The clause treats acquisition as the first pipeline stage. Training and evaluation both depend on the data, so it must suit the business purpose. Training data is often a subset of a larger population and should represent that population. Supervised tasks need labels or target variables, and the system also needs validation and test data, which can be acquired with the training data or separately. The type of data follows the problem: transactional data for forecasting, images for picture classification, and so on. Sources are identified and the data is acquired, from a store or a stream, and common sources include transactions, sensors, queries, images, documents, sound and social media.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.