pandas requirement too strict

[DominikStiller]

Pyleoclim's setup.py pins the pandas version to 2.1.4:

Pyleoclim_util/setup.py

Line 30 in ecaf89c

"pandas==2.1.4",

This requirement is too strict and makes it impossible for my package manager to resolve the conflict. Is there any chance the requirement can be relaxed to >=2.1.4?

[CommonClimate]

Hi @DominikStiller,
this is because of the non-ns datetime stuff, which we were very excited about, and then pandas broke a bunch if things in subsequent releases. This is also an issue for interoperability with xarray so we'd love to move forward, but until the bug is fixed, there isn't much we can do. If you don't care about resampling, you can manually update pandas to the latest release after installing Pyleoclim -- I'd be curious to see what breaks besides resampling.

[CommonClimate]

@DominikStiller did you try it?

[DominikStiller]

I was able to install pyleoclim manually using pip (my package manager, pixi, didn't let me because of the dependency conflict) and then upgraded to pandas==2.2.3 and numpy==2.1.3 again. I didn't have any issues with the wavelet analysis + significance test, multitaper spectra, and linear detrending. However, I'm using float timestamps, no datetime64[ns]/[s].

[CommonClimate]

Hi @DominikStiller, good to know. The code converts timestamps to datetime objects under the hood (using a s interval by default), but as far as we know only resampling is creating problems with those. It's good to have confirmation that these functionalities are fine. If you are satisfied, please close the issue.

[CommonClimate]

This issue surfaces again because the pandas requirement also limits NumPy to <2:

This is getting to the point where resample() is hanging up a lot of key packages, so I suggest we put a warning in the code that this function awaits a resolution of the aforementioned bug, and just move on with our high-grade pandas and NumPy lives.

[CommonClimate]

Dependency has been lifted in the master branch and the forthcoming release 1.2.1. I did have to keep the 2.1.4 pin in the rtd_env.yml file so the documentation would build properly for Series.resample() and GeoSeries.resample(), but otherwise we now live in a modern pandas world -- a world where resampling is unavailable until this bug is fixed.