Companion for FastAI course is Fastbook
Finish the damn course !! I have been trying to finish any version of the course since 2017, passively I did, but that is of no use as I have often found myself clueless. So this time I am going to follow how it is supposed to be done
Create a project: I plan to rewrite Jupyter notebooks for the 2018 fastai course, in recent fastai version 2
Be Tenacious !! Acknowledge the bumps and keep going despite them.
Explore Meta-Learning, I was able to finish this book and wrote a summary about it here.
Start doing something from week 1. Write Code, Do Experiments, Train Models
If you are not well versed in coding checkout CS50
Check out this structured course for understanding the tools used by developers
How to do a fastAI Lesson: Watch Lecture -> Run notebook and experiment -> Reproduce results in a fresh notebook -> Repeat the procedure with a different dataset(while going through this process we might have to revisit lecture again and again, but that's ok).
We can start by using notebook servers but along the way start using full linux servers as mentioned on fastai course page
Explore the source code of functions used in colab notebook.
You'll never finish it, if you didn't get started in the first place
To become a better developer, read code & write code
Validation set should have the same data as will be expected in the real-life application cycle.
Always start with simplest possible baseline
Successful machine learning projects are made by building the simplest machine learning model that works end to end and then gradually improving it incrementally.
Learn to use Tmux or even check out this video from 2017 fastAI course
Checkout Fastsetup