Enforce instance key to be dtype int

src/spatialdata/datasets.py

@@ -240,7 +240,7 @@ def _points_blobs(
  arr = rng.integers(padding, length - padding, size=(n_points, 2)).astype(np.int64)
  # randomly assign some values from v to the points
  points_assignment0 = rng.integers(0, 10, size=arr.shape[0]).astype(np.int64)
- genes = rng.choice(["a", "b"], size=arr.shape[0])
+ genes = rng.choice(["gene_a", "gene_b"], size=arr.shape[0])
  annotation = pd.DataFrame(
  {
  "genes": genes,

src/spatialdata/models/models.py

@@ -19,6 +19,7 @@
 from multiscale_spatial_image.multiscale_spatial_image import MultiscaleSpatialImage
 from multiscale_spatial_image.to_multiscale.to_multiscale import Methods
 from pandas import CategoricalDtype
+from pandas.errors import IntCastingNaNError
 from shapely._geometry import GeometryType
 from shapely.geometry import MultiPolygon, Point, Polygon
 from shapely.geometry.collection import GeometryCollection
@@ -857,6 +858,23 @@ def parse(
  adata.obs[region_key] = pd.Categorical(adata.obs[region_key])
  if instance_key is None:
  raise ValueError("`instance_key` must be provided.")
+ if adata.obs[instance_key].dtype != int:
+ try:
+ warnings.warn(
+ f"Converting `{cls.INSTANCE_KEY}: {instance_key}` to integer dtype.", UserWarning, stacklevel=2
+ )
+ adata.obs[instance_key] = adata.obs[instance_key].astype(int)
+ except IntCastingNaNError as exc:
+ raise ValueError("Values within table.obs[] must be able to be coerced to int dtype.") from exc
+
+ grouped = adata.obs.groupby(region_key)
+ grouped_size = grouped.size()
+ grouped_nunique = grouped.nunique()
+ not_unique = grouped_size[grouped_size != grouped_nunique[instance_key]].index.tolist()
+ if not_unique:
+ raise ValueError(
+ f"Instance key column for region(s) `{', '.join(not_unique)}` does not contain only unique integers"
+ )
 
  attr = {"region": region, "region_key": region_key, "instance_key": instance_key}
  adata.uns[cls.ATTRS_KEY] = attr

tests/conftest.py

@@ -247,7 +247,7 @@ def _get_shapes() -> dict[str, GeoDataFrame]:
  points["radius"] = rng.normal(size=(len(points), 1))
 
  out["poly"] = ShapesModel.parse(poly)
- out["poly"].index = ["a", "b", "c", "d", "e"]
+ out["poly"].index = [0, 1, 2, 3, 4]
  out["multipoly"] = ShapesModel.parse(multipoly)
  out["circles"] = ShapesModel.parse(points)
 

tests/core/operations/test_spatialdata_operations.py

@@ -376,7 +376,7 @@ def test_subset(full_sdata: SpatialData) -> None:
 
  adata = AnnData(
  shape=(10, 0),
- obs={"region": ["circles"] * 5 + ["poly"] * 5, "instance_id": [0, 1, 2, 3, 4, "a", "b", "c", "d", "e"]},
+ obs={"region": ["circles"] * 5 + ["poly"] * 5, "instance_id": [0, 1, 2, 3, 4, 0, 1, 2, 3, 4]},
  )
  del full_sdata.table
  sdata_table = TableModel.parse(adata, region=["circles", "poly"], region_key="region", instance_key="instance_id")

tests/models/test_models.py

@@ -1,6 +1,7 @@
 from __future__ import annotations
 
 import os
+import re
 import tempfile
 from copy import deepcopy
 from functools import partial
@@ -305,7 +306,7 @@ def test_table_model(
  region: str | np.ndarray,
  ) -> None:
  region_key = "reg"
- obs = pd.DataFrame(RNG.integers(0, 100, size=(10, 3)), columns=["A", "B", "C"])
+ obs = pd.DataFrame(RNG.choice(np.arange(0, 100), size=(10, 3), replace=False), columns=["A", "B", "C"])
  obs[region_key] = region
  adata = AnnData(RNG.normal(size=(10, 2)), obs=obs)
  table = model.parse(adata, region=region, region_key=region_key, instance_key="A")
@@ -319,6 +320,31 @@ def test_table_model(
  assert TableModel.REGION_KEY_KEY in table.uns[TableModel.ATTRS_KEY]
  assert table.uns[TableModel.ATTRS_KEY][TableModel.REGION_KEY] == region
 
+ obs["A"] = obs["A"].astype(str)
+ adata = AnnData(RNG.normal(size=(10, 2)), obs=obs)
+ with pytest.warns(UserWarning, match="Converting"):
+ model.parse(adata, region=region, region_key=region_key, instance_key="A")
+
+ obs["A"] = pd.Series(len([chr(ord("a") + i) for i in range(10)]))
+ adata = AnnData(RNG.normal(size=(10, 2)), obs=obs)
+ with pytest.raises(ValueError, match="Values within"):
+ model.parse(adata, region=region, region_key=region_key, instance_key="A")
+
+ @pytest.mark.parametrize("model", [TableModel])
+ @pytest.mark.parametrize("region", [["sample_1"] * 5 + ["sample_2"] * 5])
+ def test_table_instance_key_values_not_unique(self, model: TableModel, region: str | np.ndarray):
+ region_key = "region"
+ obs = pd.DataFrame(RNG.integers(0, 100, size=(10, 3)), columns=["A", "B", "C"])
+ obs[region_key] = region
+ obs["A"] = [1] * 5 + list(range(5))
+ adata = AnnData(RNG.normal(size=(10, 2)), obs=obs)
+ with pytest.raises(ValueError, match=re.escape("Instance key column for region(s) `sample_1`")):
+ model.parse(adata, region=region, region_key=region_key, instance_key="A")
+
+ adata.obs["A"] = [1] * 10
+ with pytest.raises(ValueError, match=re.escape("Instance key column for region(s) `sample_1, sample_2`")):
+ model.parse(adata, region=region, region_key=region_key, instance_key="A")
+
 
 def test_get_schema():
  images = _get_images()