The fast.ai library is a set of wrapper functions over the PyTorch code library. See http://www.fast.ai.

The idea is that PyTorch operates at a low level of abstraction and is quite difficult to learn and use. The fast.ai library provides higher level functions that are intended to make coding up deep neural systems easier. An analogy is the way the Keras library is a wrapper over TensorFlow.

I like Keras a lot so I was optimistic when I started exploring the fast.ai library. However, I didn’t like fast.ai at all for at least four reasons.

First, the code quality of the library is poor. Second, the documentation is very poor. Third, there was relatively little gained by using fast.ai in the sense that I could just write my own wrapper functions. Fourth, there was a annoying feel that fast.ai is more interested in selling their learning courses and pushing their political agendas of female victimization and diversity indoctrination than they are in creating good technology.

So, I’m aware that all four of my reasons for not liking fast.ai are subjective but that doesn’t make them irrelevant. I won’t be using fast.ai anytime soon, if ever. For now, PyTorch and Keras are my two deep neural libraries of choice.

Three paintings by Jack Gaughan (1930 – 1985). He did many covers for science fiction paperback books and magazines in the 1960s. He had a very distinctive style intended to catch people’s attention and sell, as opposed to creating art without purpose.

Update: The image on the right is attributed to Gaughan by several resources on the Internet, but I suspect that the artist may actually be John Schoenherr (1935 – 2010).