Fastai did a remarkable contribution for coding deeplearning alright . But i find most part especially where we create the dataloaders , preprocessing images etc… as intimidating . It would have been a whole lot nicer if there was a proper auto code completion ( not just showing the attributes to filled inside an object ) but also debug our approach in a step by step manner .
I think that with the current debugging tools is quite fine… perphaps it’s just matter of time and patience, and also get to know the docs properly
I tell you this, because I passed trought this in the past … and I can sure you, it’s just matter of time and patience
i understand why you may be feeling intimidated by all the technicalities that you mentioned.
Fastai is designed to give flexibiity of choices while designing your model (Choices like input, transformations,etc).
Given the variety of problems you can solve by the same pieces of code, its really important to first make all inputs and outputs into the format that the code expects. And different datasets and problems provide different types of data.
So, as a practitioner, its really fruitful to spend time understanding these small concepts, in my opinion.
Really its just 3 or 4 steps, and most of the time, all you have to do is follow a set procedure.
Rest, if you ever feel stuck in theory or dont understand why something is done, or even exists in the first place(like, What are Dataloaders in the first place, and why do we need it?), you might find the answer on the forum. Even if you dont, you can always post!
Hope this helps