Installing rapids and spatialdata in the same environment
Is your feature request related to a problem? Please describe. I am currently unable to install (with pixi) an environment having both spatialdata and rapids because of dependency issues. It would be very useful to allow this to benefit from rapids' GPU-accelerated functions
Describe the solution you'd like pixi install working
Describe alternatives you've considered Using conda kind of works because it allows conflicting packages to coexist but it isn't ideal
Additional context
pixi.toml
[workspace]
authors = ["j-bac <[email protected]>"]
channels = ["conda-forge", "rapidsai","nvidia"]
name = "tmp1"
platforms = ["linux-64"]
version = "0.1.0"
channel-priority = "disabled"
[tasks]
[dependencies]
python = "3.11.*"
rapids = "25.08"
[pypi-dependencies]
spatialdata = ">=0.5.0, <0.6"
Hi, thanks for reporting. Is this due to dask version conflict?
Yes indeed
Thanks. We are discussing a solution within scverse. But in short, either we or someone fixes this https://github.com/dask/dask/issues/11146, or we need to have some sort of redesign (with some degrees of breaking changes). We are more inclined to go for option 2 and to rely on xarray for creating a thin wrapper around points (so a breaking change, but not a large one).
Is there an update when this might be implemented? Using rapids together with spatialdata would be great for a lot of applications.
This is now fixed! Please give a try to https://github.com/scverse/spatialdata/pull/1006. If you have any feedback it would be welcome.