File Formats & Parquet — MLA-C01 Practice Question
A representative AWS Machine Learning Engineer Associate (MLA-C01) exam question on File Formats & Parquet. Work through it below, then read why each option is right or wrong.
Short answer
The correct answer is B. Convert the files to Apache Parquet with columnar compression.
Parquet is a columnar format that lets query and training engines read only the required columns (column pruning) and skip row groups using min/max statistics (predicate pushdown), directly solving the "read 8 of 40 columns" problem while also compressing better than row-oriented CSV.
The Question
A retail analytics team stores 6 months of clickstream logs in S3 as gzip-compressed CSV files. Their SageMaker training jobs spend most of their time scanning full files to select only 8 of 40 columns, and storage costs are climbing. They want a format that supports column pruning and predicate pushdown for faster, cheaper reads. Which file format should they migrate to?
Why B is correct
Parquet is a columnar format that lets query and training engines read only the required columns (column pruning) and skip row groups using min/max statistics (predicate pushdown), directly solving the "read 8 of 40 columns" problem while also compressing better than row-oriented CSV.
Why the other options are wrong
Avro is row-oriented and optimized for schema evolution and write-heavy streaming, not for selective column reads — it would not reduce the scan cost.
CSV is still a row-oriented, uncompressed-per-column format; better gzip compression reduces size slightly but does nothing for column pruning or predicate pushdown.
RecordIO-protobuf is used by SageMaker built-in algorithms for efficient streaming during training, not for reducing S3 scan costs on a data lake with many unused columns.
Key idea: File Formats & Parquet
Why B is correct: Parquet is a columnar format that lets query and training engines read only the required columns (column pruning) and skip row groups using min/max statistics (predicate pushdown), directly solving the "read 8 of 40 columns" problem while also compressing better than row-oriented CSV. Why A is wrong: Avro is row-oriented and optimized for schema evolution and write-heavy streaming, not for selective column reads — it would not reduce the scan cost. Why C is wrong: CSV is still a row-oriented, uncompressed-per-column format; better gzip compression reduces size slightly but does nothing for column pruning or predicate pushdown. Why D is wrong: RecordIO-protobuf is used by SageMaker built-in algorithms for efficient streaming during training, not for reducing S3 scan costs on a data lake with many unused columns. On the MLA-C01 exam, questions in the "Data Preparation for Machine Learning" domain test whether you can map a scenario's constraints to the right choice. Read the requirement carefully, eliminate options that violate any single constraint, and pick the one that satisfies all of them with the least operational overhead.
Ready to see how you'd score?
Take the free practice quiz and find out which AWS Certified Machine Learning Engineer - Associate domains you need to focus on. No signup required.
Practice 5 similar questions
Same cert, same or adjacent domain. Use these after reviewing the explanation.