TFX- Production ML Pipelines with TensorFlow

Explore production ML pipelines with TensorFlow Extended (TFX) in this 42-minute conference talk from TF World '19. Discover how Google's open-source ML infrastructure platform addresses deployment and scaling challenges inherent in production ML systems. Learn about TFX components, metadata storage, pipeline orchestration, and directed acyclic graphs. Gain insights into custom components, fairness indicators, and feature space coverage. Presented by Robert Crowe and Charles Chen, this talk offers valuable knowledge for ML practitioners looking to design scalable and maintainable production pipelines.

Introduction
What is TFX
Why Google created TFX
Vision of TFX
TFX Components
Components
Metadata Store
Pipeline Components
Example Gen
Orchestration
Directed Acyclic Graph
CubeFlow vs TensorFlow
Charles Chen
TFX Notebook
Overview
Custom components
Fully custom components
Example
Reality
Fairness Indicators
Feature Space Coverage
Whatif