10

Machine Learning
Operations (MLOps)

So far in this book, we have focused on the theory of neural networks (NNs), various NN architectures, and the tasks we can solve with them. This chapter is a little different because we’ll focus on some of the practical aspects of NN development. We’ll delve into this topic because the development and production deployment of ML models (and NNs in particular) have some unique challenges. We can split this process into three steps:

1. Training dataset creation: Data collection, cleanup, storage, transformations, and feature engineering.
2. Model development: Experiment with different models and training algorithms and evaluate them.
3. Deployment: Deploy trained models in the production environment and monitor their performance in computational and accuracy terms.

This multi-step complex pipeline presupposes some of the challenges when solving ML tasks:

• Diverse software toolkit: Each step has multiple competing...