This is pretty much the reverse of what happened in mainstream deep learning. Most people learned with Tensorflow, which is arguably functional, and then decide

This is pretty much the reverse of what happened in mainstream deep learning. Most people learned with Tensorflow, which is arguably functional, and then decided PyTorch (which is like numpy, imperative) was a lot more straightforward to work with, to the point where Google decided to make Tensorflow 2.0 imperative by default in order to stop losing ground