Environment creation fails using provided environment.yml

[mebauer]

Summary

I attempted to install and run the project following the instructions in the repository, but the provided conda environment file (examples/env/environment.yml) fails to build on macOS due to several pinned packages that cannot be resolved. It is also unclear what the recommended setup process is, or which Python versions / platforms are supported.

I also tried to install using pip install git+https://github.com/abpoll/unsafe in the test notebook, but had to install requests and rioxarray, and upon rioxarray, received a numpy issue.

Steps to Reproduce

run conda env create -f examples/env/environment.yml

Behavior

ResolvePackageNotFound:
  - rasterio=1.3.9
  - pandas=2.2.0
  - matplotlib=3.8.2
  - ipykernel=6.29.0

Expected Behavior

The provided environment should resolve successfully across supported platforms.

Alternatively, documentation should clarify:

• supported Python versions,
• supported OS/architectures,
• recommended channels, and whether strict version pins are truly required.

Additional Context

I’m happy to test updated environment files or help improve the documentation once I can get the environment running. It’s entirely possible I’ve missed something, and apologies if so. This package looks fantastic, and I’m really excited to use it. Thank you for all the hard work that’s gone into building it!

[abpoll]

Hi @mebauer, thank you for testing out the package and sorry for the environment issues! It is a mistake that the environment.yml and examples README do not indicate python versions. The pyproject.toml for the package does indicate python >=3.9.

Could you let me know what version of conda and python you attempted this with and I will try to reproduce and address the issue?

[mebauer]

Environment details:

• Conda version: 4.9.2
• macOS (Intel)
• Python ≥ 3.9

The only change I made to the provided environment file was adding an explicit Python version. Everything else remains the same.

Given my operating system, one possibility is that the exact package versions specified in the environment file are no longer available (or were never available) on conda-forge for macOS/Intel. Since the error shows a ResolvePackageNotFound, this seems like the first place to investigate. Some of the pinned dependencies, particularly historically tricky ones like geopandas, might be too restrictive or refer to builds that have since been removed.

These are just the potential issues I am thinking about.

[abpoll]

We did not test the examples on a Intel chip so that's a possibility, and it would be hard for me to replicate that and find the fix. I agree with your suggestions about updating documentation. For starters, strict version pins are not necessary and you might be able to get this working by changing "=" to ">=" We've been using newer versions in recent projects. If you continue to have issues, I will update the documentation about supported OS/architectures. It's on my list to update the documentation to indicate Python version as well. Thank you!

[mebauer]

I was finally able to run the partial_data_example.ipynb notebook successfully, plot included. Phew. I retrieved the package from GitHub, and here’s what I ended up doing to get a working environment:

• Created a fresh conda environment with python=3.9
• Installed most packages from conda-forge, but several core scientific and geospatial libraries ended up coming from pip
• Installed geopandas right after creating the environment, then installed unsafe
• Ran into a few dependency issues, especially a numpy issue, but was able to resolve them with pip uninstalling and reinstalling packages

I can’t say this is the cleanest or most elegant solution, but it does work. It’s still unclear why conda-forge initially reported some package versions as unavailable (I know they’re hosted there), so something odd may be happening with channel resolution. Strange!

Hopefully this helps you and other users get unsafe installed and running smoothly.