Dive deep into AutoGraph implementation, usage, and common pitfalls with TensorFlow Software Engineer Dan Moldovan in this 54-minute technical presentation from the Inside TensorFlow series. Explore topics such as function lifecycle, caching, overloading, Python scope preservation, and debugging techniques. Learn about the interplay between Python and Graph code, Eager execution, TensorFlow control flow restrictions, and exception-driven control flow. Gain insights into handling Python collections, shape invariants, type changes, and graph construction errors. Enhance your understanding of AutoGraph's power features and improve your TensorFlow development skills through this comprehensive insider's look at the technology.
Overview
Syllabus
Intro, background, motivation
The life of a function
Caching
Overloading
Preserving Python scopes
Basic example
The solution in the Python case
Power features
Debugging
Python and Graph code
Eager code
TF control flow - restrictions
TF iterators
Exception-driven control flow
Python collections
Shape invariants, type changes
More type changes
Graph construction errors
Taught by
TensorFlow
