fix?
- Configure event-time windowing with an appropriate watermark and explicit late-data correction policy in the Dataflow pipeline.
- Choose a compatible Model Garden model/runtime or perform a validated checkpoint conversion before training.
- Standardize stable entity identifiers and validated feature definitions across training and online serving pathways.
- Implement the batch aggregate in scheduled BigQuery SQL when the required transformation is expressible and fits its SLA.
Explanation
Correct Answer
C:Standardize stable entity identifiers and validated feature definitions across training and online serving pathways. This is the targeted corrective response for feature store online and offline entity alignment. Feature stores rely on consistent entity keys; incompatible identifiers lead to silent training-serving misalignment. Verify the diagnosis using: Compare entity IDs, key types, online feature-view definitions, offline joins and records for the same verified customer. Confirm success using: Known records retrieve corresponding historical and current feature values for the intended entity without mismatched accounts.
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 feature store online and offline entity alignment case. The targeted correction is: Standardize stable entity identifiers and validated feature definitions across training and online serving pathways.
B:Choose a compatible Model Garden model/runtime or perform a validated checkpoint conversion before training. This action addresses model garden jax accelerator compatibility, not the failure in the described feature store online and offline entity alignment case. The targeted correction is: Standardize stable entity identifiers and validated feature definitions across training and online serving pathways.
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 feature store online and offline entity alignment case. The targeted correction is: Standardize stable entity identifiers and validated feature definitions across training and online serving pathways.
Question 247
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. Which underlying engineering principle explains why the original approach fails?
- Unlimited retries amplify overload; backpressure and controlled concurrency are needed along with scaling.
- Coercing identifier types can change keys, particularly when numeric-looking strings have meaningful leading zeros.
- Per-request routing can mix treatments within a user and contaminate a longitudinal A/B experiment.
- Network distance can dominate inference response times when the actual model execution is already short.
Explanation
Correct Answer
D:Network distance can dominate inference response times when the actual model execution is already short. This explains the mechanism behind serving region proximity versus model compute and why the described failure is possible. A discriminating investigation is: Decompose client network, TLS, request transport, model compute and response times by user geography. The corresponding repair is: Select supported regional placement or routing close to users while respecting residency and model availability.
Incorrect Answers
A:Unlimited retries amplify overload; backpressure and controlled concurrency are needed along with scaling. This principle describes predictive serving rate limit and backpressure, which is not the cause of the observed serving region proximity versus model compute failure. The controlling explanation is: Network distance can dominate inference response times when the actual model execution is already short.
B:Coercing identifier types can change keys, particularly when numeric-looking strings have meaningful leading zeros. This principle describes feature online entity key mismatch, which is not the cause of the observed serving region proximity versus model compute failure. The controlling explanation is: Network distance can dominate inference response times when the actual model execution is already short.
C:Per-request routing can mix treatments within a user and contaminate a longitudinal A/B experiment. This principle describes a/b assignment consistency for user journeys, which is not the cause of the observed serving region proximity versus model compute failure. The controlling explanation is: Network distance can dominate inference response times when the actual model execution is already short.
Question 248
At a Google Cloud analytics team preparing a governed pilot, the following condition occurs: A high-volume classification API uses a slow reasoning-optimized foundation model for simple short labels even though a smaller supported model may suffice. The team must satisfy a reproducible low-code AI assessment. Which result is the strongest postchange evidence of success?
- A revised policy becomes answerable after document indexing without retraining, and retrieved citations match the active version.
- A controlled raw record yields the same transformed feature representation for evaluation and production SQL inference.
- Classification F1 remains above the acceptance threshold while end-to-end latency and cost per thousand requests fall.
- The chosen version meets task quality, inference cost and license rules on a documented representative evaluation set.
Explanation
Correct Answer
C:Classification F1 remains above the acceptance threshold while end-to-end latency and cost per thousand requests fall. This is specific acceptance evidence for gemini model choice for latency versus complexity. Production model selection is a task-quality and cost tradeoff; using the largest model by default is unnecessary. It confirms the controlled correction: Benchmark a supported lower-latency Gemini variant against the current model under representative traffic before routing requests.
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 gemini model choice for latency versus complexity issue. The direct test is: Classification F1 remains above the acceptance threshold while end-to-end latency and cost per thousand requests fall.
B:A controlled raw record yields the same transformed feature representation for evaluation and production SQL inference. This result could demonstrate success for bigquery ml transform serving parity but not for the stated gemini model choice for latency versus complexity issue. The direct test is: Classification F1 remains above the acceptance threshold while end-to-end latency and cost per thousand requests fall.
D:The chosen version meets task quality, inference cost and license rules on a documented representative evaluation set. This result could demonstrate success for model garden task and license selection but not for the stated gemini model choice for latency versus complexity issue. The direct test is: Classification F1 remains above the acceptance threshold while end-to-end latency and cost per thousand requests fall.
Question 249
Within a jointly maintained Google Cloud data science workspace, the following condition occurs: A distributed preprocessing job joins two huge feature tables, but one NULL-like customer category contains most rows and one reducer becomes the bottleneck. Collaborators must avoid expanding data access or breaking experiment reproducibility. Which technical explanation best accounts for the reported behavior?
- Metrics from different data distributions are not directly comparable evidence of model improvement.
- Choosing a managed SQL engine for warehouse-resident batch joins avoids unnecessary always-on stream processing.
- Partition pruning avoids irrelevant I/O when predicates match the storage partitioning scheme.
- Skew concentrates otherwise distributed work on a single hot key, which extra workers alone may not resolve.
Explanation
Correct Answer
D:Skew concentrates otherwise distributed work on a single hot key, which extra workers alone may not resolve. This explains the mechanism behind managed spark join-key skew repair and why the described failure is possible. A discriminating investigation is: Inspect join-key frequency distributions, per-task runtime, shuffle spill and the output cardinality before optimization. The corresponding repair is: Handle the dominant key separately or repartition and salt the skewed join while validating join semantics and counts.
Incorrect Answers
A:Metrics from different data distributions are not directly comparable evidence of model improvement. This principle describes experiment metric comparison with mismatched splits, which is not the cause of the observed managed spark join-key skew repair failure. The controlling explanation is: Skew concentrates otherwise distributed work on a single hot key, which extra workers alone may not resolve.
B:Choosing a managed SQL engine for warehouse-resident batch joins avoids unnecessary always-on stream processing. This principle describes bigquery versus dataflow preparation selection, which is not the cause of the observed managed spark join-key skew repair failure. The controlling explanation is: Skew concentrates otherwise distributed work on a single hot key, which extra workers alone may not resolve.
C:Partition pruning avoids irrelevant I/O when predicates match the storage partitioning scheme. This principle describes bigquery partition filter cost optimization, which is not the cause of the observed managed spark join-key skew repair failure. The controlling explanation is: Skew concentrates otherwise distributed work on a single hot key, which extra workers alone may not resolve.
Question 250
During model-training validation on a Google Cloud experimental project, the following condition occurs: An engineer runs an unbounded grid across learning rates, depths and regularization values, using expensive accelerators without a metric-driven stopping rule. The change should preserve a meaningful independent model-quality comparison. Which diagnostic check should be performed FIRST to isolate the described cause?
- Review parameter ranges, total trial count, training budget and the metric used to rank completed experiments.
- Compare training and validation curves, learning-rate schedule and the checkpoint corresponding to minimum validation loss.
- Inspect region-zone capacity, quota limits, accelerator requirements and scheduling error events.
- Inspect shutdown events, checkpoint directory, durable artifact transfers and restoration logic after simulated interruptions.
Explanation
Correct Answer
A:Review parameter ranges, total trial count, training budget and the metric used to rank completed experiments. This inspects evidence relevant to hyperparameter search budget limit. Search quality depends on comparable objectives and resource limits, not exhaustive combinations at any cost. After identifying the cause, the matching correction is: Use bounded hyperparameter tuning with an explicit objective metric, search space, concurrency and early stopping.
Incorrect Answers
B:Compare training and validation curves, learning-rate schedule and the checkpoint corresponding to minimum validation loss. This examines early stopping on validation divergence rather than the specific mechanism reported here. It is not the decisive first check for hyperparameter search budget limit. The relevant evidence is: Review parameter ranges, total trial count, training budget and the metric used to rank completed experiments.
C:Inspect region-zone capacity, quota limits, accelerator requirements and scheduling error events. This examines accelerator quota and regional availability rather than the specific mechanism reported here. It is not the decisive first check for hyperparameter search budget limit. The relevant evidence is: Review parameter ranges, total trial count, training budget and the metric used to rank completed experiments.
D:Inspect shutdown events, checkpoint directory, durable artifact transfers and restoration logic after simulated interruptions. This examines training checkpoint persistence across interruptions rather than the specific mechanism reported here. It is not the decisive first check for hyperparameter search budget limit. The relevant evidence is: Review parameter ranges, total trial count, training budget and the metric used to rank completed experiments.