Choropleth and branca colormaps do not show up with numpy 2.0 about folium HOT 10 CLOSED

I haven't tested Numpy 2 myself yet, will do so. I do wonder though, if the stable release is fine, should we worry right now? Maybe something will get fixed before the problem with the nightly release gets into stable?

from folium.

Numpy 2.0 comes with a lot of breaking changes so my bet is that whatever is causing this will not be fixed and is intentional. It may also be a bug but then we need to ensure they're aware of it.

from folium.

I installed numpy==2.0.0rc1 and tried to verify compatibility.

For testing Choropleth I need Pandas, but there's not currently a Pandas version that's compatible with Numpy 2.0.0. It's expected to come out somewhere in the coming weeks. See pandas-dev/pandas#55519.

LinearColorMap seems to work without issue. Branca doesn't actually have a dependency on Numpy. Though it does check if Pandas is installed, so if Pandas is broken it will cause an error.

I'm now thinking we pin Folium to Numpy<2.0.0, and once our dependencies like Pandas are compatible, we make sure Folium also still works. List of status of dependencies: numpy/numpy#26191 How does that sound Martin?

Maybe Geopandas is a special case, since Folium uses it, but Geopandas also uses Folium. I guess when we can run Geopandas without issue on Numpy 2.0.0, even if it's not listed as officially compatible, we can drop the pin on Numpy.

from folium.

Pandas nightly is compatible with numpy 2.0. Let me dig into this to see where it comes from. I don't think it is caused by any of the dependencies like pandas as the only change required in the environment composed of nightly versions is numpy.

from folium.

I should say I didn't install from nightly sources, but from the 2.0.0 release candidate: https://github.com/numpy/numpy/releases/tag/v2.0.0rc1

from folium.

Does Folium run CI with NumPy 2.0.0rc1?

from folium.

@jakirkham no, though we probably should.

from folium.

Thanks Martin! 🙏

Went ahead and filed issue: #1937

Please feel free to add any more context there

from folium.

BTW, running CI with latest numpy may not help much we our tests are flaky and we use numpy in a very basic manner. Hence the blank page that would result in a passing CI :-/

We never created image checks or deep HTML inspection to catch something like this.

With that said, my guess is that np.histogram may have changed. I'll check the other issues/PRs opened.

from folium.

Ah ok. Happy to defer to whatever you and Martin think is best 🙂

Thanks Filipe! 🙏

from folium.