Inside TensorFlow - TF Input Pipeline

Dive deep into TensorFlow's input pipeline with this 54-minute technical session from the TensorFlow team. Explore tf.data's architecture, including Python and C++ views, support for non-tensor types, and both static and dynamic optimizations. Learn about data sources, transformations, user-defined functions, iterator life cycle operations, and C++ abstractions. Gain insights into performance optimization techniques, supported data types, and the tf.data structure. Discover static optimizations, tf.data options, parallelism implementation, and autotuning features. Enhance your TensorFlow skills with this comprehensive look at the internal workings of tf.data, designed for those with basic familiarity of TensorFlow concepts.

Intro
Outline
Data Sources
Data Transformations
User-defined Functions
Example
Iterator Life Cycle Ops
tf.data C++ Abstractions
Performance
Supported Types
Using tf.data Structure
Static Optimizations
tf.data Options
Parallelism Quiz
Implementing
Autotuning