TensorFlow For Dummies

Chapter 11

Using Threads, Devices, and Clusters

IN THIS CHAPTER

Configuring threads in applications

Assigning operations to devices

Creating distributed applications with clusters

I feel the need … the need for speed! If you’ve ever said this about machine learning, then this chapter is for you. In my experience, you can accelerate a TensorFlow application using four methods:

• Generate multiple threads of execution
• Access high-performance devices like graphics processor units (GPUs)
• Execute an application on a cluster of networked devices
• Deploy an application to the cloud

This chapter discusses the first three options and then demonstrates how to execute a TensorFlow application in a cluster. Chapter 13 explains how to run TensorFlow in the cloud.

Executing with Multiple Threads

A thread is a sequence of operations capable of executing independently from other threads. In a TensorFlow application, you can take advantage of threads in two main ways:

• Perform time-consuming operations, such as the loading and storing of data, in separate threads. This approach lets your processing thread continue its work without interruption.
• Run a session with multiple threads. In theory, ...