Understanding model-as-a-service
In Chapters 13 and 14, we studied how to develop physics-based or data-driven analytics (machine learning (ML) or deep learning (DL) and, in general, artificial intelligence or AI). In this final chapter, we will see some analytics deployment strategies.
Once developed, analytics reduced to its most extreme essence is nothing more than a function that converts the input (X) into an output (y). Whatever the input – an image, text, or a time series – 90% of analytics can be represented by a function f such that the following applies:
y = f(X)
If you use TensorFlow, Keras, or scikit-learn, this function is implemented with the predict function; with PyTorch, it is the __call__ function.
In Chapter 8, we studied the concepts of software-as-a-service (SaaS) and microservices. The SaaS model has become increasingly prevalent thanks to its remarkable flexibility, scalability, and cost-effectiveness, making it a popular choice for businesses...
