Chapter 3: Hierarchy of TensorFlow

From lower to higher, there are five levels in TensorFlow hierarchy.

The bottom one is hardware level. TensorFlow supports adding CPU, GPU or TPU to the resource pool of computing.

The third level contains operators in written in Python, which provides low-level API instructions that packaging C++ kernels, including tensor operation, graph, automatic differentiate, etc. For example: , tf.constant, tf.function, tf.GradientTape, … If we compare a model to a house, then these third level APIs are the bricks.

The fourth level contains model components implemented in Python. They provide packaging to the low-level API functions, including model layers, loss functions, optimizers, data pipelines, feature columns, etc. For example: tf.keras.layers, tf.keras.losses, tf.keras.metrics, , tf.data.DataSet, tf.feature_column… If we compare a model to a house, then these fourth level APIs are the walls.

Please leave comments in the WeChat official account “Python与算法之美” (Elegance of Python and Algorithms) if you want to communicate with the author about the content. The author will try best to reply given the limited time available.