Implementing MLOps in the Enterprise

Chapter 7. Building a Production Grade MLOps Project from A to Z

This chapter provides an example of an ML project with all its components and the MLOps attributes required for production deployment. It follows the practices presented in Chapter 3. The example applications predict and prevent credit transaction fraud by calculating user and transaction features and feeding them into a classifier model, which will determine if the transaction is a legitimate transaction.

All the code examples presented in this chapter are stored in Git.

The project implementation consists of the following steps:

1. Exploring and analyzing the data (EDA)

2. Building the data ingestion and preparation pipeline

3. Building the model training and validation pipeline

4. Developing the application serving pipeline (intercept requests, process data, inference, and so on)

5. Monitoring the data and model (drift and more)

6. Addressing continuous operations and CI/CD

The data preparation step will be implemented in two ways: using standard Python packages and using a feature store.

Fraud prevention is a challenge as it requires processing raw transactions and events in real time and being able to respond quickly and block transactions before they occur. Consider a case where you would like to evaluate the average transaction amount. When training the model, taking a DataFrame and calculating the average is common. However, in real-time/online scenarios, the average calculation is accumulative (incremental). ...