1
AWS Certified Machine
Learning Engineer –
Associate (MLA-C01) 150
Multiple-Choice Questions
— Advanced / Mixed
Difficulty a well detailed
one written
and graded A+
upgraded
, 2
DOMAIN 1: DATA PREPARATION FOR MACHINE LEARNING (28%)
Question 1
A financial services company collects millions of streaming transactions per second from
multiple global sources. The ML engineer needs to ingest this data in real time, perform light
transformation, and store it for both real-time feature lookup and batch training. Which
combination of AWS services should the engineer use?
A. Amazon Kinesis Data Streams → AWS Lambda → Amazon S3 (batch) and Amazon DynamoDB
(online)
B. Amazon MSK → AWS Glue Streaming → Amazon S3 (batch) and SageMaker Feature Store
(online)
C. Amazon Kinesis Data Firehose → Amazon S3 → SageMaker Data Wrangler for transformation
D. Amazon SQS → AWS Lambda → Amazon Redshift for both online and batch storage
-” detailed answer 100 % correct :-”B
Rationale: Amazon MSK (managed Kafka) can handle high-volume streaming ingestion. AWS
Glue Streaming provides serverless transformation, and SageMaker Feature Store supports both
an online store (low-latency inference lookups) and an offline store (batch training data). Kinesis
Data Streams + Lambda would not scale as efficiently for millions of events per second. Kinesis
Firehose is for near-real-time delivery, not true streaming ingestion. SQS is message queuing,
not designed for streaming analytics at this scale.
Question 2
An ML engineer is preparing a dataset stored in CSV format in Amazon S3. The dataset contains
500 million rows with 200 columns. The engineer needs to perform exploratory data analysis,
detect outliers, and generate statistical summaries before training. Which approach provides
the MOST efficient processing?
A. Load the entire dataset into a SageMaker notebook instance using pandas and perform EDA
, 3
B. Use SageMaker Data Wrangler with a data flow to sample, analyze, and visualize the dataset
C. Use AWS Glue to convert CSV to Parquet, then run Athena queries for statistical analysis
D. Use Amazon EMR with Spark to process the full dataset and generate summary statistics
-” detailed answer 100 % correct :-”B
Rationale: SageMaker Data Wrangler is purpose-built for visual data preparation at scale,
providing built-in transforms, statistical summaries, and outlier detection without requiring
extensive coding. Loading 500M rows into a notebook would cause memory issues. Glue +
Athena is more suited for querying, not interactive EDA. EMR + Spark would work but requires
more operational overhead than Data Wrangler's managed experience.
Question 3
A healthcare company needs to detect potential bias in their training data before model
development. The dataset contains sensitive attributes including age, gender, and ethnicity.
Which SageMaker service should the ML engineer use?
A. SageMaker Model Monitor
B. SageMaker Clarify
C. SageMaker Data Wrangler
D. SageMaker Ground Truth
-” detailed answer 100 % correct :-”B
Rationale: SageMaker Clarify is specifically designed to detect pre-training bias by analyzing
feature distributions across sensitive attributes and providing fairness metrics. Model Monitor
detects drift after deployment, not pre-training bias. Data Wrangler is for data transformation,
not bias detection. Ground Truth is for data labeling, not bias analysis.
Question 4
An ML engineer is designing a feature store for a recommendation system. Features must be
available for real-time inference with sub-10ms latency and also for batch training jobs that run
weekly. How should the engineer configure SageMaker Feature Store?
A. Use only the online store with a large instance type for both inference and batch reads
, 4
B. Use only the offline store and cache features in Redis for inference
C. Configure both online and offline stores; write features to both simultaneously
D. Use the online store for inference and export data to S3 weekly for batch training
-” detailed answer 100 % correct :-”C
Rationale: SageMaker Feature Store provides both an online store (low-latency lookups for
inference) and an offline store (optimized for batch training). Writing to both stores
simultaneously ensures feature consistency across online and offline use cases. Option A would
not scale for batch. Option B adds unnecessary complexity. Option D creates data consistency
issues between online and offline.
Question 5
A company stores petabytes of unstructured data in Amazon S3. The ML engineer needs to
catalog this data, infer schemas, and prepare it for machine learning pipelines. Which service
should the engineer use?
A. Amazon Athena
B. AWS Glue Crawlers and Data Catalog
C. SageMaker Data Wrangler
D. Amazon EMR
-” detailed answer 100 % correct :-”B
Rationale: AWS Glue Crawlers automatically scan data in S3, infer schemas, and populate the
Glue Data Catalog. Athena queries the catalog but doesn't catalog data. Data Wrangler is for
transformation, not cataloging. EMR requires manual schema definition.
Question 6
An ML engineer needs to transform raw JSON logs into a structured format suitable for model
training. The transformation involves complex joins, aggregations, and feature engineering
across multiple data sources. The workload runs daily and must be cost-effective. Which service
should the engineer use?
A. SageMaker Data Wrangler with scheduled data flows
AWS Certified Machine
Learning Engineer –
Associate (MLA-C01) 150
Multiple-Choice Questions
— Advanced / Mixed
Difficulty a well detailed
one written
and graded A+
upgraded
, 2
DOMAIN 1: DATA PREPARATION FOR MACHINE LEARNING (28%)
Question 1
A financial services company collects millions of streaming transactions per second from
multiple global sources. The ML engineer needs to ingest this data in real time, perform light
transformation, and store it for both real-time feature lookup and batch training. Which
combination of AWS services should the engineer use?
A. Amazon Kinesis Data Streams → AWS Lambda → Amazon S3 (batch) and Amazon DynamoDB
(online)
B. Amazon MSK → AWS Glue Streaming → Amazon S3 (batch) and SageMaker Feature Store
(online)
C. Amazon Kinesis Data Firehose → Amazon S3 → SageMaker Data Wrangler for transformation
D. Amazon SQS → AWS Lambda → Amazon Redshift for both online and batch storage
-” detailed answer 100 % correct :-”B
Rationale: Amazon MSK (managed Kafka) can handle high-volume streaming ingestion. AWS
Glue Streaming provides serverless transformation, and SageMaker Feature Store supports both
an online store (low-latency inference lookups) and an offline store (batch training data). Kinesis
Data Streams + Lambda would not scale as efficiently for millions of events per second. Kinesis
Firehose is for near-real-time delivery, not true streaming ingestion. SQS is message queuing,
not designed for streaming analytics at this scale.
Question 2
An ML engineer is preparing a dataset stored in CSV format in Amazon S3. The dataset contains
500 million rows with 200 columns. The engineer needs to perform exploratory data analysis,
detect outliers, and generate statistical summaries before training. Which approach provides
the MOST efficient processing?
A. Load the entire dataset into a SageMaker notebook instance using pandas and perform EDA
, 3
B. Use SageMaker Data Wrangler with a data flow to sample, analyze, and visualize the dataset
C. Use AWS Glue to convert CSV to Parquet, then run Athena queries for statistical analysis
D. Use Amazon EMR with Spark to process the full dataset and generate summary statistics
-” detailed answer 100 % correct :-”B
Rationale: SageMaker Data Wrangler is purpose-built for visual data preparation at scale,
providing built-in transforms, statistical summaries, and outlier detection without requiring
extensive coding. Loading 500M rows into a notebook would cause memory issues. Glue +
Athena is more suited for querying, not interactive EDA. EMR + Spark would work but requires
more operational overhead than Data Wrangler's managed experience.
Question 3
A healthcare company needs to detect potential bias in their training data before model
development. The dataset contains sensitive attributes including age, gender, and ethnicity.
Which SageMaker service should the ML engineer use?
A. SageMaker Model Monitor
B. SageMaker Clarify
C. SageMaker Data Wrangler
D. SageMaker Ground Truth
-” detailed answer 100 % correct :-”B
Rationale: SageMaker Clarify is specifically designed to detect pre-training bias by analyzing
feature distributions across sensitive attributes and providing fairness metrics. Model Monitor
detects drift after deployment, not pre-training bias. Data Wrangler is for data transformation,
not bias detection. Ground Truth is for data labeling, not bias analysis.
Question 4
An ML engineer is designing a feature store for a recommendation system. Features must be
available for real-time inference with sub-10ms latency and also for batch training jobs that run
weekly. How should the engineer configure SageMaker Feature Store?
A. Use only the online store with a large instance type for both inference and batch reads
, 4
B. Use only the offline store and cache features in Redis for inference
C. Configure both online and offline stores; write features to both simultaneously
D. Use the online store for inference and export data to S3 weekly for batch training
-” detailed answer 100 % correct :-”C
Rationale: SageMaker Feature Store provides both an online store (low-latency lookups for
inference) and an offline store (optimized for batch training). Writing to both stores
simultaneously ensures feature consistency across online and offline use cases. Option A would
not scale for batch. Option B adds unnecessary complexity. Option D creates data consistency
issues between online and offline.
Question 5
A company stores petabytes of unstructured data in Amazon S3. The ML engineer needs to
catalog this data, infer schemas, and prepare it for machine learning pipelines. Which service
should the engineer use?
A. Amazon Athena
B. AWS Glue Crawlers and Data Catalog
C. SageMaker Data Wrangler
D. Amazon EMR
-” detailed answer 100 % correct :-”B
Rationale: AWS Glue Crawlers automatically scan data in S3, infer schemas, and populate the
Glue Data Catalog. Athena queries the catalog but doesn't catalog data. Data Wrangler is for
transformation, not cataloging. EMR requires manual schema definition.
Question 6
An ML engineer needs to transform raw JSON logs into a structured format suitable for model
training. The transformation involves complex joins, aggregations, and feature engineering
across multiple data sources. The workload runs daily and must be cost-effective. Which service
should the engineer use?
A. SageMaker Data Wrangler with scheduled data flows
