Hands-On Machine Learning with TensorFlow.js

GraphDef

The core component of TensorFlow is definitely the graph structure representing the operation graph literally. GraphDef is a definition to describe the graph structure in the protocol buffer. As the definition shows, it is just a collection of operation nodes:

message GraphDef {  repeated NodeDef node = 1;  VersionDef versions = 4;  FunctionDefLibrary library = 2;}

This is part of the definition of GraphDef in TensorFlow. As you can see, it contains a collection of operation node definitions. That is the most important part of GraphDef because it defines the structure of the operation graph. Understanding the information contained in GraphDef is helpful for debugging and analyzing your model in general. Next, let's see what NodeDef ...