From Experimentation to Products - The Production Machine Learning Journey

Explore the journey from machine learning experimentation to production in this comprehensive conference talk. Dive into the world of MLOps and learn about the challenges and solutions in deploying advanced ML technology. Discover the importance of continuous integration, deployment, and testing in ML applications. Gain insights into TensorFlow Extended (TFX) and its production components, understanding how it scales for large-scale ML applications. Compare TFX with Kubeflow pipelines and explore distributed pipeline processing using Apache Beam. Examine various TFX standard components, including ExampleGen, StatisticsGen, SchemaGen, and more. Learn about TFX pipeline nodes, custom components, and high-level architecture. Perfect for data scientists, ML engineers, and professionals looking to enhance their knowledge of production-ready ML systems and best practices in MLOps.

Intro
Production ML
We need MLOps
Continuous integration, deployment and testing
MLOps level 0: Manual Process
Experiment
Tales from the trenches
TensorFlow Extended TFX
TFX production components
What is a TFX component?
TFX orchestration
Difference between TFX & Kubeflow pipelines
Distributed pipeline processing: Apache Beam
TFX standard components
Components: ExampleGen, StatisticsGen & SchemaGen
Components: ExampleValidator, Transform & Trainer
Components: Tuner, Evaluator & InfraValidator
Components: Pusher & BulkInferrer
TFX pipeline nodes
TRFX custom components
Very high level architecture
Outro