Amazon AWS Certified Machine Learning Engineer - Associate MLA-C01 Practice Test 12

100% Latest & Updated Amazon AWS Certified Machine Learning Engineer - Associate MLA-C01 Practice Test Questions, Exam Dumps & Verified Answers!
30 Days Free Updates, Instant Download!

Question 110

A release gate must account for this limitation while using: Perform timestamp-aware historical feature joins using available offline Feature Store records. This relates to a situation where a model that predicts defaults at loan origination currently joins the customer's current account balance rather than the balance known on the origination date. Which limitation is most important when deciding whether this change is suitable?

An ordinal code assumes meaningful order and is wrong for unordered product labels.
Worker agreement does not prove labels are unbiased; Ground Truth ceased onboarding new customers on July 30, 2026.
Binning discards within-bin detail and can amplify fairness concerns around thresholds.
A simple latest-value join can yield impressive but invalid offline accuracy.

Explanation

Correct Answer

This is a real operational limit for the described design: A simple latest-value join can yield impressive but invalid offline accuracy. The associated control is: Perform timestamp-aware historical feature joins using available offline Feature Store records. Its intended result is: Training rows contain only feature values that existed when each prediction would have been made. This limitation must still be managed.

Incorrect Answers

That concern belongs to a different design: A risk rating has meaningful ordered tiers low, medium, and high, and a product scoring model should retain that order in a compact feature. The relevant boundary for this case is: A simple latest-value join can yield impressive but invalid offline accuracy. These are different constraints.

That concern belongs to a different design: A retailer with an existing SageMaker Ground Truth account has thousands of ambiguous product images and needs human labels with periodic consensus verification for supervised classification. The relevant boundary for this case is: A simple latest-value join can yield impressive but invalid offline accuracy. These are different constraints.

That concern belongs to a different design: A credit risk team wants a smooth age measurement grouped into stable policy bands rather than a continuously varying predictor for a regulated report. The relevant boundary for this case is: A simple latest-value join can yield impressive but invalid offline accuracy. These are different constraints.

Question 111

The technical review is assessing an important boundary of this proposal: Run the preprocessing logic as a SageMaker Processing job with versioned code and output paths. This relates to a situation where a team needs to run a one-off Python preprocessing script using controlled container libraries and write transformed data to S3 without a long-lived notebook. Which limitation is most important when deciding whether this change is suitable?

A simple latest-value join can yield impressive but invalid offline accuracy.
Incorrect event-time handling or checkpoint failures can produce late or duplicated aggregates.
A Processing job does not itself automatically deploy or train a prediction endpoint.
One-hot features can become extremely wide when cardinality is high.

Explanation

Correct Answer

This is a real operational limit for the described design: A Processing job does not itself automatically deploy or train a prediction endpoint. The associated control is: Run the preprocessing logic as a SageMaker Processing job with versioned code and output paths. Its intended result is: Transformations execute as a reproducible managed job with bounded resources. This limitation must still be managed.

Incorrect Answers

That concern belongs to a different design: A model that predicts defaults at loan origination currently joins the customer's current account balance rather than the balance known on the origination date. The relevant boundary for this case is: A Processing job does not itself automatically deploy or train a prediction endpoint. These are different constraints.

That concern belongs to a different design: A logistics stream requires stateful rolling five-minute counts from Kinesis records without launching and managing an always-on EC2 Spark cluster. The relevant boundary for this case is: A Processing job does not itself automatically deploy or train a prediction endpoint. These are different constraints.

That concern belongs to a different design: A small product category field has red, blue, and green values without any intrinsic order, but an engineer currently maps them to 1, 2, and 3. The relevant boundary for this case is: A Processing job does not itself automatically deploy or train a prediction endpoint. These are different constraints.

Question 112

During risk sign-off, the team must assess this approved design step: Fit or choose a supported text tokenizer and reuse its saved vocabulary and preprocessing settings for inference. This relates to a situation where a service classification model receives free-text customer tickets and must convert words into consistent model input tokens while preserving the same vocabulary during inference. Which limitation is most important when deciding whether this change is suitable?

A log transform does not automatically fix mislabeled records and requires care with negative values.
A recipe designed on a small sample must be tested against unusual production values.
Changing tokenization between training and serving can invalidate learned embeddings.
Worker agreement does not prove labels are unbiased; Ground Truth ceased onboarding new customers on July 30, 2026.

Explanation

Correct Answer

This is a real operational limit for the described design: Changing tokenization between training and serving can invalidate learned embeddings. The associated control is: Fit or choose a supported text tokenizer and reuse its saved vocabulary and preprocessing settings for inference. Its intended result is: Text inputs are converted into consistent token IDs accepted by the model. This limitation must still be managed.

Incorrect Answers

That concern belongs to a different design: Transaction dollar amounts span several orders of magnitude, and a linear model is dominated by the largest outliers despite consistent labels. The relevant boundary for this case is: Changing tokenization between training and serving can invalidate learned embeddings. These are different constraints.

That concern belongs to a different design: A data operations analyst cleans inconsistent dates and malformed postal codes in a structured dataset using an auditable sequence of low-code rules. The relevant boundary for this case is: Changing tokenization between training and serving can invalidate learned embeddings. These are different constraints.

That concern belongs to a different design: A retailer with an existing SageMaker Ground Truth account has thousands of ambiguous product images and needs human labels with periodic consensus verification for supervised classification. The relevant boundary for this case is: Changing tokenization between training and serving can invalidate learned embeddings. These are different constraints.

Question 113

A single ML platform team is handling two distinct changes for the same system. Workstream A: A data lake receives a daily product feed that must have nonnull customer keys and positive amounts before it enters feature processing. Workstream B: A hospital exports clinical text into a shared ML feature workspace and must remove patient identifiers while retaining useful nonidentifying signals. Select exactly TWO measures, one that directly resolves each workstream.

Deidentify or tokenize personal identifiers before shared processing and restrict the raw originals.
Run supported SageMaker Clarify pretraining bias analysis and investigate the observed disparity.
Choose a compatible shared high-performance training filesystem such as FSx for Lustre when justified.
Define a Data Quality ruleset and fail or quarantine datasets that violate required checks.
Refine annotation instructions, adjudicate ambiguous examples, and relabel affected training data.

Explanation

Correct Answers

This action fits the situation: A hospital exports clinical text into a shared ML feature workspace and must remove patient identifiers while retaining useful nonidentifying signals. It implements the needed technical control: Deidentify or tokenize personal identifiers before shared processing and restrict the raw originals. The intended outcome is that modelers can use an approved dataset without directly exposing patient identity fields. The important limitation is that removing names alone may not prevent reidentification through rare attribute combinations.

This action fits the situation: A data lake receives a daily product feed that must have nonnull customer keys and positive amounts before it enters feature processing. It implements the needed technical control: Define a Data Quality ruleset and fail or quarantine datasets that violate required checks. The intended outcome is that invalid daily datasets are identified before corrupting model training inputs. The important limitation is that a passing schema check cannot guarantee every record's business meaning is correct.

Incorrect Answers

That alternative addresses a different task: A lender already using SageMaker Clarify has much fewer approved loans from one demographic subgroup and needs a measured dataset bias check before modeling. The selected workloads instead require Rule based Glue data validation, Masked sensitive inputs.

That alternative addresses a different task: A training job repeatedly reads a shared block of feature matrices and needs much faster parallel data access than the current single-volume setup. The selected workloads instead require Rule based Glue data validation, Masked sensitive inputs.

That alternative addresses a different task: A text moderation classifier has confusing class definitions and annotators disagree about which of two labels applies to borderline tickets. The selected workloads instead require Rule based Glue data validation, Masked sensitive inputs.

Question 114

A single ML platform team is handling two distinct changes for the same system. Workstream A: A lender already using SageMaker Clarify has much fewer approved loans from one demographic subgroup and needs a measured dataset bias check before modeling. Workstream B: A regulated ML team stores customer data in S3 and wants control over encryption-key access by execution role and audited key usage. Select exactly TWO measures, one that directly resolves each workstream.

Define a Data Quality ruleset and fail or quarantine datasets that violate required checks.
Use S3 server-side encryption with an appropriately restricted customer-managed KMS key.
Run supported SageMaker Clarify pretraining bias analysis and investigate the observed disparity.
Refine annotation instructions, adjudicate ambiguous examples, and relabel affected training data.
Record immutable input references, job identifiers, schemas, and transformation versions in the data catalog lineage.

Explanation

Correct Answers

This action fits the situation: A regulated ML team stores customer data in S3 and wants control over encryption-key access by execution role and audited key usage. It implements the needed technical control: Use S3 server-side encryption with an appropriately restricted customer-managed KMS key. The intended outcome is that stored training objects are encrypted with an auditable key access boundary. The important limitation is that the processing role also needs KMS permissions for decrypting authorized objects.

This action fits the situation: A lender already using SageMaker Clarify has much fewer approved loans from one demographic subgroup and needs a measured dataset bias check before modeling. It implements the needed technical control: Run supported SageMaker Clarify pretraining bias analysis and investigate the observed disparity. The intended outcome is that data scientists can quantify group representation or label imbalance before model fitting. The important limitation is that a numerical bias score is not proof of legal fairness; Clarify is available only to existing customers after July 30, 2026.

Incorrect Answers

That alternative addresses a different task: A data lake receives a daily product feed that must have nonnull customer keys and positive amounts before it enters feature processing. The selected workloads instead require Pretraining representation bias, Encryption key segregation.

That alternative addresses a different task: A text moderation classifier has confusing class definitions and annotators disagree about which of two labels applies to borderline tickets. The selected workloads instead require Pretraining representation bias, Encryption key segregation.

That alternative addresses a different task: An auditor must trace a prediction feature back to its original S3 object, Glue job run, and data processing version after a model incident. The selected workloads instead require Pretraining representation bias, Encryption key segregation.

Question 115

A single ML platform team is handling two distinct changes for the same system. Workstream A: Only 2 percent of historical medical claims are fraudulent, and random splitting sometimes leaves too few fraud examples in evaluation data. Workstream B: Model input CSV files are written into an S3 bucket that mistakenly has a permissive ACL, and compliance forbids anonymous access regardless of the object setting. Select exactly TWO measures, one that directly resolves each workstream.

Enable S3 Block Public Access and apply a restricted bucket policy for approved principals.
Use a stratified train-validation-test split that preserves class proportions where appropriate.
Refine annotation instructions, adjudicate ambiguous examples, and relabel affected training data.
Record immutable input references, job identifiers, schemas, and transformation versions in the data catalog lineage.
Apply suitable class weights or resampling on training data and evaluate minority recall or F1.

Explanation

Correct Answers

This action fits the situation: Model input CSV files are written into an S3 bucket that mistakenly has a permissive ACL, and compliance forbids anonymous access regardless of the object setting. It implements the needed technical control: Enable S3 Block Public Access and apply a restricted bucket policy for approved principals. The intended outcome is that public ACL or policy misconfigurations cannot expose the protected dataset through blocked public access paths. The important limitation is that block Public Access does not replace least-privilege private access review.

This action fits the situation: Only 2 percent of historical medical claims are fraudulent, and random splitting sometimes leaves too few fraud examples in evaluation data. It implements the needed technical control: Use a stratified train-validation-test split that preserves class proportions where appropriate. The intended outcome is that each split retains a representative number of rare positives for evaluation. The important limitation is that stratification cannot prevent leakage from related records belonging to the same customer.

Incorrect Answers

That alternative addresses a different task: A text moderation classifier has confusing class definitions and annotators disagree about which of two labels applies to borderline tickets. The selected workloads instead require Stratified minority label split, Block accidental public buckets.

That alternative addresses a different task: An auditor must trace a prediction feature back to its original S3 object, Glue job run, and data processing version after a model incident. The selected workloads instead require Stratified minority label split, Block accidental public buckets.

That alternative addresses a different task: A defective-parts classifier predicts the majority good class for nearly every example because defective units are rare and accuracy appears high. The selected workloads instead require Stratified minority label split, Block accidental public buckets.

Question 116

A single ML platform team is handling two distinct changes for the same system. Workstream A: An energy-demand forecasting model was evaluated using randomly shuffled observations even though production predicts future days. Workstream B: A feature producer changes a numeric transaction count to a string, causing model inference preprocessing to fail despite successful S3 ingestion. Select exactly TWO measures, one that directly resolves each workstream.

Deduplicate appropriate rows and group records by entity before assigning train/test partitions.
Use a time-ordered split with training history strictly preceding the evaluation interval.
Enforce schema and type validation at ingestion and quarantine incompatible records.
Enable S3 Block Public Access and apply a restricted bucket policy for approved principals.
Record immutable input references, job identifiers, schemas, and transformation versions in the data catalog lineage.

Explanation

Correct Answers

This action fits the situation: An energy-demand forecasting model was evaluated using randomly shuffled observations even though production predicts future days. It implements the needed technical control: Use a time-ordered split with training history strictly preceding the evaluation interval. The intended outcome is that offline performance better reflects predictions on future unseen periods. The important limitation is that seasonal changes and forecasting horizon still need representative backtesting windows.

This action fits the situation: A feature producer changes a numeric transaction count to a string, causing model inference preprocessing to fail despite successful S3 ingestion. It implements the needed technical control: Enforce schema and type validation at ingestion and quarantine incompatible records. The intended outcome is that unexpected feature types are rejected before they reach the serving pipeline. The important limitation is that accepting every string as convertible can silently change feature meaning.

Incorrect Answers

That alternative addresses a different task: The same customer appears in nearly identical records across training and evaluation splits, yielding unrealistic customer churn accuracy. The selected workloads instead require Chronological evaluation window, Feature schema contract.

That alternative addresses a different task: Model input CSV files are written into an S3 bucket that mistakenly has a permissive ACL, and compliance forbids anonymous access regardless of the object setting. The selected workloads instead require Chronological evaluation window, Feature schema contract.

That alternative addresses a different task: An auditor must trace a prediction feature back to its original S3 object, Glue job run, and data processing version after a model incident. The selected workloads instead require Chronological evaluation window, Feature schema contract.

Question 117

A single ML platform team is handling two distinct changes for the same system. Workstream A: A hospital exports clinical text into a shared ML feature workspace and must remove patient identifiers while retaining useful nonidentifying signals. Workstream B: An auditor must trace a prediction feature back to its original S3 object, Glue job run, and data processing version after a model incident. Select exactly TWO measures, one that directly resolves each workstream.

Record immutable input references, job identifiers, schemas, and transformation versions in the data catalog lineage.
Define a Data Quality ruleset and fail or quarantine datasets that violate required checks.
Choose a compatible shared high-performance training filesystem such as FSx for Lustre when justified.
Apply suitable class weights or resampling on training data and evaluate minority recall or F1.
Deidentify or tokenize personal identifiers before shared processing and restrict the raw originals.

Explanation

Correct Answers

This action fits the situation: An auditor must trace a prediction feature back to its original S3 object, Glue job run, and data processing version after a model incident. It implements the needed technical control: Record immutable input references, job identifiers, schemas, and transformation versions in the data catalog lineage. The intended outcome is that incident reviewers can reconstruct which source data and processing steps produced a feature set. The important limitation is that a catalog entry alone is insufficient if actual input object versions were not retained.

Incorrect Answers

That alternative addresses a different task: A data lake receives a daily product feed that must have nonnull customer keys and positive amounts before it enters feature processing. The selected workloads instead require Masked sensitive inputs, Provenance across transformations.

That alternative addresses a different task: A training job repeatedly reads a shared block of feature matrices and needs much faster parallel data access than the current single-volume setup. The selected workloads instead require Masked sensitive inputs, Provenance across transformations.

That alternative addresses a different task: A defective-parts classifier predicts the majority good class for nearly every example because defective units are rare and accuracy appears high. The selected workloads instead require Masked sensitive inputs, Provenance across transformations.

Question 118

A cloud ML change request describes the following issue: A regulated ML team stores customer data in S3 and wants control over encryption-key access by execution role and audited key usage. Which action best addresses the actual requirement?

Apply suitable class weights or resampling on training data and evaluate minority recall or F1.
Choose a compatible shared high-performance training filesystem such as FSx for Lustre when justified.
Use S3 server-side encryption with an appropriately restricted customer-managed KMS key.
Define a Data Quality ruleset and fail or quarantine datasets that violate required checks.

Explanation

Correct Answer

Incorrect Answers

This action is appropriate for another requirement: A defective-parts classifier predicts the majority good class for nearly every example because defective units are rare and accuracy appears high. It does not solve the presented constraint: A regulated ML team stores customer data in S3 and wants control over encryption-key access by execution role and audited key usage. The matching action for this case is: Use S3 server-side encryption with an appropriately restricted customer-managed KMS key.

This action is appropriate for another requirement: A training job repeatedly reads a shared block of feature matrices and needs much faster parallel data access than the current single-volume setup. It does not solve the presented constraint: A regulated ML team stores customer data in S3 and wants control over encryption-key access by execution role and audited key usage. The matching action for this case is: Use S3 server-side encryption with an appropriately restricted customer-managed KMS key.

This action is appropriate for another requirement: A data lake receives a daily product feed that must have nonnull customer keys and positive amounts before it enters feature processing. It does not solve the presented constraint: A regulated ML team stores customer data in S3 and wants control over encryption-key access by execution role and audited key usage. The matching action for this case is: Use S3 server-side encryption with an appropriately restricted customer-managed KMS key.

Question 119

A team is choosing a specific control for this workload: Model input CSV files are written into an S3 bucket that mistakenly has a permissive ACL, and compliance forbids anonymous access regardless of the object setting. Which action best addresses the actual requirement?

Enable S3 Block Public Access and apply a restricted bucket policy for approved principals.
Record immutable input references, job identifiers, schemas, and transformation versions in the data catalog lineage.
Choose a compatible shared high-performance training filesystem such as FSx for Lustre when justified.
Use a time-ordered split with training history strictly preceding the evaluation interval.

Explanation

Correct Answer

Incorrect Answers

This action is appropriate for another requirement: An auditor must trace a prediction feature back to its original S3 object, Glue job run, and data processing version after a model incident. It does not solve the presented constraint: Model input CSV files are written into an S3 bucket that mistakenly has a permissive ACL, and compliance forbids anonymous access regardless of the object setting. The matching action for this case is: Enable S3 Block Public Access and apply a restricted bucket policy for approved principals.

This action is appropriate for another requirement: A training job repeatedly reads a shared block of feature matrices and needs much faster parallel data access than the current single-volume setup. It does not solve the presented constraint: Model input CSV files are written into an S3 bucket that mistakenly has a permissive ACL, and compliance forbids anonymous access regardless of the object setting. The matching action for this case is: Enable S3 Block Public Access and apply a restricted bucket policy for approved principals.

This action is appropriate for another requirement: An energy-demand forecasting model was evaluated using randomly shuffled observations even though production predicts future days. It does not solve the presented constraint: Model input CSV files are written into an S3 bucket that mistakenly has a permissive ACL, and compliance forbids anonymous access regardless of the object setting. The matching action for this case is: Enable S3 Block Public Access and apply a restricted bucket policy for approved principals.

Question 120

An ML engineering team faces this production constraint: A feature producer changes a numeric transaction count to a string, causing model inference preprocessing to fail despite successful S3 ingestion. Which action best addresses the actual requirement?

Enable S3 Block Public Access and apply a restricted bucket policy for approved principals.
Deduplicate appropriate rows and group records by entity before assigning train/test partitions.
Enforce schema and type validation at ingestion and quarantine incompatible records.
Choose a compatible shared high-performance training filesystem such as FSx for Lustre when justified.

Explanation

Correct Answer

Incorrect Answers

This action is appropriate for another requirement: Model input CSV files are written into an S3 bucket that mistakenly has a permissive ACL, and compliance forbids anonymous access regardless of the object setting. It does not solve the presented constraint: A feature producer changes a numeric transaction count to a string, causing model inference preprocessing to fail despite successful S3 ingestion. The matching action for this case is: Enforce schema and type validation at ingestion and quarantine incompatible records.

This action is appropriate for another requirement: The same customer appears in nearly identical records across training and evaluation splits, yielding unrealistic customer churn accuracy. It does not solve the presented constraint: A feature producer changes a numeric transaction count to a string, causing model inference preprocessing to fail despite successful S3 ingestion. The matching action for this case is: Enforce schema and type validation at ingestion and quarantine incompatible records.

This action is appropriate for another requirement: A training job repeatedly reads a shared block of feature matrices and needs much faster parallel data access than the current single-volume setup. It does not solve the presented constraint: A feature producer changes a numeric transaction count to a string, causing model inference preprocessing to fail despite successful S3 ingestion. The matching action for this case is: Enforce schema and type validation at ingestion and quarantine incompatible records.

Amazon AWS Certified Machine Learning Engineer - Associate MLA-C01 Exam Dumps, Practice Test Questions

Question 110

Explanation

Correct Answer

Incorrect Answers

Question 111

Explanation

Correct Answer

Incorrect Answers

Question 112

Explanation

Correct Answer

Incorrect Answers

Question 113

Explanation

Correct Answers

Incorrect Answers

Question 114

Explanation

Correct Answers

Incorrect Answers

Question 115

Explanation

Correct Answers

Incorrect Answers

Question 116

Explanation

Correct Answers

Incorrect Answers

Question 117

Explanation

Correct Answers

Incorrect Answers

Question 118

Explanation

Correct Answer

Incorrect Answers

Question 119

Explanation

Correct Answer

Incorrect Answers

Question 120

Explanation

Correct Answer

Incorrect Answers