Amazon AWS Certified Machine Learning Engineer - Associate MLA-C01 Practice Test 108
Question 1070
At a production AWS ML data lake supporting customer-facing predictions, the following condition is observed: A demand model treats hour 23 and hour 0 as far apart when integer hour is used as a linear continuous feature. Engineers must avoid changing unrelated source partitions. Which observation most directly identifies the root cause of the reported behavior?
- Inspect preprocessing fit order, imputation statistics, dataset split boundaries, and pipeline serialization.
- Inspect event identifiers, object versions or ETags, processing ledger, and downstream record keys for duplicate deliveries.
- Inspect prediction errors around midnight and feature representations for the periodic hour variable.
- Check job names, bookmark settings, run history, source object timestamps, and the previously committed checkpoint.
Explanation
Correct Answer
C:Inspect prediction errors around midnight and feature representations for the periodic hour variable. This directly tests time-series cyclical calendar feature in the reported AWS ML workflow. Time-of-day has periodic topology; raw integer distance does not represent wraparound. It distinguishes the problem before choosing a change. The related targeted response is: Encode hour with sine and cosine cyclic features and validate with a time-aware split. Reported signal: A demand model treats hour 23 and hour 0 as far apart when integer hour is used as a linear continuous feature. Operational setting: At a production AWS ML data lake supporting customer-facing predictions.
Incorrect Answers
A: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 time-series cyclical calendar feature failure reported here. Prioritize: Inspect prediction errors around midnight and feature representations for the periodic hour variable. The relevant case is: A demand model treats hour 23 and hour 0 as far apart when integer hour is used as a linear continuous feature. Operational setting: At a production AWS ML data lake supporting customer-facing predictions.
B:Inspect event identifiers, object versions or ETags, processing ledger, and downstream record keys for duplicate deliveries. This is a possible diagnostic for s3 object event at-least-once processing, but it does not directly test the time-series cyclical calendar feature failure reported here. Prioritize: Inspect prediction errors around midnight and feature representations for the periodic hour variable. The relevant case is: A demand model treats hour 23 and hour 0 as far apart when integer hour is used as a linear continuous feature. Operational setting: At a production AWS ML data lake supporting customer-facing predictions.
D:Check job names, bookmark settings, run history, source object timestamps, and the previously committed checkpoint. This is a possible diagnostic for glue job bookmark state error, but it does not directly test the time-series cyclical calendar feature failure reported here. Prioritize: Inspect prediction errors around midnight and feature representations for the periodic hour variable. The relevant case is: A demand model treats hour 23 and hour 0 as far apart when integer hour is used as a linear continuous feature. Operational setting: At a production AWS ML data lake supporting customer-facing predictions.
Question 1071
At a production AWS ML data lake supporting customer-facing predictions, the following condition is observed: Three SageMaker Ground Truth annotators assign conflicting labels to borderline product photos, but a training job assumes all majority votes are gold. Engineers must avoid changing unrelated source partitions. Which TWO complementary choices provide the decisive diagnosis AND corrective response? Select TWO.
- Inspect per-class confusion, worker agreement, label instructions, and disagreements on an adjudicated reference set.
- Revise instructions, route ambiguous samples for expert adjudication, and monitor label quality per class.
- Choose partition keys with enough cardinality or reshard the stream for the intended throughput, while preserving required per-key ordering.
- Inspect sampler placement, duplicate neighbor IDs, fold membership, and class-conditioned metric changes.
- Define event-time windows with appropriate lateness handling and a documented revision policy for late records.
Explanation
Correct Answers
A:Inspect per-class confusion, worker agreement, label instructions, and disagreements on an adjudicated reference set. This directly tests ground truth labeling agreement review in the reported AWS ML workflow. Majority vote does not guarantee semantic correctness when instructions are ambiguous or a class is rare. It distinguishes the problem before choosing a change. The related targeted response is: Revise instructions, route ambiguous samples for expert adjudication, and monitor label quality per class. This is one part of a select-two decision; the matching evidence and corrective action are both required. Reported signal: Three SageMaker Ground Truth annotators assign conflicting labels to borderline product photos, but a training job assumes all majority votes are gold. Operational setting: At a production AWS ML data lake supporting customer-facing predictions.
B:Revise instructions, route ambiguous samples for expert adjudication, and monitor label quality per class. This is the appropriate controlled remedy for ground truth labeling agreement review. Majority vote does not guarantee semantic correctness when instructions are ambiguous or a class is rare. Verify the cause with: Inspect per-class confusion, worker agreement, label instructions, and disagreements on an adjudicated reference set. Then confirm: Agreement on a held-out gold set rises without silently discarding rare legitimate classes. This is one part of a select-two decision; the matching evidence and corrective action are both required. Reported signal: Three SageMaker Ground Truth annotators assign conflicting labels to borderline product photos, but a training job assumes all majority votes are gold. Operational setting: At a production AWS ML data lake supporting customer-facing predictions.
Incorrect Answers
C:Choose partition keys with enough cardinality or reshard the stream for the intended throughput, while preserving required per-key ordering. This selection concerns kinesis data streams shard hot key rather than the required diagnostic-and-remedy pair for ground truth labeling agreement review. Use evidence: Inspect per-class confusion, worker agreement, label instructions, and disagreements on an adjudicated reference set. and the matching correction: Revise instructions, route ambiguous samples for expert adjudication, and monitor label quality per class. Signal: Three SageMaker Ground Truth annotators assign conflicting labels to borderline product photos, but a training job assumes all majority votes are gold. Operational setting: At a production AWS ML data lake supporting customer-facing predictions.
D:Inspect sampler placement, duplicate neighbor IDs, fold membership, and class-conditioned metric changes. This selection concerns class imbalance handled inside cross-validation rather than the required diagnostic-and-remedy pair for ground truth labeling agreement review. Use evidence: Inspect per-class confusion, worker agreement, label instructions, and disagreements on an adjudicated reference set. and the matching correction: Revise instructions, route ambiguous samples for expert adjudication, and monitor label quality per class. Signal: Three SageMaker Ground Truth annotators assign conflicting labels to borderline product photos, but a training job assumes all majority votes are gold. Operational setting: At a production AWS ML data lake supporting customer-facing predictions.
E:Define event-time windows with appropriate lateness handling and a documented revision policy for late records. This selection concerns streaming feature rolling window correctness rather than the required diagnostic-and-remedy pair for ground truth labeling agreement review. Use evidence: Inspect per-class confusion, worker agreement, label instructions, and disagreements on an adjudicated reference set. and the matching correction: Revise instructions, route ambiguous samples for expert adjudication, and monitor label quality per class. Signal: Three SageMaker Ground Truth annotators assign conflicting labels to borderline product photos, but a training job assumes all majority votes are gold. Operational setting: At a production AWS ML data lake supporting customer-facing predictions.
Question 1072
During a production MLOps reliability and security review, the following condition is observed: An endpoint's client-side response time rises, but the team only examines SageMaker ModelLatency and concludes the model is slow. The owner needs an auditable resolution without broadening privileges. What should the ML engineer inspect FIRST to distinguish this failure from unrelated issues?
- Inspect endpoint 5XXErrors, InvocationsPerInstance, instance availability, and per-minute latency distribution.
- Compare ModelLatency with OverheadLatency, client network timing, invocation counts, and endpoint 5XX responses.
- Inspect routing hashes, experiment assignments, variant weights, and request identity consistency.
- Inspect constraint threshold selection, false-positive incidents, feature importance and baseline seasonal variation.
Explanation
Correct Answer
B:Compare ModelLatency with OverheadLatency, client network timing, invocation counts, and endpoint 5XX responses. This directly tests cloudwatch invocations and modellatency distinction in the reported AWS ML workflow. Model compute time and overall endpoint latency are distinct signals; blaming model code without tracing overhead is premature. It distinguishes the problem before choosing a change. The related targeted response is: Identify the component contributing to total request time and optimize network or overhead separately from actual model inference. Reported signal: An endpoint's client-side response time rises, but the team only examines SageMaker ModelLatency and concludes the model is slow. Operational setting: During a production MLOps reliability and security review.
Incorrect Answers
A:Inspect endpoint 5XXErrors, InvocationsPerInstance, instance availability, and per-minute latency distribution. This is a possible diagnostic for cloudwatch 5xxerrors endpoint saturation, but it does not directly test the cloudwatch invocations and modellatency distinction failure reported here. Prioritize: Compare ModelLatency with OverheadLatency, client network timing, invocation counts, and endpoint 5XX responses. The relevant case is: An endpoint's client-side response time rises, but the team only examines SageMaker ModelLatency and concludes the model is slow. Operational setting: During a production MLOps reliability and security review.
C:Inspect routing hashes, experiment assignments, variant weights, and request identity consistency. This is a possible diagnostic for champion challenger a/b assignment consistency, but it does not directly test the cloudwatch invocations and modellatency distinction failure reported here. Prioritize: Compare ModelLatency with OverheadLatency, client network timing, invocation counts, and endpoint 5XX responses. The relevant case is: An endpoint's client-side response time rises, but the team only examines SageMaker ModelLatency and concludes the model is slow. Operational setting: During a production MLOps reliability and security review.
D:Inspect constraint threshold selection, false-positive incidents, feature importance and baseline seasonal variation. This is a possible diagnostic for model monitor violation alert calibration, but it does not directly test the cloudwatch invocations and modellatency distinction failure reported here. Prioritize: Compare ModelLatency with OverheadLatency, client network timing, invocation counts, and endpoint 5XX responses. The relevant case is: An endpoint's client-side response time rises, but the team only examines SageMaker ModelLatency and concludes the model is slow. Operational setting: During a production MLOps reliability and security review.
Question 1073
At a production AWS ML data lake supporting customer-facing predictions, the following condition is observed: A new marketing campaign adds hundreds of thousands of unique user IDs, making naïve one-hot features extremely sparse and too large to maintain. Engineers must avoid changing unrelated source partitions. Which TWO steps best establish the evidence and correct the identified root cause? Select TWO.
- Repartition or salt the skewed join appropriately, or use a justified broadcast join when the small side fits memory.
- Increase appropriate retention before long maintenance and persist a durable downstream historical copy for replay beyond retention.
- Inspect group sample sizes, label-positive fractions, and the appropriate SageMaker Clarify pretraining bias report.
- Drop leakage-prone IDs or adopt controlled hashing or suitable learned encodings with stable unknown-category handling.
- Inspect categorical cardinality, sparse matrix dimensions, feature contribution, and generalization to unseen identifiers.
Explanation
Correct Answers
D:Drop leakage-prone IDs or adopt controlled hashing or suitable learned encodings with stable unknown-category handling. This is the appropriate controlled remedy for high-cardinality categorical encoding. One-hot encoding every nearly unique identifier can waste capacity and overfit identities instead of learning transferable patterns. Verify the cause with: Inspect categorical cardinality, sparse matrix dimensions, feature contribution, and generalization to unseen identifiers. Then confirm: Memory use stays within limits and previously unseen categories are processed without dimension mismatch. This is one part of a select-two decision; the matching evidence and corrective action are both required. Reported signal: A new marketing campaign adds hundreds of thousands of unique user IDs, making naïve one-hot features extremely sparse and too large to maintain. Operational setting: At a production AWS ML data lake supporting customer-facing predictions.
E:Inspect categorical cardinality, sparse matrix dimensions, feature contribution, and generalization to unseen identifiers. This directly tests high-cardinality categorical encoding in the reported AWS ML workflow. One-hot encoding every nearly unique identifier can waste capacity and overfit identities instead of learning transferable patterns. It distinguishes the problem before choosing a change. The related targeted response is: Drop leakage-prone IDs or adopt controlled hashing or suitable learned encodings with stable unknown-category handling. This is one part of a select-two decision; the matching evidence and corrective action are both required. Reported signal: A new marketing campaign adds hundreds of thousands of unique user IDs, making naïve one-hot features extremely sparse and too large to maintain. Operational setting: At a production AWS ML data lake supporting customer-facing predictions.
Incorrect Answers
A:Repartition or salt the skewed join appropriately, or use a justified broadcast join when the small side fits memory. This selection concerns glue spark join skew on dominant identifier rather than the required diagnostic-and-remedy pair for high-cardinality categorical encoding. Use evidence: Inspect categorical cardinality, sparse matrix dimensions, feature contribution, and generalization to unseen identifiers. and the matching correction: Drop leakage-prone IDs or adopt controlled hashing or suitable learned encodings with stable unknown-category handling. Signal: A new marketing campaign adds hundreds of thousands of unique user IDs, making naïve one-hot features extremely sparse and too large to maintain. Operational setting: At a production AWS ML data lake supporting customer-facing predictions.
B:Increase appropriate retention before long maintenance and persist a durable downstream historical copy for replay beyond retention. This selection concerns kinesis consumer retention gap rather than the required diagnostic-and-remedy pair for high-cardinality categorical encoding. Use evidence: Inspect categorical cardinality, sparse matrix dimensions, feature contribution, and generalization to unseen identifiers. and the matching correction: Drop leakage-prone IDs or adopt controlled hashing or suitable learned encodings with stable unknown-category handling. Signal: A new marketing campaign adds hundreds of thousands of unique user IDs, making naïve one-hot features extremely sparse and too large to maintain. Operational setting: At a production AWS ML data lake supporting customer-facing predictions.
C:Inspect group sample sizes, label-positive fractions, and the appropriate SageMaker Clarify pretraining bias report. This selection concerns clarify label imbalance measurement rather than the required diagnostic-and-remedy pair for high-cardinality categorical encoding. Use evidence: Inspect categorical cardinality, sparse matrix dimensions, feature contribution, and generalization to unseen identifiers. and the matching correction: Drop leakage-prone IDs or adopt controlled hashing or suitable learned encodings with stable unknown-category handling. Signal: A new marketing campaign adds hundreds of thousands of unique user IDs, making naïve one-hot features extremely sparse and too large to maintain. Operational setting: At a production AWS ML data lake supporting customer-facing predictions.
Question 1074
During an approved rollout of an AWS ML inference workload, the following condition is observed: A data-science team tests only that a training script executes, but deployment repeatedly breaks because client JSON fields changed. The team must meet its documented availability and rollback constraints. What targeted change should be implemented after verifying the stated cause?
- Use a measured canary or linear traffic-shift deployment with alarms and immediate rollback conditions.
- Compile the supported model for the specific edge target using SageMaker Neo or choose a compatible conversion route.
- Filter appropriate validated data-ready events and trigger retraining only after the required dataset manifest is complete.
- Add schema, integration and sample endpoint inference tests to the ML build and deployment pipeline.
Explanation
Correct Answer
D:Add schema, integration and sample endpoint inference tests to the ML build and deployment pipeline. This is the appropriate controlled remedy for continuous integration unit and endpoint contract tests. A successful training job does not prove the deployed inference API still satisfies consumer contracts. Verify the cause with: Inspect versioned request schema, unit-test coverage, container /invocations behavior, and pipeline gate logs. Then confirm: A deliberately incompatible payload causes CI to block promotion and a compatible fixture passes end to end. Reported signal: A data-science team tests only that a training script executes, but deployment repeatedly breaks because client JSON fields changed. Operational setting: During an approved rollout of an AWS ML inference workload.
Incorrect Answers
A:Use a measured canary or linear traffic-shift deployment with alarms and immediate rollback conditions. This addresses a different issue involving canary variant rollout for model changes. It does not correct the stated continuous integration unit and endpoint contract tests mechanism. The targeted response is: Add schema, integration and sample endpoint inference tests to the ML build and deployment pipeline. Case evidence: A data-science team tests only that a training script executes, but deployment repeatedly breaks because client JSON fields changed. Operational setting: During an approved rollout of an AWS ML inference workload.
B:Compile the supported model for the specific edge target using SageMaker Neo or choose a compatible conversion route. This addresses a different issue involving sagemaker neo edge compiler compatibility. It does not correct the stated continuous integration unit and endpoint contract tests mechanism. The targeted response is: Add schema, integration and sample endpoint inference tests to the ML build and deployment pipeline. Case evidence: A data-science team tests only that a training script executes, but deployment repeatedly breaks because client JSON fields changed. Operational setting: During an approved rollout of an AWS ML inference workload.
C:Filter appropriate validated data-ready events and trigger retraining only after the required dataset manifest is complete. This addresses a different issue involving eventbridge conditional retraining trigger. It does not correct the stated continuous integration unit and endpoint contract tests mechanism. The targeted response is: Add schema, integration and sample endpoint inference tests to the ML build and deployment pipeline. Case evidence: A data-science team tests only that a training script executes, but deployment repeatedly breaks because client JSON fields changed. Operational setting: During an approved rollout of an AWS ML inference workload.
Question 1075
While testing a second region for recoverable deployment of the same ML service, the following condition is observed: An ML system retrains on every S3 object event even when the object is a debug file, wasting resources and changing model versions unexpectedly. The deployment must be reproducible without manual console-only fixes. Which TWO complementary choices provide the decisive diagnosis AND corrective response? Select TWO.
- Filter appropriate validated data-ready events and trigger retraining only after the required dataset manifest is complete.
- Use a compatible SageMaker inference pipeline with ordered containers or a single packaged equivalent for shared preprocessing.
- Add schema, integration and sample endpoint inference tests to the ML build and deployment pipeline.
- Inspect EventBridge rule patterns, source events, object prefixes, and pipeline execution histories.
- Inspect model sizes, framework compatibility, per-model request rates, load latency, and isolation requirements.
Explanation
Correct Answers
A:Filter appropriate validated data-ready events and trigger retraining only after the required dataset manifest is complete. This is the appropriate controlled remedy for eventbridge conditional retraining trigger. Event-driven automation needs selective event patterns and readiness checks; every object-write event is not a retraining signal. Verify the cause with: Inspect EventBridge rule patterns, source events, object prefixes, and pipeline execution histories. Then confirm: Only the intended completed data batch launches one controlled training workflow, while unrelated objects do not. This is one part of a select-two decision; the matching evidence and corrective action are both required. Reported signal: An ML system retrains on every S3 object event even when the object is a debug file, wasting resources and changing model versions unexpectedly. Operational setting: While testing a second region for recoverable deployment of the same ML service.
D:Inspect EventBridge rule patterns, source events, object prefixes, and pipeline execution histories. This directly tests eventbridge conditional retraining trigger in the reported AWS ML workflow. Event-driven automation needs selective event patterns and readiness checks; every object-write event is not a retraining signal. It distinguishes the problem before choosing a change. The related targeted response is: Filter appropriate validated data-ready events and trigger retraining only after the required dataset manifest is complete. This is one part of a select-two decision; the matching evidence and corrective action are both required. Reported signal: An ML system retrains on every S3 object event even when the object is a debug file, wasting resources and changing model versions unexpectedly. Operational setting: While testing a second region for recoverable deployment of the same ML service.
Incorrect Answers
B:Use a compatible SageMaker inference pipeline with ordered containers or a single packaged equivalent for shared preprocessing. This selection concerns multi-container inference pipeline responsibilities rather than the required diagnostic-and-remedy pair for eventbridge conditional retraining trigger. Use evidence: Inspect EventBridge rule patterns, source events, object prefixes, and pipeline execution histories. and the matching correction: Filter appropriate validated data-ready events and trigger retraining only after the required dataset manifest is complete. Signal: An ML system retrains on every S3 object event even when the object is a debug file, wasting resources and changing model versions unexpectedly. Operational setting: While testing a second region for recoverable deployment of the same ML service.
C:Add schema, integration and sample endpoint inference tests to the ML build and deployment pipeline. This selection concerns continuous integration unit and endpoint contract tests rather than the required diagnostic-and-remedy pair for eventbridge conditional retraining trigger. Use evidence: Inspect EventBridge rule patterns, source events, object prefixes, and pipeline execution histories. and the matching correction: Filter appropriate validated data-ready events and trigger retraining only after the required dataset manifest is complete. Signal: An ML system retrains on every S3 object event even when the object is a debug file, wasting resources and changing model versions unexpectedly. Operational setting: While testing a second region for recoverable deployment of the same ML service.
E:Inspect model sizes, framework compatibility, per-model request rates, load latency, and isolation requirements. This selection concerns multi-model endpoint hosting tradeoff rather than the required diagnostic-and-remedy pair for eventbridge conditional retraining trigger. Use evidence: Inspect EventBridge rule patterns, source events, object prefixes, and pipeline execution histories. and the matching correction: Filter appropriate validated data-ready events and trigger retraining only after the required dataset manifest is complete. Signal: An ML system retrains on every S3 object event even when the object is a debug file, wasting resources and changing model versions unexpectedly. Operational setting: While testing a second region for recoverable deployment of the same ML service.
Question 1076
While reproducing a training incident in an independent SageMaker experiment, the following condition is observed: A pretrained image encoder loses useful general features after fine-tuning on a small labeled dataset using an excessively high learning rate. The experiment must be compared fairly against its prior baseline. Which TWO complementary choices provide the decisive diagnosis AND corrective response? Select TWO.
- Select a licensed, task-appropriate model and validate its performance and resource needs before a controlled deployment.
- Inspect the supervised labels, text representation, model-input format, and the desired classification output.
- Use a smaller learning rate, optionally freeze early layers, and assess staged unfreezing against a held-out set.
- Compare pretrained validation metrics, fine-tuning learning curves, and layer-wise parameter updates.
- Configure a supported SageMaker checkpoint location and resume training state on managed Spot interruption recovery.
Explanation
Correct Answers
C:Use a smaller learning rate, optionally freeze early layers, and assess staged unfreezing against a held-out set. This is the appropriate controlled remedy for fine-tuning pretrained model learning-rate stability. Large updates on small data can overwrite pretrained features; careful transfer learning limits destructive forgetting. Verify the cause with: Compare pretrained validation metrics, fine-tuning learning curves, and layer-wise parameter updates. Then confirm: The fine-tuned model improves task accuracy without collapsing pretrained representations. This is one part of a select-two decision; the matching evidence and corrective action are both required. Reported signal: A pretrained image encoder loses useful general features after fine-tuning on a small labeled dataset using an excessively high learning rate. Operational setting: While reproducing a training incident in an independent SageMaker experiment.
D:Compare pretrained validation metrics, fine-tuning learning curves, and layer-wise parameter updates. This directly tests fine-tuning pretrained model learning-rate stability in the reported AWS ML workflow. Large updates on small data can overwrite pretrained features; careful transfer learning limits destructive forgetting. It distinguishes the problem before choosing a change. The related targeted response is: Use a smaller learning rate, optionally freeze early layers, and assess staged unfreezing against a held-out set. This is one part of a select-two decision; the matching evidence and corrective action are both required. Reported signal: A pretrained image encoder loses useful general features after fine-tuning on a small labeled dataset using an excessively high learning rate. Operational setting: While reproducing a training incident in an independent SageMaker experiment.
Incorrect Answers
A:Select a licensed, task-appropriate model and validate its performance and resource needs before a controlled deployment. This selection concerns sagemaker jumpstart template provenance rather than the required diagnostic-and-remedy pair for fine-tuning pretrained model learning-rate stability. Use evidence: Compare pretrained validation metrics, fine-tuning learning curves, and layer-wise parameter updates. and the matching correction: Use a smaller learning rate, optionally freeze early layers, and assess staged unfreezing against a held-out set. Signal: A pretrained image encoder loses useful general features after fine-tuning on a small labeled dataset using an excessively high learning rate. Operational setting: While reproducing a training incident in an independent SageMaker experiment.
B:Inspect the supervised labels, text representation, model-input format, and the desired classification output. This selection concerns sagemaker built-in blazingtext choice rather than the required diagnostic-and-remedy pair for fine-tuning pretrained model learning-rate stability. Use evidence: Compare pretrained validation metrics, fine-tuning learning curves, and layer-wise parameter updates. and the matching correction: Use a smaller learning rate, optionally freeze early layers, and assess staged unfreezing against a held-out set. Signal: A pretrained image encoder loses useful general features after fine-tuning on a small labeled dataset using an excessively high learning rate. Operational setting: While reproducing a training incident in an independent SageMaker experiment.
E:Configure a supported SageMaker checkpoint location and resume training state on managed Spot interruption recovery. This selection concerns managed spot training checkpoint recovery rather than the required diagnostic-and-remedy pair for fine-tuning pretrained model learning-rate stability. Use evidence: Compare pretrained validation metrics, fine-tuning learning curves, and layer-wise parameter updates. and the matching correction: Use a smaller learning rate, optionally freeze early layers, and assess staged unfreezing against a held-out set. Signal: A pretrained image encoder loses useful general features after fine-tuning on a small labeled dataset using an excessively high learning rate. Operational setting: While reproducing a training incident in an independent SageMaker experiment.
Question 1077
During a production MLOps reliability and security review, the following condition is observed: An endpoint's client-side response time rises, but the team only examines SageMaker ModelLatency and concludes the model is slow. The owner needs an auditable resolution without broadening privileges. What postchange observation provides the strongest acceptance evidence?
- Measured training cost drops while interrupted test jobs resume from checkpoints within the delivery window.
- Representative benign fluctuations stay quiet while a known severe schema or distribution break pages the owner.
- The chosen fix reduces measured end-to-end p95 latency and component metrics explain the change.
- A test write to an approved prefix succeeds, while writes to unapproved external destinations are denied.
Explanation
Correct Answer
C:The chosen fix reduces measured end-to-end p95 latency and component metrics explain the change. This is the acceptance evidence for the described cloudwatch invocations and modellatency distinction scenario. Model compute time and overall endpoint latency are distinct signals; blaming model code without tracing overhead is premature. The corrective implementation it validates is: Identify the component contributing to total request time and optimize network or overhead separately from actual model inference. Reported signal: An endpoint's client-side response time rises, but the team only examines SageMaker ModelLatency and concludes the model is slow. Operational setting: During a production MLOps reliability and security review.
Incorrect Answers
A:Measured training cost drops while interrupted test jobs resume from checkpoints within the delivery window. This could validate spot training cost versus interruption recovery, but it does not establish that cloudwatch invocations and modellatency distinction is resolved for this workload. Confirm instead: The chosen fix reduces measured end-to-end p95 latency and component metrics explain the change. Observed problem: An endpoint's client-side response time rises, but the team only examines SageMaker ModelLatency and concludes the model is slow. Operational setting: During a production MLOps reliability and security review.
B:Representative benign fluctuations stay quiet while a known severe schema or distribution break pages the owner. This could validate model monitor violation alert calibration, but it does not establish that cloudwatch invocations and modellatency distinction is resolved for this workload. Confirm instead: The chosen fix reduces measured end-to-end p95 latency and component metrics explain the change. Observed problem: An endpoint's client-side response time rises, but the team only examines SageMaker ModelLatency and concludes the model is slow. Operational setting: During a production MLOps reliability and security review.
D:A test write to an approved prefix succeeds, while writes to unapproved external destinations are denied. This could validate cross-account s3 bucket exfiltration protection, but it does not establish that cloudwatch invocations and modellatency distinction is resolved for this workload. Confirm instead: The chosen fix reduces measured end-to-end p95 latency and component metrics explain the change. Observed problem: An endpoint's client-side response time rises, but the team only examines SageMaker ModelLatency and concludes the model is slow. Operational setting: During a production MLOps reliability and security review.
Question 1078
During a controlled replay of the data-preparation workflow in a separate ML staging account, the following condition is observed: New event producers change a numerical source field to strings while legacy objects remain numeric, causing the shared Glue transformation to fail. The dataset contract must remain reproducible for a backfill. What targeted change should be implemented after verifying the stated cause?
- Compact the partitioned dataset into appropriately sized Parquet objects without changing the record keys or schema.
- Remove or pseudonymize unnecessary PII before annotation and restrict access to the token mapping under separate controls.
- Define explicit compatible conversions and schema contracts, isolate incompatible versions, and validate before publishing features.
- Perform point-in-time correct joins using the most recent eligible pre-prediction feature record, not the globally latest value.
Explanation
Correct Answer
C:Define explicit compatible conversions and schema contracts, isolate incompatible versions, and validate before publishing features. This is the appropriate controlled remedy for glue schema evolution with mixed input types. A catalog entry does not automatically reconcile incompatible field types across all source partitions. Verify the cause with: Compare source schemas by partition, Glue Data Catalog types, and failed cast rows for the evolving field. Then confirm: Both supported source versions produce the agreed typed feature, and invalid values enter an auditable reject path. Reported signal: New event producers change a numerical source field to strings while legacy objects remain numeric, causing the shared Glue transformation to fail. Operational setting: During a controlled replay of the data-preparation workflow in a separate ML staging account.
Incorrect Answers
A:Compact the partitioned dataset into appropriately sized Parquet objects without changing the record keys or schema. This addresses a different issue involving columnar file compaction for small objects. It does not correct the stated glue schema evolution with mixed input types mechanism. The targeted response is: Define explicit compatible conversions and schema contracts, isolate incompatible versions, and validate before publishing features. Case evidence: New event producers change a numerical source field to strings while legacy objects remain numeric, causing the shared Glue transformation to fail. Operational setting: During a controlled replay of the data-preparation workflow in a separate ML staging account.
B:Remove or pseudonymize unnecessary PII before annotation and restrict access to the token mapping under separate controls. This addresses a different issue involving pii masking before model labeling. It does not correct the stated glue schema evolution with mixed input types mechanism. The targeted response is: Define explicit compatible conversions and schema contracts, isolate incompatible versions, and validate before publishing features. Case evidence: New event producers change a numerical source field to strings while legacy objects remain numeric, causing the shared Glue transformation to fail. Operational setting: During a controlled replay of the data-preparation workflow in a separate ML staging account.
D:Perform point-in-time correct joins using the most recent eligible pre-prediction feature record, not the globally latest value. This addresses a different issue involving point-in-time feature data leakage. It does not correct the stated glue schema evolution with mixed input types mechanism. The targeted response is: Define explicit compatible conversions and schema contracts, isolate incompatible versions, and validate before publishing features. Case evidence: New event producers change a numerical source field to strings while legacy objects remain numeric, causing the shared Glue transformation to fail. Operational setting: During a controlled replay of the data-preparation workflow in a separate ML staging account.
Question 1079
During a production MLOps reliability and security review, the following condition is observed: A recommendation endpoint drops peak requests with HTTP 5xx as invocations per instance rise, while the dashboard averages latency across an entire day. The owner needs an auditable resolution without broadening privileges. Which measured outcome would confirm that the correction worked as intended?
- A query isolates the failing job versions and recurring error class within the incident interval.
- Peak-window 5xx rate and p95 latency fall below service objectives with verified alarms.
- Auditors see reproducible subgroup statistics and alarms for material disparities rather than only global totals.
- An authorized training read emits attributable S3 object-access events while alerting detects unexpected access.
Explanation
Correct Answer
B:Peak-window 5xx rate and p95 latency fall below service objectives with verified alarms. This is the acceptance evidence for the described cloudwatch 5xxerrors endpoint saturation scenario. Daily averages can conceal short capacity failures; monitoring must use an incident-relevant interval. The corrective implementation it validates is: Configure actionable alarms, capacity, and supported scaling or backpressure based on the burst window. Reported signal: A recommendation endpoint drops peak requests with HTTP 5xx as invocations per instance rise, while the dashboard averages latency across an entire day. Operational setting: During a production MLOps reliability and security review.
Incorrect Answers
A:A query isolates the failing job versions and recurring error class within the incident interval. This could validate cloudwatch logs insights error correlation, but it does not establish that cloudwatch 5xxerrors endpoint saturation is resolved for this workload. Confirm instead: Peak-window 5xx rate and p95 latency fall below service objectives with verified alarms. Observed problem: A recommendation endpoint drops peak requests with HTTP 5xx as invocations per instance rise, while the dashboard averages latency across an entire day. Operational setting: During a production MLOps reliability and security review.
C:Auditors see reproducible subgroup statistics and alarms for material disparities rather than only global totals. This could validate clarify postdeployment group bias monitoring, but it does not establish that cloudwatch 5xxerrors endpoint saturation is resolved for this workload. Confirm instead: Peak-window 5xx rate and p95 latency fall below service objectives with verified alarms. Observed problem: A recommendation endpoint drops peak requests with HTTP 5xx as invocations per instance rise, while the dashboard averages latency across an entire day. Operational setting: During a production MLOps reliability and security review.
D:An authorized training read emits attributable S3 object-access events while alerting detects unexpected access. This could validate cloudtrail training data access audit coverage, but it does not establish that cloudwatch 5xxerrors endpoint saturation is resolved for this workload. Confirm instead: Peak-window 5xx rate and p95 latency fall below service objectives with verified alarms. Observed problem: A recommendation endpoint drops peak requests with HTTP 5xx as invocations per instance rise, while the dashboard averages latency across an entire day. Operational setting: During a production MLOps reliability and security review.
Question 1080
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 corrective action most directly addresses this ML workflow problem?
- Compact the partitioned dataset into appropriately sized Parquet objects without changing the record keys or schema.
- Drop leakage-prone IDs or adopt controlled hashing or suitable learned encodings with stable unknown-category handling.
- Define event-time windows with appropriate lateness handling and a documented revision policy for late records.
- Use controlled incremental extraction or a suitable read replica and throttle export jobs within the approved data freshness target.
Explanation
Correct Answer
A:Compact the partitioned dataset into appropriately sized Parquet objects without changing the record keys or schema. This is the appropriate controlled remedy for columnar file compaction for small objects. Excessive small files create metadata and request overhead even with a columnar data format. Verify the cause with: Measure file-size distribution, object request count, Spark task scheduling overhead, and read throughput. Then confirm: Training read time and object operations decline while the resulting row count and checksums match. 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
B:Drop leakage-prone IDs or adopt controlled hashing or suitable learned encodings with stable unknown-category handling. This addresses a different issue involving high-cardinality categorical encoding. It does not correct the stated columnar file compaction for small objects mechanism. The targeted response is: Compact the partitioned dataset into appropriately sized Parquet objects without changing the record keys or schema. Case evidence: 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.
C:Define event-time windows with appropriate lateness handling and a documented revision policy for late records. This addresses a different issue involving streaming feature rolling window correctness. It does not correct the stated columnar file compaction for small objects mechanism. The targeted response is: Compact the partitioned dataset into appropriately sized Parquet objects without changing the record keys or schema. Case evidence: 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:Use controlled incremental extraction or a suitable read replica and throttle export jobs within the approved data freshness target. This addresses a different issue involving rds extraction overload into ml. It does not correct the stated columnar file compaction for small objects mechanism. The targeted response is: Compact the partitioned dataset into appropriately sized Parquet objects without changing the record keys or schema. Case evidence: 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.
