Hi all,
We are only a day away from starting the wonderful part 2 and I’m really excited. I did my preparation by revising all the part 1 lessons and reading the amazing notes shared by people in the forum. During this revision, I also took a moment to reflect on my journey for last 2 years with fastai and things that could have helped me a little more, when I was starting. Forums is the best place to learn and share all the knowledge but I remember, as a newbie, I was always intimidated by the idea of looking through all the posts and what’s going on. It was the wrong approach because it is almost impossible for anyone to catch-up with everything going on and things grow exponentially in the forum. This lead me to skipping things in the forums and just trying to search my way out of the topics. In a way, I was only reading what I wanted to and nothing else, typical exploitation vs exploration, due to the sheer size of the information.
To address this issue partially (which might not be an issue for many out there), today I read through most of the topic titles from all the courses since 2017 and derived a structure from it. This structure can act as a meta-data or FLAG associated with the post and improve the search results while helping people to explore at the same time. The idea is not to reduce the number of posts but to categorize them during construction. This would require a marginal effort from the post creator, the idea is to simply write the title as
[CATEGORY: info] Whatever title I want to give to this post
For exmaple:
[Conceptual] Why use loss vs lr and not accuracy vs lr in lr_find?
[DataSet] How to label ImageNet dataset?
[Notebooks: Lesson 2] name ‘classes’ is not defined
An almost exhaustive list of category along with the expected content is posted below, always read [Anything] as “Anything questions” in your head
- [Setup]: Questions related to IDE, Unix, python, git clones, cuda or anything local with setup
- [Platform]: Questions related to AWS, Crestle, Slamandar, Paperspace or anything new, add info
- [Documentation]: Questions related fastai, pytorch or any other documentation
- [PyTorch]: Specefic Pytorch issues or errors that cannot be associated with fastai library code
- [Swift/TF]: Questions around Swift and TF (For last few lessons)
- [FastaiLib]: Questions around anything that is written(has source code) inside the library
- [Notebooks]: Questions related to the notebooks, add info around lesson# or personal
- [Conceptual]: Questions around DL, CV, NLP, RL etc. which are fundamental in nature
- [Competitions]: Kaggle, MURA or anything else that you want to discuss, try to add info
- [Implementation]: Questions related to implementation of a concept/paper, add paper info like
[Implementation: arXiv:1708.07120v3] - [Reading]: Suggestions, Shares or Questions around any reading content, adapt the info
- [Datasets]: Questions around fastai or public datasets with relevant info
- [Notes]: Q/A for notes and sharing notes
- [StudyGroup]: Discussion and questions for a study group, add timezone in info
- [Others]: Try your best to minimze the use of this tag, I’m using it as a zero point to start this
As this is just an idea, I request your feedback if this sounds logical and ready to be used / possible but need changes / a waste of time. I am only tagging @jeremy for his thoughts first but your contribution is equally valuable. Let’s try to bring this to life along with lesson one 
Thanks for reading
It’s an interesting idea. I don’t think we could use topic flags like that, because I think most people wouldn’t add them. But we could create all those as sub-forums.