Understanding model development

In this section, we’ll discuss various tools that will help us manage the model development phase of the ML solution life cycle. Let’s start with the most important question – which NN framework should we choose?

Choosing an NN framework

So far in this book, we’ve mostly used PyTorch and TensorFlow. We can refer to them as foundational frameworks as these are the most important components of the entire NN software stack. They serve as a base for other components in the ML NN ecosystem, such as Keras or HF Transformers, which can use either of them as a backend (multi-backend support will come with Keras 3.0). In addition to TF, Google has also released JAX (https://github.com/google/jax), a foundational library that supports GPU-accelerated NumPy operations and Autograd. Other popular libraries such as NumPy, pandas, and scikit-learn (https://scikit-learn.org) go beyond the scope of this book as they are not strictly...