10 Path to production
This chapter covers
- Preliminary work and tasks before productionizing deep learning models
- Productionizing deep learning models with a deep learning system
- Model deployment strategies for experimentation in production
As the concluding chapter of the book, we think it makes sense to return to a high-level view, and connect all the dots from previous chapters. We have now discussed in detail each service in a deep learning system. In this chapter, we will talk about how the services work together to support the deep learning product development cycle we introduced in chapter 1. That cycle, if you remember, brings the efforts of research and data science all the way through productionization to the end products that customers use.
To help jog your memory, figure 10.1, borrowed from chapter 1, shows that product development cycle. Our main focus in this chapter will be on three phases that come toward the end of the process: DL research, prototyping, and productionization. This focus means we’ll ignore the cycles of experimentation, testing, training, and exploration, and look at how to take a final product from the research phase to making it ready to be released to the public.