Implementing MLOps in the Enterprise

Chapter 2. The Stages of MLOps

MLOps is not about tracking local experiments and is not about placing an ML model behind an API endpoint. Instead, MLOps is about building an automated environment and processes for continuously delivering ML projects to production.

MLOps consists of four major components (and is not confined to model training):

• Data collection and preparation

• Model development and training

• ML service deployment

• Continuous feedback and monitoring

This chapter explores these components in detail.

Getting Started

Begin with the end in mind. The first step in any ML project is to articulate:

• The problem that needs to be solved using ML.

• What you want to predict.

• How to extract business value from the answer. Examples of business value we might require include decreasing fraud, increasing revenue by attracting new customers, cutting operational costs by automating various manual processes, and so on.

Once you define the goal, don’t rush straight into implementation. First, consider the following:

• Which historical and operational data can be gathered and used in both the training and serving pipelines

• How to incorporate the ML model results in a new or existing application in a way that can make an impact

• How to verify and reliably measure that the ML model meets the target and generates valuable business outcomes

Figure 2-1 illustrates the different stages in an ML project. Note the feedback loop where the observations are used to recalibrate ...