How and when to move from fast-ai to pytorch?


Have been following the fast-ai lectures (lessons 1 & 2 till now) and really want to get into writing my own code and trying out different problems / datasets on my own. Eventually, I see myself using pytorch and not fast-ai (as I believe it was created specifically keeping the course in mind).

So what is the suggested method ? Should i start looking at how he has coded these functions using pytorch ? or does he himself take us through this transition and for now I should just stick to playing around with the fast-ai version ?


If you follow the lectures through to the end, Jeremy eventually peels back the layers and gets to PyTorch code after the 5th lecture or so.

I won’t suggest deep diving into the fast ai library. The devs are kickass coders and getting a hang of the library can be very intimidating, plus it’s a topic for advanced discussions and will be covered in the Part 2 of the course.

I’d say play around with Fast AI.

But if you want to learn about PyTorch, We’ll be discussing PyTorch basics today in slack. (Basics + Linear and Logistic Regression) from 8-10PM IST.

I’ll share the resources here too soon.