How should I get started with fast.ai library?


#1

Hi all,

Is it the best way for me to watch Jeremy video and working with the notebook? I tried to look around any written “tutorial” for fast.ai library or documentation but can’t find any. Even some starter scripts would be good too…

I have very little background on ML (have done a few MOOC and tried out a few ML library) and have done programming for quite a long time. So I usually read tutorial and the api/ documentation myself to proceed. So may be that’s not the ideal way to learn fast.ai?

Any recommendations/ suggestions are welcomed.

Thanks,


#2

Under the root of the repo, in directory fastai/courses/ you will find plenty of example notebooks.

If you are interested in deep learning and have a GPU you can use (or willing to set up something in the cloud) I would recommend starting with watching the lectures here If you are more interested in ML, you could go this route.

Being able to read and write code is super helpful but there are quite advanced things happening in the notebooks and the context in the recorded lectures can be of immense value - regardless what you coding / ML background is, I would start with the videos.


#3

Hi radek,

Thx for your reply, appreciated that. Yes, I did finish the ML one but haven’t started 2018 DL one yet.
Yes I think I would probably start the DL video now. But I asked earlier to make sure I didn’t miss out anything obvious (like docu, tutorial etc).


(urmas pitsi) #4

I would suggest first watch all videos with 1.5x speedup: oversaturate your brain with information. Then relax, rehab and pick favorite subject and dig into that. Meaning rewatch related lessons and working through related notebooks, digging into source if necessary.
Happy exploring :slight_smile:)

Let me just add one more thought, that fast.ai library is not a ‘code first’ library. It is more of a ‘concept first’ library. Meaning that it is better to understand concept first and then you understand what the code is doing. Not the other way around. That is how I feel about it.


(Phil Weslow) #5

I think your idea of going through all the videos at 1.5 speed to start is great. However, I had no idea that there was a way to speed up youtube vids. Then I found this video explaining the HTML5 player and how to alter settings… MIND = BLOWN!


(urmas pitsi) #6

for me the speedup functionality addition in youtube some years ago was like a life saver. you could really skim through a lot of more material in the same amount time


(Phil Weslow) #7

Evidently, I’ve been living in a hole for the last couple of years. Glad to finally come out and see the light of day! :grinning:


#8

can you go faster than 2x? would love that!
or like 1.75x for more control to fine tune some videos
seems at the moment youtube gives you 1.25x 1.5x 2x


#9

Yes. I quite agree with what you said fast.ai is “concept first” library. I think it helps you to get up with with deep learning fairly quickly – but if you need to learn more, you need spend a considerable effort in look at the source code or review the course videos… :slight_smile: