What's New In TensorFlow 2.x?

Chapter 6. TensorBoard

Neural networks are amazingly complex and can be endlessly fascinating. Experts and novices alike find themselves in bewildered awe when trying to understand the determinants of neural network and optimal performance. How many steps are best in the optimization process, and what other settings are important? How many neurons are best for each layer? How many layers are best? How is information encoded in a neural network? These and many more questions about the neural network are a source of puzzlement and wonder, and their deeper exploration leads not only to exciting revelations about the model at hand, but to new models and ways of thinking about deep learning.

Before we began our journey into writing TensorFlow code, we understood neural networks at a mathematical level, with a distinct emphasis on visual descriptions to organize our thinking. To make progress in our understanding, we will return to the visual representations of deep-learning principles: enter the TensorBoard visualization framework, which contains an incredibly rich set of visualization capabilities. As we will see, it forms the one-stop shop for all visualization needs within TensorFlow and also provides interactive capabilities for a low code exploratory environment.

What Is TensorBoard?

TensorBoard is a visualization framework that is launched separately from TensorFlow. TensorBoard can be launched as a service, and you simply point your browser to the location. The original design ...