Backport PR #4045 on branch 1.12.x (fix: use weights to calculate modularity) (#4048)

Author: flying-sheep

.github/workflows/check-pr.yml

@@ -60,7 +60,7 @@ jobs:
         id: changes
         with:
           filters: | # this is intentionally a string
-            relnotes: 'docs/release-notes/${{ github.event.pull_request.number }}.*.md'
+            relnotes: 'docs/release-notes/${{ github.event.pull_request.number }}.${{ (contains(github.event.pull_request.title, '!') && 'breaking') || needs.check-milestone.outputs.type }}.md'
       - name: Check if a relevant release fragment is added
         uses: flying-sheep/check@v1
         with:

docs/release-notes/4045.fix.md

@@ -0,0 +1 @@
+Make {func}`scanpy.metrics.modularity` actually use edge weights {smaller}`P Angerer`

pyproject.toml

@@ -63,17 +63,17 @@ dependencies = [
     "numba>=0.60",
     "numpy>=2",
     "packaging>=25",
-    "pandas>=2.2.2",
+    "pandas>=2.3",
     "patsy",
     "pynndescent>=0.5.13",
-    "scikit-learn>=1.4.2",
+    "scikit-learn>=1.6",
     "scipy>=1.13",
     "seaborn>=0.13.2",
     "session-info2",
     "statsmodels>=0.14.5",
     "tqdm",
     "typing-extensions; python_version<'3.13'",
-    "umap-learn>=0.5.7",
+    "umap-learn>=0.5.12",
 ]
 urls.Bluesky = "https://bsky.app/profile/scverse.bsky.social"
 urls.Discourse = "https://discourse.scverse.org/c/help/scanpy/37"

src/scanpy/metrics/_metrics.py

@@ -212,7 +212,7 @@ def modularity_array(
         raise
     igraph_mode: str = ig.ADJ_DIRECTED if is_directed else ig.ADJ_UNDIRECTED
     graph: ig.Graph = ig.Graph.Weighted_Adjacency(connectivities, mode=igraph_mode)
-    return graph.modularity(_codes(labels))
+    return graph.modularity(_codes(labels), "weight")
 
 
 def _codes(labels: AnyArrayLike) -> AnyArrayLike:

src/testing/scanpy/_pytest/marks.py

@@ -1,11 +1,9 @@
 from __future__ import annotations
 
 from enum import Enum, auto
-from importlib.metadata import version
 from importlib.util import find_spec
 
 import pytest
-from packaging.version import Version
 
 
 class QuietMarkDecorator(pytest.MarkDecorator):
@@ -71,9 +69,3 @@ def skip_reason(self) -> str | None:
         if self._name_.casefold() != self.mod.casefold().replace("-", "_"):
             reason = f"{reason} (`pip install {self.mod}`)"
         return reason
-
-
-# TODO: remove once https://github.com/numba/numba/issues/10319 is fixed
-skip_numba_0_63 = pytest.mark.skipif(
-    Version(version=version("numba")) >= Version("0.63b0"), reason="numba 0.63 bug"
-)

tests/test_highly_variable_genes.py

@@ -19,7 +19,7 @@
 from scanpy._compat import CSRBase
 from testing.scanpy._helpers import _check_check_values_warnings
 from testing.scanpy._helpers.data import pbmc3k, pbmc68k_reduced
-from testing.scanpy._pytest.marks import needs, skip_numba_0_63
+from testing.scanpy._pytest.marks import needs
 from testing.scanpy._pytest.params import ARRAY_TYPES
 
 if TYPE_CHECKING:
@@ -166,7 +166,6 @@ def _check_pearson_hvg_columns(output_df: pd.DataFrame, n_top_genes: int):
     assert np.nanmax(output_df["highly_variable_rank"].to_numpy()) <= n_top_genes - 1
 
 
-@skip_numba_0_63
 def test_pearson_residuals_inputchecks(
     pbmc3k_parametrized_small: Callable[[], AnnData],
 ) -> None:
@@ -203,7 +202,6 @@ def test_pearson_residuals_inputchecks(
         )
 
 
-@skip_numba_0_63
 @pytest.mark.parametrize("subset", [True, False], ids=["subset", "full"])
 @pytest.mark.parametrize(
     "clip", [None, np.inf, 30], ids=["noclip", "infclip", "30clip"]
@@ -297,7 +295,6 @@ def test_pearson_residuals_general(
     _check_pearson_hvg_columns(output_df, n_top_genes)
 
 
-@skip_numba_0_63
 @pytest.mark.parametrize("subset", [True, False], ids=["subset", "full"])
 @pytest.mark.parametrize("n_top_genes", [100, 200], ids=["100n", "200n"])
 def test_pearson_residuals_batch(

tests/test_metrics.py

@@ -11,14 +11,11 @@
 import pytest
 import threadpoolctl
 from anndata import AnnData
-from packaging.version import Version
 from scipy import sparse
 
 import scanpy as sc
-from scanpy._compat import pkg_version
 from scanpy.metrics import modularity
 from testing.scanpy._helpers.data import pbmc3k, pbmc68k_reduced
-from testing.scanpy._pytest.context import xfail
 from testing.scanpy._pytest.marks import needs
 from testing.scanpy._pytest.params import ARRAY_TYPES
 
@@ -336,15 +333,8 @@ def test_modularity_adata(
         with subtests.test("bounds", score=name):
             assert 0 <= s <= 1
     for (n0, s0), (n1, s1) in combinations(scores.items(), 2):
-        approx = {n0, n1} != {"update", "calculate"}
-        with (
-            subtests.test("equality", l=n0, r=n1),
-            xfail(
-                approx and pkg_version("igraph") < Version("1"),
-                reason="igraph 0.x has different modularity behavior",
-            ),
-        ):
-            assert pytest.approx(s0, rel=1e-3 if approx else 1e-6) == s1
+        with subtests.test("equality", l=n0, r=n1):
+            assert pytest.approx(s0, rel=1e-6) == s1
     with subtests.test("update"):
         assert adata.uns["leiden"]["modularity"] is scores["update"]
 

tests/test_neighbors.py

@@ -5,27 +5,19 @@
 import numpy as np
 import pytest
 from anndata import AnnData
-from packaging.version import Version
 from scipy import sparse
 from sklearn.neighbors import KNeighborsTransformer
 
 import scanpy as sc
 from scanpy import Neighbors
-from scanpy._compat import CSBase, pkg_version
+from scanpy._compat import CSBase
 from testing.scanpy._helpers.data import pbmc68k_reduced
 
 if TYPE_CHECKING:
     from typing import Literal
 
     from pytest_mock import MockerFixture
 
-# https://github.com/lmcinnes/umap/issues/1216
-XFAIL_IF_UMAP_BROKEN = pytest.mark.xfail(
-    pkg_version("umap-learn") < Version("0.6a0.dev0")
-    and pkg_version("numba") >= Version("0.62.0rc1"),
-    reason="umap<0.6 is broken with numba≥0.62.0rc1",
-)
-
 # the input data
 X = [[1, 0], [3, 0], [5, 6], [0, 4]]
 n_neighbors = 3  # includes data points themselves
@@ -207,7 +199,6 @@ def test_distances_all(neigh: Neighbors, transformer, knn):
             connectivities_umap,
             transitions_umap,
             transitions_sym_umap,
-            marks=XFAIL_IF_UMAP_BROKEN,
             id="umap",
         ),
         pytest.param(

tests/test_normalization.py

@@ -18,7 +18,6 @@
     check_rep_mutation,
     check_rep_results,
 )
-from testing.scanpy._pytest.marks import skip_numba_0_63
 
 # TODO: Add support for sparse-in-dask
 from testing.scanpy._pytest.params import ARRAY_TYPES, ARRAY_TYPES_DENSE
@@ -210,7 +209,6 @@ def _check_pearson_pca_fields(ad, n_cells, n_comps):
     ), "Wrong shape of PCA output in `X_pca`"
 
 
-@skip_numba_0_63
 @pytest.mark.parametrize("n_hvgs", [100, 200])
 @pytest.mark.parametrize("n_comps", [30, 50])
 @pytest.mark.parametrize(
@@ -283,7 +281,6 @@ def test_normalize_pearson_residuals_pca(
     np.testing.assert_array_equal(adata.obsm["X_pca"], adata_pca.obsm["X_pca"])
 
 
-@skip_numba_0_63
 @pytest.mark.parametrize("n_hvgs", [100, 200])
 @pytest.mark.parametrize("n_comps", [30, 50])
 def test_normalize_pearson_residuals_recipe(