Minimum dependency test job

ci/scripts/min-deps.py

@@ -0,0 +1,98 @@
+#!python3
+from __future__ import annotations
+
+import argparse
+import sys
+from pathlib import Path
+
+if sys.version_info >= (3, 11):
+    import tomllib
+else:
+    import tomli as tomllib
+from packaging.requirements import Requirement
+from packaging.version import Version
+
+
+def min_dep(req: Requirement) -> str:
+    """
+    Given a requirement, return the minimum version specifier.
+
+    Example
+    -------
+
+    >>> min_dep(Requirement("numpy>=1.0"))
+    "numpy==1.0"
+    """
+    req_name = req.name
+    if req.extras:
+        req_name = f"{req_name}[{','.join(req.extras)}]"
+
+    # TODO: Should this be allowed?
+    if not req.specifier:
+        return req_name
+
+    min_version = Version("0.0.0.a1")
+    for spec in req.specifier:
+        if spec.operator in [">", ">=", "~-"]:
+            min_version = max(min_version, Version(spec.version))
+        elif spec.operator == "==":
+            min_version = Version(spec.version)
+
+    # TODO: should this return `~=` or `==`?
+    return f"{req_name}=={min_version}.*"
+
+
+def extract_min_deps(
+        dependencies: list[str], 
+        *, 
+        pyproject
+    ) -> list[str]:
+    dependencies = dependencies.copy()  # We'll be mutating this
+    requirements: list[Requirement] = []
+    project_name = pyproject["project"]["name"]
+
+    while len(dependencies) > 0:
+        req = Requirement(dependencies.pop())
+
+        # If we are reffering to other optional dependency lists, resolve them
+        if req.name == project_name:
+            assert req.extras, f"Project included itself as dependency, without specifying extras: {req}"
+            for extra in req.extras:
+                dependencies.extend(pyproject["project"]["optional-dependencies"][extra])
+        else:
+            requirements.append(min_dep(req))
+
+    return requirements
+
+
+def main():
+    # TODO: Allow optional dependencies
+    parser = argparse.ArgumentParser(
+        prog="min-deps",
+        description="""Parse a pyproject.toml file and output a list of minimum dependencies.
+
+        Output is directly passable to `pip install`.""",
+        usage="pip install `python min-deps.py pyproject.toml`",
+    )
+    parser.add_argument(
+        "path", type=Path, help="pyproject.toml to parse minimum dependencies from"
+    )
+    parser.add_argument("--extras", type=str, nargs="*", help="extras to install")
+
+    args = parser.parse_args()
+
+    pyproject = tomllib.loads(args.path.read_text())
+
+    project_name = pyproject["project"]["name"]
+    deps = pyproject["project"]["dependencies"]
+
+    for extra in args.extras:
+        deps.append(f"{project_name}[{extra}]")
+
+    min_deps = extract_min_deps(deps, pyproject=pyproject)
+
+    print(" ".join(min_deps))
+
+
+if __name__ == "__main__":
+    main()

ci/scripts/run-min-deps.sh

@@ -0,0 +1,19 @@
+set -e
+
+mamba env remove -yn scanpy-min-deps-test
+mamba create -yn scanpy-min-deps-test "python=3.9"
+
+PACKAGES=`python3 ci/scripts/min-deps.py pyproject.toml --extra dev test`
+
+# conda activate anndata-min-deps-test
+# conda run -n anndata-min-deps-test pip install cupy-cuda12x
+
+
+echo Installing $PACKAGES
+conda run -n scanpy-min-deps-test pip install $PACKAGES
+conda run -n scanpy-min-deps-test pip install pytest-xdist # cupy-cuda12x
+conda run -n scanpy-min-deps-test pip install -e . --no-deps
+echo "Starting tests"
+conda run -n scanpy-min-deps-test pytest -n auto
+
+conda list -n scanpy-min-deps-tests

pyproject.toml

@@ -46,24 +46,24 @@ classifiers = [
     "Topic :: Scientific/Engineering :: Visualization",
 ]
 dependencies = [
-    "anndata>=0.7.4",
+    "anndata>=0.10",
     # numpy needs a version due to #1320
-    "numpy>=1.17.0",
+    "numpy>=1.23",
     "matplotlib>=3.6",
-    "pandas >=2.1.3",
-    "scipy>=1.4",
-    "seaborn>=0.13.0",
-    "h5py>=3",
+    "pandas >=1.5",
+    "scipy>=1.8",
+    "seaborn>=0.13",
+    "h5py>=3.1",
     "tqdm",
     "scikit-learn>=0.24",
-    "statsmodels>=0.10.0rc2",
+    "statsmodels>=0.10",
     "patsy",
-    "networkx>=2.3",
+    "networkx>=2.6",
     "natsort",
     "joblib",
-    "numba>=0.41.0",
+    "numba>=0.56",
     "umap-learn>=0.3.10",
-    "packaging",
+    "packaging>=20.0",
     "session-info",
     "legacy-api-wrap>=1.4",  # for positional API deprecations
     "get-annotations; python_version < '3.10'",
@@ -133,8 +133,8 @@ dev = [
 ]
 # Algorithms
 paga = ["igraph"]
-louvain = ["igraph", "louvain>=0.6,!=0.6.2"]  # Louvain community detection
-leiden = ["igraph>=0.10", "leidenalg>=0.9"]  # Leiden community detection
+louvain = ["igraph", "louvain>=0.6.0,!=0.6.2"]  # Louvain community detection
+leiden = ["igraph>=0.10", "leidenalg>=0.9.0"]  # Leiden community detection
 bbknn = ["bbknn"]  # Batch balanced KNN (batch correction)
 magic = ["magic-impute>=2.0"]  # MAGIC imputation method
 skmisc = ["scikit-misc>=0.1.3"]  # highly_variable_genes method 'seurat_v3'
@@ -143,7 +143,7 @@ scanorama = ["scanorama"]  # Scanorama dataset integration
 scrublet = ["scrublet"]  # Doublet detection
 # Acceleration
 rapids = ["cudf>=0.9", "cuml>=0.9", "cugraph>=0.9"]  # GPU accelerated calculation of neighbors
-dask = ["dask[array]!=2.17.0"]  # Use the Dask parallelization engine
+dask = ["dask[array]>=2022.09"]  # Use the Dask parallelization engine
 dask-ml = ["dask-ml", "scanpy[dask]"]  # Dask-ML for sklearn-like API
 
 [tool.hatch.build]

scanpy/neighbors/_backends/rapids.py

@@ -3,8 +3,9 @@
 from typing import TYPE_CHECKING, Any, Literal
 
 import numpy as np
-from sklearn.base import BaseEstimator, TransformerMixin, check_is_fitted
+from sklearn.base import BaseEstimator, TransformerMixin
 from sklearn.exceptions import NotFittedError
+from sklearn.utils.validation import check_is_fitted
 
 from ..._settings import settings
 from ._common import TransformerChecksMixin

scanpy/plotting/_tools/scatterplots.py

@@ -12,6 +12,7 @@
 
 import numpy as np
 import pandas as pd
+from packaging.version import Version
 from anndata import AnnData  # noqa: TCH002
 from cycler import Cycler  # noqa: TCH002
 from matplotlib import colormaps, colors, patheffects, rcParams
@@ -1256,8 +1257,10 @@ def _color_vector(
     }
     # If color_map does not have unique values, this can be slow as the
     # result is not categorical
-    color_vector = pd.Categorical(values.map(color_map, na_action="ignore"))
-
+    if Version(pd.__version__) < Version("2.1.0"):
+        color_vector = pd.Categorical(values.map(color_map))
+    else:
+        color_vector = pd.Categorical(values.map(color_map, na_action="ignore"))
     # Set color to 'missing color' for all missing values
     if color_vector.isna().any():
         color_vector = color_vector.add_categories([to_hex(na_color)])

scanpy/plotting/_utils.py

@@ -381,7 +381,7 @@ def _validate_palette(adata: AnnData, key: str) -> None:
                 break
         _palette.append(color)
     # Don't modify if nothing changed
-    if _palette is None or np.equal(_palette, adata.uns[color_key]).all():
+    if _palette is None or _palette == list(adata.uns[color_key]):
         return
     adata.uns[color_key] = _palette