Applied Natural Language Processing in the Enterprise

Chapter 11. Productionization

The difficulty of the move from prototyping to production is where many companies fail and is one of the main reasons many companies derive such a low return on investment on machine learning initiatives they launch. In the previous chapter, we discussed how to productionize machine learning as a web app. However, the primary way for companies to productionize machine learning and truly unlock the value of these models in a production setting is not via a simple web app; it is via APIs and automated pipelines, both of which we will cover in this chapter. We will also discuss the various roles that are involved in deploying, maintaining, and monitoring machine learning models in production, and explore Databricks, one of the current market-leading platforms to perform data science and machine learning work in the enterprise.

Data Scientists, Engineers, and Analysts

Before we dive into how to productionize machine learning models, let’s review the different individuals who will be involved during the entire machine learning development and deployment cycle. Understanding the roles of these individuals and their preferences for programming language and programming environment is important because we want to reduce the friction in moving from prototyping models to deploying them in production; in other words, we need to consider ease of collaboration to ensure success in running machine learning in production.

Prototyping, Deployment, and Maintenance ...