Add CaBuAr dataset

[DarthReca]

This PR extends the ChaBuD dataset introduced in #1259.

It is based on data presented in both CaBuAr: California Burned Areas dataset for delineation and ChaBuD challenge. Train and Validation are taken from CaBuAr, while the Test is from ChaBuD.

The files are hosted on HuggingFace.

These are some samples taken from the dataset:

[DarthReca]

@microsoft-github-policy-service agree

[adamjstewart]

Missing tests for the new datamodule. Usually we put these in the trainer tests, but we don't yet have a trainer for change detection (feel free to work on this if you want). For now, you'll have to create a new tests/datamodules test.

[DarthReca]

I have added the datamodule test, but the project coverage is decreasing for some reason, while the patch one is ok.

Ruff mentions some issues with the cabuar dataset file, but no changes were made, and the last run was ok.

For the datasets test, should I add another importorskip to the datamodule test?

[adamjstewart]

• ruff: I just merged Ruff: enable ruff-specific rules #2218 which adds extra checks for certain coding patterns to avoid. You can rebase/merge main to get the tests to fail locally. I can comment directly on the lines with the suggested changes if that helps.
• datasets: Copy the pytest.importorskip to your datamodules tests as well
• coverage: No idea what's going on here, probably a random bug. If you still see the issue after you push a couple more commits I'll investigate.

[DarthReca]

Everything fine. Do you suggest keeping both implementations (ChaBuD and CaBuAr)?

[adamjstewart]

I see no reason not to keep both. Even if the newer dataset is more useful, someone might want to reproduce the results on the previous dataset for comparison against older papers.

[adamjstewart]

Depending on how much code is the same, it may be possible to either subclass ChaBuD:

from .chabud import ChaBuD

class CaBuAr(ChaBuD):
    ...

or create a shared abstract base class. We could either have both datasets in a single file or a different dataset in each file. It's borderline since it's more like a ChaBuD v2, but it has a unique name.

[DarthReca]

Originally the data was proposed in CaBuAr, then we extended the dataset for the challenge. However, the challenge page is not working anymore.

[DarthReca]

I can probably do the reverse if that seems appropriate: I can make ChaBuD an extension of CaBuAr since the latter deals already with more files and subsets.

[adamjstewart]

@DarthReca do you have time to address these change requests? Trying to finalize the 0.6.0 release this week.

[DarthReca]

I will start working today to resolve any issues.

[adamjstewart]

Everything looks fine to me now. Can you add an example figure created by the plot method to the PR description? This will help validate that the plot method works correctly. Feel free to make one dataset a subclass of the other if you want, but I'm also fine keeping things the way they are. Would like to merge this by the end of the day so we can prepare the release.

[DarthReca]

I have added some plots. If this is fine for now, everything should be ready.