Announcing Walk with fastai: A Collective Resource of fastai Tips and Tricks

Today is a very big day, as walkwithfastai is now officially released for the public to both read and contribute!

What is Walk with fastai?

As some may know, I ran a “course” of sorts earlier this year called Walk with fastai2. It was aimed at showing various different and non-standard examples of using the fastai library. My long-term goal for the project was to turn it into a book, however after the release of fastbook I suddenly realized that instead, a living breaking community-fed resource was the way to go, with no-holds-barred!

Current Chapters:

Currently there are four chapters up, covering basic introductions to the vision and tabular modules, walking through what an integration with the timm library would look like, as well as walking you through the basic pets example

A Living Resource

On top of it being a central knowledge bank, my other goal for the project is to have it as a standalone library.

What does this mean?

If you’ve written a cool function or integration, we can export that to a module as WWF was written in nbdev. For an example here is the timm integration:

How do I contribute?

There is a contribution guide here in regards to the general format, as well as a basic outline of topics I’ve planned and am currently working on here

I’ll do my best to include topics that everyone wants to do, however if it is a bit too generic or close to the way the course or documentation has it, I may ask for some rewrites in order to make it something truly special :slight_smile: That being said reposts are encouraged. If you write a chapter, please put it on your own personal blog too! My own blog acts as a living resume for employers to look at so I believe that maintaining this status-quo is extremely important

Closing Remarks

If you want to receive email updates with new chapters posted I have a mailchimp feed you can sign up for here

I really do hope you all will utilize this resource and potentially contribute, my heart and soul has gone into making Walk with fastai something I can truly be proud of, and I am ecstatic for what happens from here :slight_smile:

I’d also like to thank all of you who have helped me get to this point, especially @jeremy, @sgugger, @morgan, @LessW2020, @ilovescience, @init_27 and so many others for keeping me motivated and pushing me throughout this journey. Thank you! :slight_smile:


Nice! Change your linky from https, chrome is saying that it’s not secure.


Thanks! I’ve worked around that as best I can, there is a certification it has but I’ll look into it. (Tried testing it w/ other folks before the release and they never got a warning)

Ah, has an ssl cert, does not. That’s why it was complaining, you’d linked to www.

1 Like

The mystery has been solved. TYSM! :raised_hands:t2:

1 Like

Alright, it’s been a month. Have I been slacking off?

Not quite :slight_smile: We have three new articles for folks to explore, one of which was co-authored by @etremblay. This set of articles are tabular focused, with the first discussing how to use TabularPandas in production, the second showing how to use precalculated stats for your tabular procs, and the last discussing AutoEncoders inside of fastai!

TabularPandas in Production
Custom proc Stats
AutoEncoders (co-authored by etremblay)


The entirety of my original Walk with fastai course is now live on here as well and updated to work with the latest libraries + more detailed explanations in most places


Hi muellerzr hope all is well!

As usual great work.
Cheers mrfabulous1 :smiley: :smiley:

1 Like

A new article was just posted, co-authored by @vrodriguezf! :slight_smile:

We added in a show_at function to the interpretation class to take a look at individual items, as well as a __getitem__ to quickly grab the stored inputs, preds, targets, decoded outputs, and losses at particular indices you may be interested in (such as those returned by top_losses)


1 Like

Hi muellerzr@vrodriguezf Hi hope you are both well!

Another great piece of work by two fastai Jedi’s.

Thanks mrfabulous1 :smiley: :smiley:

1 Like

Article time :smiley:

This time I’ve introduced a more verbose way for us to look at the training loop. Simply passing in learn.show_training_loop(verbose=True) will run an alternate version of fastai’s show_training_loop, where a description of what happens during each Callaback that is called during a particular event will be shown (so long as a doc string is provided!)

Article can be found here:

And to use simply do from wwf.basics.training_loop import * :slight_smile:

1 Like