An ML engineer will use Amazon SageMaker Canvas to train a model with complex-structured data stored in Amazon S3. Which file format will minimize pre-processing time for that data?

Apache Parquet is a columnar storage format optimized for analytical queries and large-scale data processing. SageMaker Canvas can efficiently read and process Parquet files, minimizing pre-processing time because it allows for predicate pushdown and column pruning, meaning only necessary columns and rows are read. CSV files, even when compressed, are row-oriented and require reading entire rows, which is less efficient for complex-structured data. JSON files, whether compressed with gzip or in JSONL format, are also row-oriented and typically require more parsing overhead compared to columnar formats, increasing pre-processing time.