fix: support umap-learn 0.6 (#4480)

Author: flying-sheep

docs/release-notes/4480.fix.md

@@ -0,0 +1 @@
+Fix {func}`~scanpy.pp.neighbors` and {func}`~scanpy.tl.ingest` with `umap-learn` 0.6 {smaller}`P Angerer`

src/scanpy/neighbors/_connectivity.py

@@ -129,8 +129,9 @@ def umap(
         n_neighbors,
         None,
         None,
-        knn_indices=knn_indices,
-        knn_dists=knn_dists,
+        # umap ≥0.6 requires C-contiguous arrays
+        knn_indices=np.ascontiguousarray(knn_indices),
+        knn_dists=np.ascontiguousarray(knn_dists),
         set_op_mix_ratio=set_op_mix_ratio,
         local_connectivity=local_connectivity,
     )

src/scanpy/tools/_ingest.py

@@ -2,6 +2,7 @@
 
 import contextlib
 from collections.abc import MutableMapping
+from inspect import signature
 from typing import TYPE_CHECKING
 
 import numpy as np
@@ -238,7 +239,14 @@ def _init_umap(self, adata: AnnData, keys: _EmbeddingKeys) -> None:
         from umap import UMAP
 
         rng = self._get_rng(adata.uns[keys.uns]["params"])
+        # we replicate the legacy (0.5.x) transform pipeline below
+        compat = (
+            dict(compatibility_layout=True)
+            if "compatibility_layout" in signature(UMAP).parameters
+            else {}
+        )
         self._umap = UMAP(
+            **compat,
             metric=self._metric,
             random_state=_legacy_random_state(rng),
             # umap can’t be run in parallel with `random_state is not None`

tests/test_ingest.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+from inspect import signature
 from typing import TYPE_CHECKING
 
 import anndata
@@ -215,7 +216,13 @@ def test_ingest_map_embedding_umap() -> None:
     ing.fit(adata_new)
     ing.map_embedding(method="umap")
 
-    reducer = UMAP(min_dist=0.5, random_state=0, n_neighbors=4, n_jobs=1)
+    # ingest replicates the legacy (0.5.x) pipeline
+    compat = (
+        dict(compatibility_layout=True)
+        if "compatibility_layout" in signature(UMAP).parameters
+        else {}
+    )
+    reducer = UMAP(min_dist=0.5, random_state=0, n_neighbors=4, n_jobs=1, **compat)
     reducer.fit(X)
     umap_transformed_t = reducer.transform(T)