In this chapter, two of the most widely used high-level APIs were presented. tf.estimator and tf.data APIs have maintained almost the same structure they had in TensorFlow 1.x since they were designed with simplicity in mind.
The tf.data API, through tf.data.Dataset, allows you to define a high-efficiency data input pipeline by chaining transformations in an ETL fashion, using the method chaining paradigm. tf.data.Dataset objects are integrated with every part of TensorFlow, from eager execution to AutoGraph, passing through the training methods of Keras models and the Estimator API. The ETL process is made easy and the complexity is hidden.
TensorFlow Datasets is the preferred way of creating a new tf.data.Dataset object, and is the perfect tool to use when a machine learning model has been developed, and it is time to measure the performance on every publicly available...