AWS MLA-C01 Practice Test

MLA-C01 practice test: what's covered

• SageMaker
• Data Prep
• Training
• Deployment
• MLOps
• Monitoring

Free MLA-C01 sample questions

1. A team stores 4 TB of raw clickstream logs as JSON files in Amazon S3 and needs to run SQL aggregations to build training features without provisioning or managing any servers. Which AWS service most directly satisfies this requirement?

   • A. Amazon Athena querying the data in place on S3
   • B. An Amazon EMR cluster running Apache Hive
   • C. Amazon Redshift with data loaded through COPY
   • D. AWS Glue DataBrew interactive recipes
   Show answer

   Correct: A. Athena is a serverless, pay-per-query engine that runs standard SQL directly against data in S3 with no infrastructure to manage, which matches the requirement exactly. EMR with Hive requires provisioning and managing a cluster, so it is not serverless. Redshift COPY would require loading the data into a provisioned warehouse first. DataBrew is a visual no-code transformation tool, not a SQL aggregation engine, so it does not fit the SQL requirement.

2. An ML engineer is preparing a tabular dataset in Amazon SageMaker Data Wrangler and notices that the categorical column 'region' has five string values that must be converted into numeric input for an XGBoost model. Which transform should the engineer apply to avoid implying a false ordinal relationship between the regions?

   • A. One-hot encoding of the region column
   • B. Ordinal (label) encoding mapping each region to an integer
   • C. Standard scaling of the region column
   • D. Min-max normalization of the region column
   Show answer

   Correct: A. One-hot encoding creates a separate binary indicator for each region, so the model treats them as unordered and no false ranking is introduced. Ordinal encoding assigns integers like 0–4, which implies that one region is greater than another and creates a spurious order. Standard scaling and min-max normalization are numeric transforms that cannot be applied to raw string categories and would not solve the encoding problem.

3. A company must ingest a continuous stream of IoT sensor readings and land them in Amazon S3 in near real time so that a nightly SageMaker training job can consume them, while keeping operational overhead low. Which two AWS services together best meet this requirement? (Choose TWO)

   • A. Amazon Kinesis Data Streams to capture the sensor events
   • B. Amazon Data Firehose to buffer and deliver records to S3
   • C. Amazon SQS standard queues polled by a custom consumer
   • D. AWS Snowball Edge to batch-transfer the readings
   • E. Amazon EBS volumes attached to the training instance
   Show answer

   Correct: A and B. Kinesis Data Streams ingests high-volume streaming sensor events, and Data Firehose buffers those records and delivers them to S3 with no servers to manage, giving a low-overhead near-real-time pipeline. SQS with a custom consumer adds operational burden and is message-oriented rather than a streaming-to-S3 delivery service. Snowball Edge is for offline bulk transfer, not streaming. EBS is block storage for an instance and cannot serve as a streaming landing zone in S3.

4. During exploratory analysis an ML engineer finds that a numeric 'income' feature contains about 3 percent missing values that appear to be missing at random, and the downstream linear model cannot accept nulls. Which approach is the most appropriate first choice to handle the gaps while preserving as many rows as possible?

   • A. Impute the missing values using the column median
   • B. Drop every row that contains a missing income value
   • C. Replace each missing value with the constant zero
   • D. Remove the entire income feature from the dataset
   Show answer

   Correct: A. Median imputation fills the small fraction of missing values with a robust central estimate, keeps all rows, and resists skew from outliers, making it the appropriate first choice. Dropping rows discards usable data unnecessarily for only 3 percent missingness. Replacing with zero distorts the distribution because zero is not a realistic income. Removing the feature throws away potentially predictive signal and is excessive for a minor amount of missing data.

5. A data science team wants a central repository where engineered features can be stored once, served at low latency for real-time inference, and also retrieved in bulk for batch training, ensuring the same feature definitions are reused across multiple models. Which SageMaker capability should they adopt?

   • A. Amazon SageMaker Feature Store with online and offline stores
   • B. Amazon SageMaker Model Registry versioned model packages
   • C. Amazon S3 buckets organized by feature folder prefixes
   • D. Amazon SageMaker Experiments tracking entities
   Show answer

   Correct: A. Feature Store provides an online store for low-latency real-time serving and an offline store for bulk historical retrieval used in training, with shared feature group definitions that promote reuse and prevent training-serving skew. Model Registry versions trained models, not features. Plain S3 prefixes give no low-latency online serving or governed feature metadata. Experiments tracks runs and metrics and is not a feature serving repository.

Last updated: 2026-10-19

MLA-C01 practice test FAQ

Related practice tests

• AWS AIF-C01 practice test
• AWS DEA-C01 practice test
• AWS SAA-C03 practice test
• Databricks Machine Learning Associate practice test
• AWS MLS-C01 practice test
• NCA-GENL practice test