If you haven't already, please read through our contributing guidelines before opening your PR

What is the purpose of this PR?

Fixes several issues plauging CI.

1. Fixes the example dataset not being found across macOS and Windows GitHub Actions Runners.
2. Adds the EZ Segmentation dataset to the the ./github/scripts/get_example_dataset.py download script.
3. Updated outdated GitHub Actions Workflows

How did you implement your changes

Example Dataset

Simplified the CI download script.

conftest.py directly access the GITHUB_WORKSPACE environment variable which is set in CI, thus making the path OS agnostic.

Modified the example_dataset.ExampleDataset class to:

• Explicitly set the cache_dir to the HuggingFace default ~/.cache/huggingface/datasets if None is provided.
• Converted the dataset_paths from DatasetDict to a regular dictionary in order to make the example datasets to be OS agnostic, and it's easier to access the paths.

For example:

'/Users/user/.cache/huggingface/datasets/downloads/extracted/<hash>'

'pathlib.path(self.dataset_cache) / downloads/extracted/<hash>/<feature_name>'

CI

• Added GITHUB_WORKSPACE as an environment variable in CI.
• Updated the macOS runner from macos-latest (where latest is 12) to macos-13
• Updated the following actions:
  • actions/checkout: v3 $\rightarrow$ v4
  • docker/setup-qemu-action: v2 $\rightarrow$ v3
  • docker/setup-buildx-action: v2 $\rightarrow$ v3
  • docker/login-action: v2 $\rightarrow$ v3
  • docker/metadata-action: v4 $\rightarrow$ v5
  • docker/build-push-action: v4 $\rightarrow$ v5
  • pypa/cibuildwheel: v2.13 $\rightarrow$ v2.16
  • actions/upload-artifact: v3 $\rightarrow$ v4
  • actions/setup-python: v3 $\rightarrow$ v5
  • actions/download-artifact: v3 $\rightarrow$ v4
  • pypa/gh-action-pypi-publish: v1.6 $\rightarrow$ v1.8
• actions/upload-artifact@v4 artifacts are immutable, so the artifact upload process has been adjusted for coverage (test) and distributions (build).
  • For coverage, each matrix configuration (runner + python version) uploads a coverage file: coverage-<python version>-<runner>, (e.x. coverage-3.9-windows-latest.lcov, artifact has the same name).
  • Similar for distributions.
• Adjusted Branch Protection rules for main, where macos-13 status checks for test and build are required. Removed macos-latest status checks.

Dependencies

Updated the following:

• alpineer: 0.1.10 $\rightarrow$ 0.1.12
• Cython: 0.29.Z $\rightarrow$ >3.Y.Z (for both buld-system and dependencies)
• pyFlowSOM: 0.1.15 $\rightarrow$ 0.1.16
• scikit-image: <=0.19.3 $\rightarrow$ <0.19.3 (0.19.3 causes issues)

The valid datset configs are gathered from the HuggingFace repo itself now, and .github/scripts/get_example_dataset.py is simplified.

Pixel Clsutering

Adds natsort calls throughout the Pixie pipeline to avoid issues with channel ordering.
Removed the parameter channels in pixel_som_clustering.py::cluster_pixels as it is an unused parameter. Reflected the change in Notebook 2.

Misc

Adjusted the runtime Protocol definition for ClusterClassTemplate to be syntactically correct.
Removed a few cibuildwheel flags that are not needed.

Remaining issues

CI / Dependencies

• macOS 14 Sonoma Runner are available:
  • Runs exclusivly on M1 silicon, therefore we do not need to use macOS-12/13's emulated arm64 build process for cibuildwheel.
  • Requires us to drop Python 3.9 support.
• Python 3.12 Support: We should add Python 3.12 to the suite of supported versions.
  • This requires us to:
    1. Update to later versions of numpy to make use of these runners, however this requires us to drop the spatial-lda requirement as it is not compatible with numpy versions 1.24 and newer.
    2. Update scikit-image also require the more recent numpy versions, therefore we would need to update scikit-image to 0.20 or later. Requires us to re-address Bad filtering in fiber segmentation #1055.
• NEP 29 - Recent Future NumPy Deprecations
  1. By April 05 2024 support for Python 3.9 for numpy 1.23+ is deprecated.
  2. By of June 22 2024 support for numpy is limited to 1.24+. (spatial-lda and scikit-image hinder this one)
• PyPI has 2FA, need to set that up so we can make releases. We should figure out how it works and get it set up, as we might not be able to upload new releases see here.

Dataset

There have been several improvements to the general dataset workflow with the HuggingFace API. We should consider seeing what new features exist to make the maintenance of it easier for us.