Keras to Kubernetes

CHAPTER 10A Platform for Machine Learning

In the previous chapter, we talked about the Machine Learning model lifecycle. We saw how model development is a piece of the bigger puzzle that includes problem definition, data collection, cleansing, preparation, hyper‐parameter tuning, and deployment. A good data science or Machine Learning platform should provide tools that can drive automation in these different phases so that the data scientist can drive the end‐to‐end cycle without engagement with software development. This is like the DevOps for Machine Learning. Once the models are released in production, they should be consumed by software applications without special integration.

In this chapter, we will look at some tools and technologies that are being extensively adopted for building ML platforms. We will discuss the common concerns that a data scientist has to deal with while deploying an AI solution. We will see some of the best‐in‐class tools that address each of the concerns. We will also see how these individual products can be tied together to form a bigger data science platform hosted on Kubernetes.

Machine Learning Platform Concerns

We saw in the last chapter how the actual algorithm selection and model development is a key activity in solving an Artificial Intelligence problem. However, it is usually not the most time‐consuming. We have powerful libraries and platforms that simplify this activity and help us build and train models with a few lines of code. Some ...