The first gateway to a machine learning pipeline is "where and how do we get the data?" No matter how sophisticated a model architecture is, if data doesn't come in, nothing can be learned. In the MLA-C01 exam, Domain 1 (Data Preparation for Machine Learning) accounts for approximately 28% of the total, and the starting point for that is precisely ingestion.
Today we examine three pathways through which data enters AWS — S3, the heart of the data lake; Kinesis, for real-time streaming; and periodic batch ingestion — and finally, the choice of data format (Parquet/CSV/JSON) that determines ML training performance.