Amazon AWS Certified Machine Learning Engineer - Associate MLA-C01 Practice Test 102
Question 1010
While testing a second region for recoverable deployment of the same ML service, the following condition is observed: An edge vision model is deployed to a constrained device with a framework binary and operators assume any model artifact runs on its target accelerator. The deployment must be reproducible without manual console-only fixes. What postchange observation provides the strongest acceptance evidence?
- The compiled artifact runs on the target device with validated outputs and acceptable memory and latency.
- The production endpoint matches the reviewed model package and auditor can trace its evaluation and approver.
- A simulated green regression returns traffic to the last healthy model without extended outage.
- Low-frequency requests succeed within agreed latency while unused provisioned compute spend is reduced.
Explanation
Correct Answer
A:The compiled artifact runs on the target device with validated outputs and acceptable memory and latency. This is the acceptance evidence for the described sagemaker neo edge compiler compatibility scenario. Edge optimization depends on target hardware and supported operations, not merely copying a cloud training artifact. The corrective implementation it validates is: Compile the supported model for the specific edge target using SageMaker Neo or choose a compatible conversion route. Reported signal: An edge vision model is deployed to a constrained device with a framework binary and operators assume any model artifact runs on its target accelerator. Operational setting: While testing a second region for recoverable deployment of the same ML service.
Incorrect Answers
B:The production endpoint matches the reviewed model package and auditor can trace its evaluation and approver. This could validate model registry version promotion provenance, but it does not establish that sagemaker neo edge compiler compatibility is resolved for this workload. Confirm instead: The compiled artifact runs on the target device with validated outputs and acceptable memory and latency. Observed problem: An edge vision model is deployed to a constrained device with a framework binary and operators assume any model artifact runs on its target accelerator. Operational setting: While testing a second region for recoverable deployment of the same ML service.
C:A simulated green regression returns traffic to the last healthy model without extended outage. This could validate blue-green endpoint rollback after health failure, but it does not establish that sagemaker neo edge compiler compatibility is resolved for this workload. Confirm instead: The compiled artifact runs on the target device with validated outputs and acceptable memory and latency. Observed problem: An edge vision model is deployed to a constrained device with a framework binary and operators assume any model artifact runs on its target accelerator. Operational setting: While testing a second region for recoverable deployment of the same ML service.
D:Low-frequency requests succeed within agreed latency while unused provisioned compute spend is reduced. This could validate serverless inference for intermittent requests, but it does not establish that sagemaker neo edge compiler compatibility is resolved for this workload. Confirm instead: The compiled artifact runs on the target device with validated outputs and acceptable memory and latency. Observed problem: An edge vision model is deployed to a constrained device with a framework binary and operators assume any model artifact runs on its target accelerator. Operational setting: While testing a second region for recoverable deployment of the same ML service.
Question 1011
As a regulated ML workload prepares for an automated regional rollout, the following condition is observed: An asynchronous inference queue's oldest pending request age keeps increasing, yet GPU utilization alerts remain normal because requests are waiting before execution. The team must demonstrate a repeatable monitoring or security control. Which observation most directly identifies the root cause of the reported behavior?
- Compare incoming captured feature statistics against a stored training baseline and review SageMaker Model Monitor drift alerts.
- Inspect async inference backlog size, age of oldest request, worker capacity and completion/error metrics.
- Inspect the assumed SageMaker execution role, bucket/KMS policies, object path, and access-denied CloudTrail evidence.
- Inspect consistent log group names, job identifiers, timestamps, and failure codes in CloudWatch Logs.
Explanation
Correct Answer
B:Inspect async inference backlog size, age of oldest request, worker capacity and completion/error metrics. This directly tests asynchronous inference queue backlog alert in the reported AWS ML workflow. Queued work can miss response deadlines without saturating compute; queue age is its own performance signal. It distinguishes the problem before choosing a change. The related targeted response is: Scale compatible asynchronous inference capacity or tune workload and queue alarms for pending-request age. Reported signal: An asynchronous inference queue's oldest pending request age keeps increasing, yet GPU utilization alerts remain normal because requests are waiting before execution. Operational setting: As a regulated ML workload prepares for an automated regional rollout.
Incorrect Answers
A:Compare incoming captured feature statistics against a stored training baseline and review SageMaker Model Monitor drift alerts. This is a possible diagnostic for sagemaker model monitor feature drift baseline, but it does not directly test the asynchronous inference queue backlog alert failure reported here. Prioritize: Inspect async inference backlog size, age of oldest request, worker capacity and completion/error metrics. The relevant case is: An asynchronous inference queue's oldest pending request age keeps increasing, yet GPU utilization alerts remain normal because requests are waiting before execution. Operational setting: As a regulated ML workload prepares for an automated regional rollout.
C:Inspect the assumed SageMaker execution role, bucket/KMS policies, object path, and access-denied CloudTrail evidence. This is a possible diagnostic for sagemaker execution role s3 least privilege, but it does not directly test the asynchronous inference queue backlog alert failure reported here. Prioritize: Inspect async inference backlog size, age of oldest request, worker capacity and completion/error metrics. The relevant case is: An asynchronous inference queue's oldest pending request age keeps increasing, yet GPU utilization alerts remain normal because requests are waiting before execution. Operational setting: As a regulated ML workload prepares for an automated regional rollout.
D:Inspect consistent log group names, job identifiers, timestamps, and failure codes in CloudWatch Logs. This is a possible diagnostic for cloudwatch logs insights error correlation, but it does not directly test the asynchronous inference queue backlog alert failure reported here. Prioritize: Inspect async inference backlog size, age of oldest request, worker capacity and completion/error metrics. The relevant case is: An asynchronous inference queue's oldest pending request age keeps increasing, yet GPU utilization alerts remain normal because requests are waiting before execution. Operational setting: As a regulated ML workload prepares for an automated regional rollout.
Question 1012
While reproducing a training incident in an independent SageMaker experiment, the following condition is observed: A support team needs text classification using labeled messages, but engineers choose a clustering algorithm that produces no class-specific predictions. The experiment must be compared fairly against its prior baseline. Which TWO steps best establish the evidence and correct the identified root cause? Select TWO.
- Reduce per-device batch size or use appropriate gradient accumulation or mixed precision after compatibility validation.
- Use an unsupervised clustering approach such as k-means after appropriate feature preparation and assess whether groups are meaningful.
- Use an appropriate text classification model or supported SageMaker BlazingText supervised approach with validated labels.
- Inspect training metrics per trial, search bounds, tuning-objective direction, and compute constraints.
- Inspect the supervised labels, text representation, model-input format, and the desired classification output.
Explanation
Correct Answers
C:Use an appropriate text classification model or supported SageMaker BlazingText supervised approach with validated labels. This is the appropriate controlled remedy for sagemaker built-in blazingtext choice. Supervised text classification requires labels and task-specific evaluation rather than unlabeled similarity clusters. Verify the cause with: Inspect the supervised labels, text representation, model-input format, and the desired classification output. Then confirm: Held-out messages receive the intended class predictions and macro-averaged quality is measurable. This is one part of a select-two decision; the matching evidence and corrective action are both required. Reported signal: A support team needs text classification using labeled messages, but engineers choose a clustering algorithm that produces no class-specific predictions. Operational setting: While reproducing a training incident in an independent SageMaker experiment.
E:Inspect the supervised labels, text representation, model-input format, and the desired classification output. This directly tests sagemaker built-in blazingtext choice in the reported AWS ML workflow. Supervised text classification requires labels and task-specific evaluation rather than unlabeled similarity clusters. It distinguishes the problem before choosing a change. The related targeted response is: Use an appropriate text classification model or supported SageMaker BlazingText supervised approach with validated labels. This is one part of a select-two decision; the matching evidence and corrective action are both required. Reported signal: A support team needs text classification using labeled messages, but engineers choose a clustering algorithm that produces no class-specific predictions. Operational setting: While reproducing a training incident in an independent SageMaker experiment.
Incorrect Answers
A:Reduce per-device batch size or use appropriate gradient accumulation or mixed precision after compatibility validation. This selection concerns training batch size causes gpu memory exhaustion rather than the required diagnostic-and-remedy pair for sagemaker built-in blazingtext choice. Use evidence: Inspect the supervised labels, text representation, model-input format, and the desired classification output. and the matching correction: Use an appropriate text classification model or supported SageMaker BlazingText supervised approach with validated labels. Signal: A support team needs text classification using labeled messages, but engineers choose a clustering algorithm that produces no class-specific predictions. Operational setting: While reproducing a training incident in an independent SageMaker experiment.
B:Use an unsupervised clustering approach such as k-means after appropriate feature preparation and assess whether groups are meaningful. This selection concerns clustering for unlabeled customer segments rather than the required diagnostic-and-remedy pair for sagemaker built-in blazingtext choice. Use evidence: Inspect the supervised labels, text representation, model-input format, and the desired classification output. and the matching correction: Use an appropriate text classification model or supported SageMaker BlazingText supervised approach with validated labels. Signal: A support team needs text classification using labeled messages, but engineers choose a clustering algorithm that produces no class-specific predictions. Operational setting: While reproducing a training incident in an independent SageMaker experiment.
D:Inspect training metrics per trial, search bounds, tuning-objective direction, and compute constraints. This selection concerns hyperparameter bayesian tuning strategy rather than the required diagnostic-and-remedy pair for sagemaker built-in blazingtext choice. Use evidence: Inspect the supervised labels, text representation, model-input format, and the desired classification output. and the matching correction: Use an appropriate text classification model or supported SageMaker BlazingText supervised approach with validated labels. Signal: A support team needs text classification using labeled messages, but engineers choose a clustering algorithm that produces no class-specific predictions. Operational setting: While reproducing a training incident in an independent SageMaker experiment.
Question 1013
At the model-governance gate before promotion of a new ML artifact, the following condition is observed: A model reaches near-perfect training accuracy but substantially worse holdout performance as feature coefficients grow unusually large. The team needs a defensible technical decision rather than an untracked workaround. Which TWO steps best establish the evidence and correct the identified root cause? Select TWO.
- Increase justified L2 regularization or reduce model capacity and revalidate with an untouched holdout.
- Evaluate PR-AUC and precision at required recall alongside ROC-AUC, and select a deployment threshold for actual prevalence.
- Reduce per-device batch size or use appropriate gradient accumulation or mixed precision after compatibility validation.
- Compare training and validation curves, coefficient magnitudes, and error by split.
- Inspect target types, business loss function, sample labels, and whether the required output is continuous or categorical.
Explanation
Correct Answers
A:Increase justified L2 regularization or reduce model capacity and revalidate with an untouched holdout. This is the appropriate controlled remedy for overfitting reduced with l2 regularization. L2 penalizes large weights and can reduce variance but excessive regularization also causes underfitting. Verify the cause with: Compare training and validation curves, coefficient magnitudes, and error by split. Then confirm: The train-validation gap narrows without losing acceptable validation performance. This is one part of a select-two decision; the matching evidence and corrective action are both required. Reported signal: A model reaches near-perfect training accuracy but substantially worse holdout performance as feature coefficients grow unusually large. Operational setting: At the model-governance gate before promotion of a new ML artifact.
D:Compare training and validation curves, coefficient magnitudes, and error by split. This directly tests overfitting reduced with l2 regularization in the reported AWS ML workflow. L2 penalizes large weights and can reduce variance but excessive regularization also causes underfitting. It distinguishes the problem before choosing a change. The related targeted response is: Increase justified L2 regularization or reduce model capacity and revalidate with an untouched holdout. This is one part of a select-two decision; the matching evidence and corrective action are both required. Reported signal: A model reaches near-perfect training accuracy but substantially worse holdout performance as feature coefficients grow unusually large. Operational setting: At the model-governance gate before promotion of a new ML artifact.
Incorrect Answers
B:Evaluate PR-AUC and precision at required recall alongside ROC-AUC, and select a deployment threshold for actual prevalence. This selection concerns roc-auc versus pr-auc for rare events rather than the required diagnostic-and-remedy pair for overfitting reduced with l2 regularization. Use evidence: Compare training and validation curves, coefficient magnitudes, and error by split. and the matching correction: Increase justified L2 regularization or reduce model capacity and revalidate with an untouched holdout. Signal: A model reaches near-perfect training accuracy but substantially worse holdout performance as feature coefficients grow unusually large. Operational setting: At the model-governance gate before promotion of a new ML artifact.
C:Reduce per-device batch size or use appropriate gradient accumulation or mixed precision after compatibility validation. This selection concerns training batch size causes gpu memory exhaustion rather than the required diagnostic-and-remedy pair for overfitting reduced with l2 regularization. Use evidence: Compare training and validation curves, coefficient magnitudes, and error by split. and the matching correction: Increase justified L2 regularization or reduce model capacity and revalidate with an untouched holdout. Signal: A model reaches near-perfect training accuracy but substantially worse holdout performance as feature coefficients grow unusually large. Operational setting: At the model-governance gate before promotion of a new ML artifact.
E:Inspect target types, business loss function, sample labels, and whether the required output is continuous or categorical. This selection concerns regression versus classification problem formulation rather than the required diagnostic-and-remedy pair for overfitting reduced with l2 regularization. Use evidence: Compare training and validation curves, coefficient magnitudes, and error by split. and the matching correction: Increase justified L2 regularization or reduce model capacity and revalidate with an untouched holdout. Signal: A model reaches near-perfect training accuracy but substantially worse holdout performance as feature coefficients grow unusually large. Operational setting: At the model-governance gate before promotion of a new ML artifact.
Question 1014
During an approved rollout of an AWS ML inference workload, the following condition is observed: A pipeline deploys a model directly from an unversioned notebook S3 path and cannot prove which evaluated candidate was approved. The team must meet its documented availability and rollback constraints. Which verification result BEST demonstrates that the issue is resolved?
- Large inputs complete without client request timeout and clients receive tracked result locations or failure notifications.
- A synthetic load surge meets error and latency SLOs without rejected provisioning due to quota or startup lag.
- The build uploads signed artifacts successfully while unauthorized buckets and prefixes remain denied.
- The production endpoint matches the reviewed model package and auditor can trace its evaluation and approver.
Explanation
Correct Answer
D:The production endpoint matches the reviewed model package and auditor can trace its evaluation and approver. This is the acceptance evidence for the described model registry version promotion provenance scenario. Immutable package versions and approval metadata connect evaluation with deployment; mutable S3 nicknames do not. The corrective implementation it validates is: Promote an explicitly approved SageMaker Model Registry package version through repeatable pipeline stages. Reported signal: A pipeline deploys a model directly from an unversioned notebook S3 path and cannot prove which evaluated candidate was approved. Operational setting: During an approved rollout of an AWS ML inference workload.
Incorrect Answers
A:Large inputs complete without client request timeout and clients receive tracked result locations or failure notifications. This could validate asynchronous endpoint for large payloads, but it does not establish that model registry version promotion provenance is resolved for this workload. Confirm instead: The production endpoint matches the reviewed model package and auditor can trace its evaluation and approver. Observed problem: A pipeline deploys a model directly from an unversioned notebook S3 path and cannot prove which evaluated candidate was approved. Operational setting: During an approved rollout of an AWS ML inference workload.
B:A synthetic load surge meets error and latency SLOs without rejected provisioning due to quota or startup lag. This could validate endpoint scale-to-zero cold capacity planning, but it does not establish that model registry version promotion provenance is resolved for this workload. Confirm instead: The production endpoint matches the reviewed model package and auditor can trace its evaluation and approver. Observed problem: A pipeline deploys a model directly from an unversioned notebook S3 path and cannot prove which evaluated candidate was approved. Operational setting: During an approved rollout of an AWS ML inference workload.
C:The build uploads signed artifacts successfully while unauthorized buckets and prefixes remain denied. This could validate codebuild iam least-privilege artifact upload, but it does not establish that model registry version promotion provenance is resolved for this workload. Confirm instead: The production endpoint matches the reviewed model package and auditor can trace its evaluation and approver. Observed problem: A pipeline deploys a model directly from an unversioned notebook S3 path and cannot prove which evaluated candidate was approved. Operational setting: During an approved rollout of an AWS ML inference workload.
Question 1015
During a controlled replay of the data-preparation workflow in a separate ML staging account, the following condition is observed: A source feed silently begins emitting missing customer IDs, and the downstream pipeline trains on rows with null entity keys instead of failing. The dataset contract must remain reproducible for a backfill. What postchange observation provides the strongest acceptance evidence?
- Historical evaluation reproduces features available at prediction time, and production freshness is monitored against the same contract.
- Invalid batches are rejected with attributable rule failures and accepted training data meets the entity-key contract.
- Annotation workers can label required content without viewing direct identifiers and authorized audits can trace approved mappings.
- Both supported source versions produce the agreed typed feature, and invalid values enter an auditable reject path.
Explanation
Correct Answer
B:Invalid batches are rejected with attributable rule failures and accepted training data meets the entity-key contract. This is the acceptance evidence for the described glue data quality non-null contract scenario. Schema existence alone cannot establish semantic completeness or identifier uniqueness in ML inputs. The corrective implementation it validates is: Enforce a required non-null or uniqueness contract with thresholds; quarantine violating partitions before feature publication. Reported signal: A source feed silently begins emitting missing customer IDs, and the downstream pipeline trains on rows with null entity keys instead of failing. Operational setting: During a controlled replay of the data-preparation workflow in a separate ML staging account.
Incorrect Answers
A:Historical evaluation reproduces features available at prediction time, and production freshness is monitored against the same contract. This could validate sagemaker feature store online offline consistency, but it does not establish that glue data quality non-null contract is resolved for this workload. Confirm instead: Invalid batches are rejected with attributable rule failures and accepted training data meets the entity-key contract. Observed problem: A source feed silently begins emitting missing customer IDs, and the downstream pipeline trains on rows with null entity keys instead of failing. Operational setting: During a controlled replay of the data-preparation workflow in a separate ML staging account.
C:Annotation workers can label required content without viewing direct identifiers and authorized audits can trace approved mappings. This could validate pii masking before model labeling, but it does not establish that glue data quality non-null contract is resolved for this workload. Confirm instead: Invalid batches are rejected with attributable rule failures and accepted training data meets the entity-key contract. Observed problem: A source feed silently begins emitting missing customer IDs, and the downstream pipeline trains on rows with null entity keys instead of failing. Operational setting: During a controlled replay of the data-preparation workflow in a separate ML staging account.
D:Both supported source versions produce the agreed typed feature, and invalid values enter an auditable reject path. This could validate glue schema evolution with mixed input types, but it does not establish that glue data quality non-null contract is resolved for this workload. Confirm instead: Invalid batches are rejected with attributable rule failures and accepted training data meets the entity-key contract. Observed problem: A source feed silently begins emitting missing customer IDs, and the downstream pipeline trains on rows with null entity keys instead of failing. Operational setting: During a controlled replay of the data-preparation workflow in a separate ML staging account.
Question 1016
As a regional data-pipeline rollout is validated against approved AWS storage and streams, the following condition is observed: Several distributed workers repeatedly read small shared checkpoints with POSIX file semantics, but the pipeline treats object-storage paths as a mutable shared filesystem. The data team must preserve event-time meaning and lineage. Which investigation provides the most decisive FIRST evidence for the underlying problem?
- Compare vocabulary hashes, normalization steps, token IDs for test strings, and serialized preprocessing versions.
- Inspect categorical cardinality, sparse matrix dimensions, feature contribution, and generalization to unseen identifiers.
- Inspect workload read/write semantics, shared access requirements, filesystem operations, and object-store request patterns.
- Inspect sample raw records, quoting rules, expected headers, and parsed column counts in the ingestion job.
Explanation
Correct Answer
C:Inspect workload read/write semantics, shared access requirements, filesystem operations, and object-store request patterns. This directly tests s3 versus efs training storage choice in the reported AWS ML workflow. S3 object semantics differ from mounted shared filesystems; choose storage for workload access behavior. It distinguishes the problem before choosing a change. The related targeted response is: Place shared POSIX-style working files on suitable EFS or FSx storage and keep immutable large artifacts in S3. Reported signal: Several distributed workers repeatedly read small shared checkpoints with POSIX file semantics, but the pipeline treats object-storage paths as a mutable shared filesystem. Operational setting: As a regional data-pipeline rollout is validated against approved AWS storage and streams.
Incorrect Answers
A:Compare vocabulary hashes, normalization steps, token IDs for test strings, and serialized preprocessing versions. This is a possible diagnostic for text tokenization mismatch in serving, but it does not directly test the s3 versus efs training storage choice failure reported here. Prioritize: Inspect workload read/write semantics, shared access requirements, filesystem operations, and object-store request patterns. The relevant case is: Several distributed workers repeatedly read small shared checkpoints with POSIX file semantics, but the pipeline treats object-storage paths as a mutable shared filesystem. Operational setting: As a regional data-pipeline rollout is validated against approved AWS storage and streams.
B:Inspect categorical cardinality, sparse matrix dimensions, feature contribution, and generalization to unseen identifiers. This is a possible diagnostic for high-cardinality categorical encoding, but it does not directly test the s3 versus efs training storage choice failure reported here. Prioritize: Inspect workload read/write semantics, shared access requirements, filesystem operations, and object-store request patterns. The relevant case is: Several distributed workers repeatedly read small shared checkpoints with POSIX file semantics, but the pipeline treats object-storage paths as a mutable shared filesystem. Operational setting: As a regional data-pipeline rollout is validated against approved AWS storage and streams.
D:Inspect sample raw records, quoting rules, expected headers, and parsed column counts in the ingestion job. This is a possible diagnostic for csv schema ambiguity during ingestion, but it does not directly test the s3 versus efs training storage choice failure reported here. Prioritize: Inspect workload read/write semantics, shared access requirements, filesystem operations, and object-store request patterns. The relevant case is: Several distributed workers repeatedly read small shared checkpoints with POSIX file semantics, but the pipeline treats object-storage paths as a mutable shared filesystem. Operational setting: As a regional data-pipeline rollout is validated against approved AWS storage and streams.
Question 1017
As a regional data-pipeline rollout is validated against approved AWS storage and streams, the following condition is observed: A daily S3 dataset consists of millions of tiny Parquet files, and training startup spends more time opening objects than scanning data. The data team must preserve event-time meaning and lineage. Which investigation provides the most decisive FIRST evidence for the underlying problem?
- Compare per-shard incoming records, write throttles, partition-key cardinality, and enhanced fan-out consumer lag.
- Inspect feature event timestamps, target prediction cutoffs, and join predicates in the historical training query.
- Measure file-size distribution, object request count, Spark task scheduling overhead, and read throughput.
- Inspect preprocessing fit order, imputation statistics, dataset split boundaries, and pipeline serialization.
Explanation
Correct Answer
C:Measure file-size distribution, object request count, Spark task scheduling overhead, and read throughput. This directly tests columnar file compaction for small objects in the reported AWS ML workflow. Excessive small files create metadata and request overhead even with a columnar data format. It distinguishes the problem before choosing a change. The related targeted response is: Compact the partitioned dataset into appropriately sized Parquet objects without changing the record keys or schema. Reported signal: A daily S3 dataset consists of millions of tiny Parquet files, and training startup spends more time opening objects than scanning data. Operational setting: As a regional data-pipeline rollout is validated against approved AWS storage and streams.
Incorrect Answers
A:Compare per-shard incoming records, write throttles, partition-key cardinality, and enhanced fan-out consumer lag. This is a possible diagnostic for kinesis data streams shard hot key, but it does not directly test the columnar file compaction for small objects failure reported here. Prioritize: Measure file-size distribution, object request count, Spark task scheduling overhead, and read throughput. The relevant case is: A daily S3 dataset consists of millions of tiny Parquet files, and training startup spends more time opening objects than scanning data. Operational setting: As a regional data-pipeline rollout is validated against approved AWS storage and streams.
B:Inspect feature event timestamps, target prediction cutoffs, and join predicates in the historical training query. This is a possible diagnostic for point-in-time feature data leakage, but it does not directly test the columnar file compaction for small objects failure reported here. Prioritize: Measure file-size distribution, object request count, Spark task scheduling overhead, and read throughput. The relevant case is: A daily S3 dataset consists of millions of tiny Parquet files, and training startup spends more time opening objects than scanning data. Operational setting: As a regional data-pipeline rollout is validated against approved AWS storage and streams.
D:Inspect preprocessing fit order, imputation statistics, dataset split boundaries, and pipeline serialization. This is a possible diagnostic for median imputation fitted on training split, but it does not directly test the columnar file compaction for small objects failure reported here. Prioritize: Measure file-size distribution, object request count, Spark task scheduling overhead, and read throughput. The relevant case is: A daily S3 dataset consists of millions of tiny Parquet files, and training startup spends more time opening objects than scanning data. Operational setting: As a regional data-pipeline rollout is validated against approved AWS storage and streams.
Question 1018
As a regulated ML workload prepares for an automated regional rollout, the following condition is observed: A real-time recommender fetches cached customer history updated hourly while offline training used a daily snapshot with different aggregation boundaries. The team must demonstrate a repeatable monitoring or security control. Which verification result BEST demonstrates that the issue is resolved?
- Peak-window 5xx rate and p95 latency fall below service objectives with verified alarms.
- The monitor joins each known labeled request to the corresponding earlier prediction and computes expected metrics.
- The pipeline deploys with the authorized role and cannot assign an unrelated privileged role.
- Parity checks match transformations within documented freshness tolerances for test entities.
Explanation
Correct Answer
D:Parity checks match transformations within documented freshness tolerances for test entities. This is the acceptance evidence for the described training-serving feature skew incident scenario. Mismatched feature windows and update cadence can degrade production predictions even if training accuracy is stable. The corrective implementation it validates is: Use a consistent versioned feature definition and test offline/online values at the same observation cutoff. Reported signal: A real-time recommender fetches cached customer history updated hourly while offline training used a daily snapshot with different aggregation boundaries. Operational setting: As a regulated ML workload prepares for an automated regional rollout.
Incorrect Answers
A:Peak-window 5xx rate and p95 latency fall below service objectives with verified alarms. This could validate cloudwatch 5xxerrors endpoint saturation, but it does not establish that training-serving feature skew incident is resolved for this workload. Confirm instead: Parity checks match transformations within documented freshness tolerances for test entities. Observed problem: A real-time recommender fetches cached customer history updated hourly while offline training used a daily snapshot with different aggregation boundaries. Operational setting: As a regulated ML workload prepares for an automated regional rollout.
B:The monitor joins each known labeled request to the corresponding earlier prediction and computes expected metrics. This could validate model monitor ground truth association, but it does not establish that training-serving feature skew incident is resolved for this workload. Confirm instead: Parity checks match transformations within documented freshness tolerances for test entities. Observed problem: A real-time recommender fetches cached customer history updated hourly while offline training used a daily snapshot with different aggregation boundaries. Operational setting: As a regulated ML workload prepares for an automated regional rollout.
C:The pipeline deploys with the authorized role and cannot assign an unrelated privileged role. This could validate iam passrole constraint for model deployment, but it does not establish that training-serving feature skew incident is resolved for this workload. Confirm instead: Parity checks match transformations within documented freshness tolerances for test entities. Observed problem: A real-time recommender fetches cached customer history updated hourly while offline training used a daily snapshot with different aggregation boundaries. Operational setting: As a regulated ML workload prepares for an automated regional rollout.
Question 1019
At the model-governance gate before promotion of a new ML artifact, the following condition is observed: A text classifier includes thousands of weak correlated features and becomes unstable across repeated training runs. The team needs a defensible technical decision rather than an untracked workaround. Which measured outcome would confirm that the correction worked as intended?
- Holdout quality stays acceptable while model size and run-to-run instability decrease.
- A production endpoint resolves to one approved model package version with traceable training lineage.
- The evaluation report shows both typical error and extreme-error impact with a justified primary metric.
- Stable clusters exhibit interpretable differences in held-out descriptors without claiming supervised classification accuracy.
Explanation
Correct Answer
A:Holdout quality stays acceptable while model size and run-to-run instability decrease. This is the acceptance evidence for the described feature selection to reduce overparameterization scenario. Unnecessary high-dimensional inputs can increase variance and compute cost; selection itself must avoid evaluation leakage. The corrective implementation it validates is: Use justified feature selection or sparsity constraints fitted on training folds before retraining. Reported signal: A text classifier includes thousands of weak correlated features and becomes unstable across repeated training runs. Operational setting: At the model-governance gate before promotion of a new ML artifact.
Incorrect Answers
B:A production endpoint resolves to one approved model package version with traceable training lineage. This could validate sagemaker model registry candidate approval, but it does not establish that feature selection to reduce overparameterization is resolved for this workload. Confirm instead: Holdout quality stays acceptable while model size and run-to-run instability decrease. Observed problem: A text classifier includes thousands of weak correlated features and becomes unstable across repeated training runs. Operational setting: At the model-governance gate before promotion of a new ML artifact.
C:The evaluation report shows both typical error and extreme-error impact with a justified primary metric. This could validate rmse versus mae outlier sensitivity, but it does not establish that feature selection to reduce overparameterization is resolved for this workload. Confirm instead: Holdout quality stays acceptable while model size and run-to-run instability decrease. Observed problem: A text classifier includes thousands of weak correlated features and becomes unstable across repeated training runs. Operational setting: At the model-governance gate before promotion of a new ML artifact.
D:Stable clusters exhibit interpretable differences in held-out descriptors without claiming supervised classification accuracy. This could validate clustering for unlabeled customer segments, but it does not establish that feature selection to reduce overparameterization is resolved for this workload. Confirm instead: Holdout quality stays acceptable while model size and run-to-run instability decrease. Observed problem: A text classifier includes thousands of weak correlated features and becomes unstable across repeated training runs. Operational setting: At the model-governance gate before promotion of a new ML artifact.
Question 1020
As a regional data-pipeline rollout is validated against approved AWS storage and streams, the following condition is observed: A daily AWS Glue ingestion job reprocesses a month of files after a deployment because its job bookmark state was reset or the job identity changed. The data team must preserve event-time meaning and lineage. Which observation most directly identifies the root cause of the reported behavior?
- Check job names, bookmark settings, run history, source object timestamps, and the previously committed checkpoint.
- Inspect sampler placement, duplicate neighbor IDs, fold membership, and class-conditioned metric changes.
- Inspect the stream retention setting, consumer checkpoint timestamps, and the oldest available sequence or time position.
- Inspect S3 replication rules, region-specific bucket policies, Glue job destinations, KMS keys, and applicable residency controls.
Explanation
Correct Answer
A:Check job names, bookmark settings, run history, source object timestamps, and the previously committed checkpoint. This directly tests glue job bookmark state error in the reported AWS ML workflow. Glue job bookmarks track supported processed state and must be managed deliberately during job changes. It distinguishes the problem before choosing a change. The related targeted response is: Preserve the intended bookmark state across deployments and run an explicit controlled backfill only when required. Reported signal: A daily AWS Glue ingestion job reprocesses a month of files after a deployment because its job bookmark state was reset or the job identity changed. Operational setting: As a regional data-pipeline rollout is validated against approved AWS storage and streams.
Incorrect Answers
B:Inspect sampler placement, duplicate neighbor IDs, fold membership, and class-conditioned metric changes. This is a possible diagnostic for class imbalance handled inside cross-validation, but it does not directly test the glue job bookmark state error failure reported here. Prioritize: Check job names, bookmark settings, run history, source object timestamps, and the previously committed checkpoint. The relevant case is: A daily AWS Glue ingestion job reprocesses a month of files after a deployment because its job bookmark state was reset or the job identity changed. Operational setting: As a regional data-pipeline rollout is validated against approved AWS storage and streams.
C:Inspect the stream retention setting, consumer checkpoint timestamps, and the oldest available sequence or time position. This is a possible diagnostic for kinesis consumer retention gap, but it does not directly test the glue job bookmark state error failure reported here. Prioritize: Check job names, bookmark settings, run history, source object timestamps, and the previously committed checkpoint. The relevant case is: A daily AWS Glue ingestion job reprocesses a month of files after a deployment because its job bookmark state was reset or the job identity changed. Operational setting: As a regional data-pipeline rollout is validated against approved AWS storage and streams.
D:Inspect S3 replication rules, region-specific bucket policies, Glue job destinations, KMS keys, and applicable residency controls. This is a possible diagnostic for model training data residency boundary, but it does not directly test the glue job bookmark state error failure reported here. Prioritize: Check job names, bookmark settings, run history, source object timestamps, and the previously committed checkpoint. The relevant case is: A daily AWS Glue ingestion job reprocesses a month of files after a deployment because its job bookmark state was reset or the job identity changed. Operational setting: As a regional data-pipeline rollout is validated against approved AWS storage and streams.
