just a general non-technical question :P

[blackmirag3]

hello guys im a rando kid in uni who just happened to stumble upon this

pretty impressed with the work and congrats on winning

just curious about how you guys learn to do ML, ie whats the motivation, any recommended resources online
noob here with minimal experience ML-wise so any direction regarding what libraries are useful to learn, which algos/patterns to read up on would be nice

thanks and have a nice week ahead!

[aliencaocao]

Hi, thanks for your kind words. We learn ML out of interest and we are from the same CCA in high school which focuses on computing and ML. As for resources, we usually do peer teaching, but coursera and github are great self learning resources. When we did this competition, ChatGPT did not exist yet, but now it does, and it is very powerful to beginners too.
I would say that participating in competitions like this often teaches us the most, as many things taught in theory or on videos are not directly application in production. It is only when you start facing errors and debug your code then you will learn the fastest.
As for libraries, I recommend starting with pytorch. Personally we started with TensorFlow but that was back when PyTorch was harder to use and learn. It has gotten a lot better and a lot more popular. TF remains a good option because it simplifies a lot of the underlying concepts like custom training loops, which can be good for production, but making transition or learning from TF to Torch will be much harder than vice versa.
You can start with simpler tasks like regression, then image classification, then onto language based tasks. This is a repo where I document my learning during my CCA time, and also contains some teaching materials for training my CCA juniors: https://github.com/aliencaocao/IRS-ML
And this is the repo for 2023 version of this competition: https://github.com/aliencaocao/TIL-2023. This TIL-2022 repo is for 2022 edition. 2023 is harder and more fun, I also included much more detailed write up.

[blackmirag3]

Thanks for the time taken to type this man much appreciated!