TL;DR; rewrite it in #Rust! 😛 Just kidding, there are lots of things we can do as #Django / #Python tech people to build leaner. More nuance coming up in the full podcast

This is just from the books3 dataset (included in The Pile). obvs goes well beyond this dataset, it’s just unique because we can name exactly which torrented books were part of training data [3/3]

And more!

- Beginning django CMS
- Django By Example
- Django Design Patterns and Best Practices
- Django: Web Development with Python
- Learning Django Web Development
- Mastering Django: Core
- Pro Django

[2/3]

Another option I’ve seen occasionally is "ask people to run your analysis script on their projects and share results". Yet another type of bias in the data so not suitable for much of anything. Anyway, I think "top 8000 PyPI" is a pretty good pick for this specific analysis! ⭐️ ty for the context

re corporate codebases, I guess the simplest I’d have thought is open source code on GitHub? Select repos on GitHub based on number of stars or activity levels. I have a small dataset of @wagtail.org projects for that reason.

ty! yeah my assumption is packages probs have more scrutiny on avg than apps, so probs if your analysis shows a clear enough problem with top packages, it’d only be worse with other packages, and with apps? No hard data to validate my assumption but seems intuitive enough.

I really liked your talk, hope it’s ok to ask a bonus question! I wanted to ask how/why you selected the code of the "top 8000 PyPI packages" as a dataset to analyze? Why PyPI packages and why top 8000? Feels valid but skewed towards packages, and specifically ones that have more scrutiny than avg

one of the questions Jake got started with "this isn’t a question more of a comment" 💯💥 someone knows what they’re doing

Ticket #36389 gets a shout-out!