1 Accelerating machine learning innovation using MLOps and Kubernetes

 

This chapter covers

  • Bringing DevOps to machine learning development using MLOps
  • Machine learning development lifecycle
  • The anatomy of an MLOps platform
  • The role of Kubernetes in MLOps

Machine learning has left the confines of academia and has entered our everyday lives. In a remarkably short period, practical applications of machine learning have left virtually nothing untouched, growing from identifying hot dogs and predicting the next song to play to produce intricate and innovative outcomes across a wide spectrum of domains.

While machine learning systems may appear magical to users, building such systems has proven to be difficult, even for some of the largest organizations. One Gartner study[1] predicted that as many as 85% of machine learning projects will fail to reach the intended outcome. The high failure rate is driven by the fact that these systems require overcoming challenges that didn’t exist in traditional software systems. Those that underestimated the complexity of operationalizing machine learning soon learned that building a model is one thing, integrating it into real world systems is a whole different ballgame.

1.1 The need for speed

1.2 Developing machine learning systems

1.2.1 Establishing business goals

1.2.2 Data engineering

1.2.3 Model development and training

1.2.4 Evaluation

1.2.5 Inference

1.2.6 Observability

1.2.7 Rinse and repeat

1.3 MLOps platform

1.4 Benefits of building MLOps on Kubernetes

1.5 When to choose Kubernetes as your ML platform

1.6 What does this book teach?

1.7 What does this book not teach?

1.8 Summary