make it opt-in

src/scanpy/neighbors/__init__.py

@@ -643,8 +643,8 @@ def _handle_transformer(
         use_dense_distances = (
             kwds["metric"] == "euclidean" and self._adata.n_obs < 8192
         ) or not knn
-        shortcut = transformer is None and (
-            use_dense_distances or self._adata.n_obs < 4096
+        shortcut = transformer in {"sklearn", "sklearn-pairwise"} or (
+            transformer is None and (use_dense_distances or self._adata.n_obs < 4096)
         )
 
         # Coerce `method` to 'gauss' or 'umap'
@@ -668,16 +668,19 @@ def _handle_transformer(
             raise ValueError(msg)
 
         # Coerce `transformer` to an instance
-        if shortcut or transformer == "sklearn":
+        if shortcut:
             from sklearn.neighbors import KNeighborsTransformer
 
-            assert transformer in {None, "sklearn"}
+            assert transformer in {None, "sklearn", "sklearn-pairwise"}
             n_neighbors = self._adata.n_obs - 1
             if knn:  # only obey n_neighbors arg if knn set
                 n_neighbors = min(n_neighbors, kwds["n_neighbors"])
 
+            # sklearn-pairwise is opt-in, because it takes more memory
             transformer_cls = (
-                PairwiseDistancesTransformer if shortcut else KNeighborsTransformer
+                PairwiseDistancesTransformer
+                if transformer == "sklearn-pairwise"
+                else KNeighborsTransformer
             )
 
             transformer = transformer_cls(

src/scanpy/neighbors/_types.py

@@ -13,7 +13,7 @@
 
 # These two are used with get_literal_vals elsewhere
 _Method = Literal["umap", "gauss"]
-_KnownTransformer = Literal["pynndescent", "sklearn", "rapids"]
+_KnownTransformer = Literal["pynndescent", "sklearn", "sklearn-pairwise", "rapids"]
 
 # sphinx-autodoc-typehints can’t transitively import types from if TYPE_CHECKING blocks,
 # so these four needs to be importable

tests/test_neighbors.py

@@ -1,13 +1,11 @@
 from __future__ import annotations
 
 import warnings
-from pathlib import Path
 from typing import TYPE_CHECKING
 
 import numpy as np
 import pytest
 from anndata import AnnData
-from scipy import sparse
 from scipy.sparse import csr_matrix, issparse
 from sklearn.neighbors import KNeighborsTransformer
 
@@ -21,9 +19,6 @@
     from pytest_mock import MockerFixture
 
 
-DATA_DIR = Path(__file__).parent / "_data"
-
-
 # the input data
 X = [[1, 0], [3, 0], [5, 6], [0, 4]]
 n_neighbors = 3  # includes data points themselves
@@ -242,41 +237,3 @@ def test_restore_n_neighbors(neigh: Neighbors, conv):
         ad.uns["neighbors"] = dict(connectivities=conv(neigh.connectivities))
     neigh_restored = Neighbors(ad)
     assert neigh_restored.n_neighbors == 1
-
-
-def test_regression_shortcut(monkeypatch: pytest.MonkeyPatch):
-    from scanpy.neighbors._backends import pairwise
-
-    monkeypatch.setattr(pairwise, "_DEBUG", True)
-    adata_ref = sc.read_h5ad(DATA_DIR / "neighbors_shortcut_ref.h5ad")
-
-    adata = AnnData(shape=(100, 5), obsm=adata_ref.obsm)
-    sc.pp.neighbors(adata, use_rep="normalized_X", random_state=0, n_neighbors=20)
-
-    mats: dict[
-        Literal["distances", "connectivities"], tuple[np.ndarray, np.ndarray]
-    ] = {
-        key: tuple(ad.obsp[key].toarray() for ad in [adata, adata_ref])
-        for key in ["distances", "connectivities"]
-    }
-
-    assert_allclose(*mats["distances"], rtol=1e-7, atol=1e-7)
-    assert_allclose(*mats["connectivities"], rtol=1e-7, atol=1e-7)
-
-
-def assert_allclose(
-    a: np.ndarray, b: np.ndarray, *, rtol: float = 1e-7, atol: float = 0
-) -> None:
-    diff = a - b
-    diff[np.isclose(a, b, rtol=rtol, atol=atol)] = 0
-    diff = sparse.coo_matrix(diff)
-    diff.eliminate_zeros()
-
-    msg_nnz = f"{diff.getnnz(0)=}\n{diff.getnnz(1)=}"
-    msg_elems = "\n".join(
-        f"(a-b)[{i:2}, {j:2}] = {d:.8f}" for i, j, d in zip(*diff.coords, diff.data)
-    )
-
-    np.testing.assert_allclose(
-        a, b, rtol=rtol, atol=atol, err_msg=f"{msg_nnz}\n{msg_elems}"
-    )