TensorFlow Extended - Machine Learning Pipelines and Model Understanding

Explore the creation of production machine learning pipelines using TensorFlow Extended (TFX) in this Google I/O'19 conference talk. Dive into implementing TFX pipelines capable of processing large datasets for modeling and inference. Learn about data wrangling, feature engineering, detailed model analysis, and versioning. Discover how to implement a TFX pipeline and gain insights into current topics in model understanding. Explore key concepts such as data quality, validation, component configuration, dependency graphs, metadata stores, lineage, and warm starting. Understand the importance of model analysis, reusable components, and open-source orchestrators. Gain valuable knowledge on model understanding techniques, including data validation, slicing, and what-if analysis. Enhance your skills in developing robust machine learning pipelines for production environments.

Introduction
Overview
TFX Introduction
Accessibility
Contract
Systems View
Monitoring Systems
Building Better Software
User Journey
Data Quality
Massage
Validation
Separation
Components
How many know TensorFlow
Model Validator
MLMetadata
Component Configuration
Dependency Graph
DataDependency Graph
Metadata Store
Lineage
Warm Starting
Model Analysis
Reuse Components
Developing Models
Open Source Orchestrators
Putting it all together
Model Understanding
What Went Wrong
Data Validation
Data Validation Example
Questions
Slices
Whatif
CTR
Custom Executor
Assumptions
Baseline Trainer
Conclusion