Fix rtol for scipy iterative solver, bump version (theislab#1237)

* Fix rtol for sciPy iterative solver, bump version

* Test on 3.13

* Try patching `.A`

* Add 3.13 runner

* Don't run on 3.13 (numba)

* Test on `macos-15`

.github/workflows/test.yml

@@ -22,7 +22,7 @@ jobs:
         python: ['3.9', '3.10', '3.11', '3.12']
         slepc: [noslepc]
         include:
-        - os: macos-14
+        - os: macos-15
           python: '3.10'
           slepc: noslepc
         - os: ubuntu-latest

pyproject.toml

@@ -57,7 +57,7 @@ dependencies = [
     "pygpcca>=1.0.4",
     "scanpy>=1.7.2",
     "scikit-learn>=0.24.0",
-    "scipy>=1.2.0",
+    "scipy>=1.12.0",
     "scvelo>=0.2.5",
     "seaborn>=0.10.0",
     "wrapt>=1.12.1",
@@ -239,7 +239,7 @@ legacy_tox_ini = """
 # TODO(michalk8): upgrade to `tox>=4.0` once `tox-conda` supports it
 requires = tox-conda
 isolated_build = true
-envlist = lint-code,py{3.9,3.10,3.11,3.12}-{slepc,noslepc}
+envlist = lint-code,py{3.9,3.10,3.11,3.12,3.13}-{slepc,noslepc}
 skip_missing_interpreters = true
 
 [testenv]

src/cellrank/__init__.py

@@ -1,5 +1,7 @@
 from importlib import metadata
 
+import scipy.sparse as sp
+
 from cellrank import datasets, estimators, kernels, logging, models, pl
 from cellrank._utils._lineage import Lineage
 from cellrank.settings import settings
@@ -14,4 +16,7 @@
 except ImportError:
     md = None
 
-del metadata, md
+# pygam uses `.A`
+sp.spmatrix.A = property(lambda self: self.toarray())
+
+del metadata, md, sp

src/cellrank/_utils/_linear_solver.py

@@ -223,11 +223,9 @@ def _solve_many_sparse_problems(
     """
     # initialise solution list and info list
     x_list, n_converged = [], 0
-    kwargs = {} if solver is not sp.linalg.gmres else {"atol": "legacy"}  # get rid of the warning
-
     for b in mat_b:
         # actually call the solver for the current sub-problem
-        x, info = solver(mat_a, b.toarray().flatten(), tol=tol, x0=None, **kwargs)
+        x, info = solver(mat_a, b.toarray().flatten(), rtol=tol, x0=None)
 
         # append solution and info
         x_list.append(np.atleast_1d(x))