Production ML systems

Estimated module length: 70 minutes

Learning objectives

• Appreciate the breadth of components in a production ML system.
• Pick the ideal training paradigm (static versus dynamic).
• Pick the ideal inference paradigm (static versus dynamic).
• Test your machine learning deployment.
• Ask the right questions about your production ML system.
• Determine flaws in real-world ML models.
• Monitor the components in a production ML system.

Prerequisites:

This module assumes you are familiar with the concepts covered in the following modules:

• Introduction to Machine Learning
• Linear regression
• Working with numerical data
• Working with categorical data
• Datasets, generalization, and overfitting

So far, this course has focused on building machine learning (ML) models. However, as Figure 1 suggests, real-world production ML systems are large ecosystems and the model is just a single, relatively small part.

Figure 1. A real-world production ML system comprises many components.

 

At the heart of a real-world machine learning production system is the ML model code, but it often represents only 5% or less of the total codebase in the system. That's not a misprint; it's significantly less than you might expect. Notice that an ML production system devotes considerable resources to the input data: collecting it, verifying it, and extracting features from it.