refactor(datasets): use the shared scverse-misc dataset registry + downloader

.scripts/ci/download_data.py

@@ -19,10 +19,7 @@ def main(args: argparse.Namespace) -> None:
     from anndata import AnnData
 
     import squidpy as sq
-    from squidpy.datasets._downloader import get_downloader
-
-    downloader = get_downloader()
-    registry = downloader.registry
+    from squidpy.datasets._registry import dataset_names
 
     # Visium samples tested in CI
     visium_samples_to_cache = [
@@ -35,23 +32,23 @@ def main(args: argparse.Namespace) -> None:
         logger.info("Cache: %s", settings.datasetdir)
         logger.info(
             "Would download: %d AnnData, %d images, %d SpatialData, %d Visium",
-            len(registry.anndata_datasets),
-            len(registry.image_datasets),
-            len(registry.spatialdata_datasets),
+            len(dataset_names("anndata")),
+            len(dataset_names("image")),
+            len(dataset_names("spatialdata")),
             len(visium_samples_to_cache),
         )
         return
 
     # Download all datasets - the downloader handles caching
-    for name in registry.anndata_datasets:
+    for name in dataset_names("anndata"):
         obj = getattr(sq.datasets, name)()
         assert isinstance(obj, AnnData)
 
-    for name in registry.image_datasets:
+    for name in dataset_names("image"):
         obj = getattr(sq.datasets, name)()
         assert isinstance(obj, sq.im.ImageContainer)
 
-    for name in registry.spatialdata_datasets:
+    for name in dataset_names("spatialdata"):
         getattr(sq.datasets, name)()
 
     for sample in visium_samples_to_cache:

pyproject.toml

@@ -47,7 +47,7 @@ dependencies = [
   "centrosome>=1.2.3",
   "cp-measure>=0.1.19,<0.2",
   "cycler>=0.11",
-  "dask[array,distributed]>=2021.2", # distributed is the only scheduler that forks for GIL-bound featurization
+  "dask[array,distributed]>=2021.2",    # distributed is the only scheduler that forks for GIL-bound featurization
   "dask-image>=0.5",
   "docrep>=0.3.1",
   "fast-array-utils",
@@ -69,10 +69,12 @@ dependencies = [
   "scikit-image>=0.25",
   # due to https://github.com/scikit-image/scikit-image/issues/6850 breaks rescale ufunc
   "scikit-learn>=0.24",
-  "spatialdata>=0.7.2",              # 0.7.2 dropped xarray-schema (pkg_resources break, #1115)
+  # dataset registry + downloader now live in scverse-misc
+  "scverse-misc[datasets]>=0.1.1,<0.2",
+  "spatialdata>=0.7.2",                 # 0.7.2 dropped xarray-schema (pkg_resources break, #1115)
   "spatialdata-plot>=0.4.1",
   "statsmodels>=0.12",
-  "threadpoolctl>=3",                # clamp BLAS/OpenMP per worker in tiled featurization
+  "threadpoolctl>=3",                   # clamp BLAS/OpenMP per worker in tiled featurization
   # https://github.com/scverse/squidpy/issues/526
   "tifffile!=2022.4.22",
   "tqdm>=4.50.2",

src/squidpy/datasets/_datasets.py

@@ -10,10 +10,8 @@
 from pathlib import Path
 from typing import TYPE_CHECKING, Any, Literal
 
-from scanpy import settings
-
-from squidpy.datasets._downloader import get_downloader
-from squidpy.datasets._registry import DatasetType, get_registry
+from squidpy.datasets._downloader import download
+from squidpy.datasets._registry import dataset_names, get_registry
 from squidpy.read._utils import PathLike
 
 if TYPE_CHECKING:
@@ -122,19 +120,15 @@ def visium(
     :class:`anndata.AnnData`
         Spatial AnnData object.
     """
-    # Validate sample_id against known names
-    downloader = get_downloader()
-
-    if sample_id not in downloader.registry:
-        msg = f"Unknown Visium sample: {sample_id}. "
-        msg += f"Available samples: {downloader.registry.visium_datasets}"
-        raise ValueError(msg)
-
-    # Use scanpy.settings.datasetdir/visium if base_dir not specified
-    if base_dir is None:
-        base_dir = Path(settings.datasetdir) / "visium"
+    # guard against the visium_10x names specifically: a valid-but-wrong-type name
+    # (e.g. "imc", an AnnData dataset) would otherwise pass and fail deep in the
+    # anndata loader with a confusing ``unexpected keyword argument 'include_hires_tiff'``.
+    visium_samples = dataset_names("visium_10x")
+    if sample_id not in visium_samples:
+        raise ValueError(f"Unknown Visium sample: {sample_id}. Available samples: {visium_samples}")
 
-    return downloader.download(sample_id, base_dir, include_hires_tiff=include_hires_tiff)
+    # downloads land in <datasetdir>/visium_10x/<sample_id>/
+    return download(sample_id, base_dir, include_hires_tiff=include_hires_tiff)
 
 
 def visium_hne_sdata(folderpath: Path | str | None = None) -> sd.SpatialData:
@@ -152,8 +146,7 @@ def visium_hne_sdata(folderpath: Path | str | None = None) -> sd.SpatialData:
     :class:`spatialdata.SpatialData`
         The downloaded and extracted Visium H&E dataset.
     """
-    downloader = get_downloader()
-    return downloader.download("visium_hne_sdata", folderpath)
+    return download("visium_hne_sdata", folderpath)
 
 
 def cells(folderpath: Path | str | None = None) -> sd.SpatialData:
@@ -171,8 +164,7 @@ def cells(folderpath: Path | str | None = None) -> sd.SpatialData:
     :class:`spatialdata.SpatialData`
         The downloaded and extracted cells dataset.
     """
-    downloader = get_downloader()
-    return downloader.download("cells", folderpath)
+    return download("cells", folderpath)
 
 
 # =============================================================================
@@ -204,9 +196,9 @@ class _DocParts:
     return_type=":class:`squidpy.im.ImageContainer`\n        The image data.",
 )
 
-_DOC_PARTS_BY_TYPE: dict[DatasetType, _DocParts] = {
-    DatasetType.ANNDATA: _ANNDATA_DOC,
-    DatasetType.IMAGE: _IMAGE_DOC,
+_DOC_PARTS_BY_TYPE: dict[str, _DocParts] = {
+    "anndata": _ANNDATA_DOC,
+    "image": _IMAGE_DOC,
 }
 
 
@@ -225,12 +217,12 @@ def _make_loader(dataset_name: str):
         raise ValueError(f"Unsupported type for loader factory: {entry.type}")
 
     def loader(path: PathLike | None = None, **kwargs: Any):
-        return get_downloader().download(dataset_name, path, **kwargs)
+        return download(dataset_name, path, **kwargs)
 
     loader.__doc__ = f"""
-    {entry.doc_header}
+    {entry.metadata.get("doc_header")}
 
-    {doc_parts.shape_prefix} ``{entry.shape}``.
+    {doc_parts.shape_prefix} ``{entry.metadata.get("shape")}``.
 
     Parameters
     ----------