TensorFlow For Dummies

Chapter 4

Executing Graphs in Sessions

IN THIS CHAPTER

Creating graphs and accessing their data

Serializing data from a graph into a GraphDef

Creating and launching sessions

Printing messages to the log

Visualizing summary data with TensorBoard

The preceding chapter introduced a plethora of functions that create, transform, and process tensors. Most of these functions return a tensor, and this may lead you to believe that the function performs its operation as soon as it’s called. This is how Python functions usually work, but this is not how TensorFlow functions work.

When an application executes a TensorFlow function that creates, transforms, or processes a tensor, the function doesn’t execute its operation. Instead, it stores its operation in a data structure called a graph. A graph can hold many operations, and they're not executed until the application executes the graph in a session. When a session executes a graph, it performs the graph's operations in order.

The benefit of ...