Backport PR #1321: (fix): empty boolean mask on backed sparse matrix (#…

…1336)

Co-authored-by: Ilan Gold <[email protected]>

anndata/_core/sparse_dataset.py

@@ -180,6 +180,8 @@ def _get_sliceXslice(self, row: slice, col: slice) -> ss.csr_matrix:
 
     def _get_arrayXslice(self, row: Sequence[int], col: slice) -> ss.csr_matrix:
         idxs = np.asarray(row)
+        if len(idxs) == 0:
+            return ss.csr_matrix((0, self.shape[1]))
         if idxs.dtype == bool:
             idxs = np.where(idxs)
         return ss.csr_matrix(
@@ -214,6 +216,8 @@ def _get_sliceXslice(self, row: slice, col: slice) -> ss.csc_matrix:
 
     def _get_sliceXarray(self, row: slice, col: Sequence[int]) -> ss.csc_matrix:
         idxs = np.asarray(col)
+        if len(idxs) == 0:
+            return ss.csc_matrix((self.shape[0], 0))
         if idxs.dtype == bool:
             idxs = np.where(idxs)
         return ss.csc_matrix(
@@ -290,10 +294,12 @@ def mean_slice_length(slices):
         return floor((slices[-1].stop - slices[0].start) / len(slices))
 
     # heuristic for whether slicing should be optimized
-    if mean_slice_length(slices) <= 7:
-        return get_compressed_vectors(mtx, np.where(mask)[0])
-    else:
-        return get_compressed_vectors_for_slices(mtx, slices)
+    if len(slices) > 0:
+        if mean_slice_length(slices) <= 7:
+            return get_compressed_vectors(mtx, np.where(mask)[0])
+        else:
+            return get_compressed_vectors_for_slices(mtx, slices)
+    return [], [], [0]
 
 
 def get_format(data: ss.spmatrix) -> str:

anndata/tests/test_backed_sparse.py

@@ -27,6 +27,10 @@ def diskfmt(request):
     return request.param
 
 
+M = 50
+N = 50
+
+
 @pytest.fixture(scope="function")
 def ondisk_equivalent_adata(
     tmp_path: Path, diskfmt: Literal["h5ad", "zarr"]
@@ -37,7 +41,7 @@ def ondisk_equivalent_adata(
 
     write = lambda x, pth, **kwargs: getattr(x, f"write_{diskfmt}")(pth, **kwargs)
 
-    csr_mem = ad.AnnData(X=sparse.random(50, 50, format="csr", density=0.1))
+    csr_mem = ad.AnnData(X=sparse.random(M, N, format="csr", density=0.1))
     csc_mem = ad.AnnData(X=csr_mem.X.tocsc())
     dense_mem = ad.AnnData(X=csr_mem.X.toarray())
 
@@ -77,6 +81,25 @@ def callback(func, elem_name, elem, iospec):
     return csr_mem, csr_disk, csc_disk, dense_disk
 
 
+@pytest.mark.parametrize(
+    "empty_mask", [[], np.zeros(M, dtype=bool)], ids=["empty_list", "empty_bool_mask"]
+)
+def test_empty_backed_indexing(
+    ondisk_equivalent_adata: tuple[AnnData, AnnData, AnnData, AnnData],
+    empty_mask,
+):
+    csr_mem, csr_disk, csc_disk, _ = ondisk_equivalent_adata
+
+    assert_equal(csr_mem.X[empty_mask], csr_disk.X[empty_mask])
+    assert_equal(csr_mem.X[:, empty_mask], csc_disk.X[:, empty_mask])
+
+    # The following do not work because of https://github.com/scipy/scipy/issues/19919
+    # Our implementation returns a (0,0) sized matrix but scipy does (1,0).
+
+    # assert_equal(csr_mem.X[empty_mask, empty_mask], csr_disk.X[empty_mask, empty_mask])
+    # assert_equal(csr_mem.X[empty_mask, empty_mask], csc_disk.X[empty_mask, empty_mask])
+
+
 def test_backed_indexing(
     ondisk_equivalent_adata: tuple[AnnData, AnnData, AnnData, AnnData],
     subset_func,