SOURCE CODE: Mid-Level API

Part 1 (2020) Study Groups
81

Thanks and sorry about the mistake. I am replying using a mobile and emojis look much smaller.

82

There’s also a chapter in the book discussing it specifically. It starts with text and goes to Siamese:

83

I don’t think anyone has mentioned this here yet (sorry if it was already discussed), but reading the fastai paper is an excellent first step to understanding the library.

Jumping on the source code can be extremely confusing, the library is multi-layered by design and that means you will need to jump a lot between functions, notebooks and even repos!! (The core functionality )

The paper goes over the most important abstractions and can give you a really solid base to start understanding the whole library =)

3 Likes
84

Oh? I see tutorials on how to deploy a Streamlit app though. Quick search shows this: https://towardsdatascience.com/quickly-build-and-deploy-an-application-with-streamlit-988ca08c7e83

85

FYI @arora_aman @init_27 I’ve made the top post a wiki now (as per our thoughts) :slight_smile:

2 Likes
86

May I also suggest the 1-hr video on the “design of fastai v2” — a conference between Jeremy and the swift team:

fastai v2 overview - at a S4TF meeting, November 2019

3 Likes
87

By the way are we going to write those blogposts on our separate blog pages (mine is still not up and running) or are we going to built a new blog page to gather all of our posts? (I think the later would be better)

88

Great!
I discussed about exploring the mid-level and top-level api with @harish3110, I’m not surprised that so many of us are on the same page, as @akashpalrecha said, maybe we should work toward making the blog for the classes and functions involved so as to develop fine insights and later on maybe we can combine our notes and come up with a great documentation, looking forward to working with you all :smiley:

1 Like
89

I think we could initially just have github repos with our jupyter notebooks that could later be converted to a blog. Once everyone is done, we can then meetup, and possibly pick out the best parts from the repos and co-author something (I don’t know how this works on sites like medium, if at all).
Nonetheless, I have my blog setup using fastpages over at: akashpalrecha.me
It’s still a bit rough though.

90

Maybe we should also put an emphasis on the writing and pedagogy aspect of the exercise.
As Jeremy and Rachel often say: write to yourself when you started fastai.
To write pretty good blog posts that will have good relevance even when fastai libreary will move on.

2 Likes
91

I agree with this idea, perhaps someone can make a new FastPages blog that we use that acts as the main “hub”. Also, while everyone is doing this once you make your articles, please update the first post with where your blogs are located (please only do this after you have made your initial blog post on lesson 1) this way we can all read them :slight_smile:

92

I have created my blog using fastpages and its located at https://ganesh3.github.io/

The blogs are way down the page? Does anyone have better template that can be used?

Regards
Ganesh Bhat

93

@ganesh.bhat You can edit the index.md file and remove it’s content if you simply want to push up your blog posts.

3 Likes
94

I’ll go ahead and setup a fastpages blog where we can possibly keep spamming our posts. It’ll then live on a domain that looks something like : akashpalrecha.me/fastai-explained/ (because all my github pages live on my custom domain)

Do people agree with this? Or perhaps we should name it something else?

1 Like
95

I think it’s a good idea. Especially once we get into the more narrowed-down ideas/topics to centralize them there. I still believe that the original authors should post onto their blogs (as it can act like a living resume for many folks!) and then cross post into our main blog. Perhaps something like “fastai2-explained”?

1 Like
96

Yes. That was exactly what I was concerned about. I wouldn’t want to implicitly take credit for everyone else’s contributions. What @muellerz suggests seems fine: post it on your own blog / medium, and add a link to the main blog with everything else.

2 Likes
97

Also, @muellerzr I think jeremy suggested that the library will be renamed to fastai in the future, so I’d be wary of naming the blog to fastai2-explained.

1 Like
98

Great point, I forgot about that. fastai-explained works fine then :slight_smile:

1 Like
99

We should have all the code to read different types of datasets added as examples in the docs. For example we would have examples for single lablel case(like pets), the multilable case, regression, multiple regression etc. so when someone new wants to use the datablocks it’s all in one place. From each example we can have links to the rest of the notebook( this will have the architecture and training etc). I feel this will help navigate and help keep everything consolidated

100

I’ll just add (though I’m sure you’re aware), we have the holy-grail notebook that is nb_50, perhaps we can take some inspiration from it!

1 Like