This chapter has shown an overall approach to designing and constructing an input data pipeline using TF 2.0 APIs in a simple and suggestive manner. It has provided the building blocks of the different components of the data pipeline and given details of the APIs that are required to build the pipeline. A comparison between TF 1.x APIs and TF 2.0 APIs has been provided.
The overall flow can be summarized in two major passes: raw data management and dataset manipulation. Raw data management deals with raw data; splitting data into train, validation, and test sets; and the creation of TFRecords. Typically, this is a one-time process, which can also include offline data transformation. Dataset manipulation is an online transformation process that creates dataset objects, applies transformations, shuffles the data, and then repeats this and creates batches of the data...