Cannot express how relieved I am to see the CI finally green 😅
be087d2

Thanks for this example! Can we add this to the category case studies instead of distributions? If the uncertainty was computed within Altair (not yet possible) than it was a good example for the distribution section.

On-the-fly computation of these confidence regions by doing selections would be really great though. Continue dreaming for vega/vega-lite#6043.

Can we remove this line regarding the import annotation from __future__? Probably also requires removing other types from the function, but that is fine for this example.

Can we rename the functions? Eg can we use confidence_region_2d and grouped_confidence_regions instead?

I also tried to reduce the usage of pandas by using dicts directly. I think that can work, but the readability is slightly less than current situation. But doing more with dicts as containers is always good 😊

Thanks for this example!

No problem @mattijn, although I feel my role in this was mostly making the connection between #3715, #514 and ggplot2.stat_ellipse.
It was interesting though brushing up on some scipy/numpy - as I don't usually interact with these packages directly.

Can we add this to the category case studies instead of distributions? If the uncertainty was computed within Altair (not yet possible) than it was a good example for the distribution section.
On-the-fly computation of these confidence regions by doing selections would be really great though. Continue dreaming for vega/vega-lite#6043.

Huh, I guess I'd never considered where the computation occured being a factor in the category.
Yeah I can recategorise if you think that would be a better fit

Can we remove this line regarding the import annotation from __future__? Probably also requires removing other types from the function, but that is fine for this example.

Yeah I think we might not need that line at all actually.
I added it out of habit when I used tuple[...] here, but that should be valid since https://docs.python.org/3.9/whatsnew/3.9.html#type-hinting-generics-in-standard-collections

I'll take a look today and see what works 👍

Can we rename the functions? Eg can we use confidence_region_2d and grouped_confidence_regions instead?

Yeah those seem like much better names!
Bikeshedding a little here, but maybe confidence_region_2d and confidence_region(s?)_grouped?.
I don't feel too strongly about it since this is an example - but I'd usually try to match the prefix for related functionality

I also tried to reduce the usage of pandas by using dicts directly. I think that can work, but the readability is slightly less than current situation. But doing more with dicts as containers is always good 😊

Interesting, I hadn't thought to try this without a dataframe at all 🤔
I'm not a fan of pandas, but since we're already using it (vega_datasets) for iris - I think it is simpler just to keep the data in that container.
We'll also be able to easily adapt this example to penguins (#2231) after (#3631) and in #2213

@mattijn hopefully I've addressed all of your points in (#3747 (commits))

Thanks for your help with this @mattijn