I am reading the book and of course, it is pretty much detailed. It is an excellent book, one of the best books when it comes to programming and APIs. Even though I can understand everything, I can’t seem to remember much, I think it does a lot of hand-holding, and explanation are pretty long and longer.
I am a computer programmer, programming C, C++, and Linux for 5 years now and if I want to learn any new language or software, I learn best when rather than going to a tutorial I go to the reference manuals. The same thing happened with the fastai book. I like the book but I can remember and work better directly from the docs:
I want to switch to the docs because I find a lesser explanation and more code gives me more focus and engagement with the material. If I don’t understand something, I search and figure things out myself (SGD e.g). I have good experience in computer programming but I am just a beginner in ML, DL, and AI. I want to know if I will miss anything by replacing the book with the docs?
The book does not only present the code but also how to approach the different problems, image processing, different NLP tasks etc. It explains very well why we do things, while with only the docs you would know the different functions but would not necessarily know how to best apply them.
One good approach could be to read the notebooks once and try to redo them entirely yourself later just looking at it if you are blocked or try what you learned on other datasets.
If you can reuse what you learned then you really master it.
You’re not alone alone very few people can remember coding from just reading.
The docs are there for you to use, if they help you understand and remember better then that’s what you should use.
There is no need to replace the book with the docs use them both. In fact with fastai you have a book, videos, doc’s/notebooks and tutorials/notebooks.
I use all four, and combine them to help aid my memory.
I think the book is essential for example chapter 3 on ethics. I would be happy if no one was allowed to create/deploy a model until they had shown they had thought about the ethics of their model.
Chapter 2 contains a section on how to avoid disaster when deploying models in production. So there are concepts in the book which are not covered in the docs which is why I think you should use all the tools Jeremy has made available.
Also according to one of the greatest teachers of modern times, if you want to know how much you really know about something, you should take a blank sheet of paper, a pen or pencil and without any notes, make notes on the paper from memory that you can use to teach someone else what it is you have learned.
I find that the above method is the only way for me, that I can confirm what I know, and I have to use all four resources that Jeremy has provided to do it, as well as building my own apps.