fix

Author: flying-sheep

.github/workflows/pytest.yml

@@ -10,7 +10,7 @@ jobs:
   test:
     strategy:
       matrix:
-        python-version: ["3.9", "3.13"]
+        python-version: ["3.10", "3.14"]
     runs-on: ubuntu-latest
     timeout-minutes: 60
     steps:
@@ -27,7 +27,7 @@ jobs:
           cache: pip
 
       - name: Install project and dependencies
-        run: pip install .[annlibs,tests]
+        run: pip install --group=tests .[annlibs]
 
       - name: Run test suite
         run: pytest -v --color=yes

.pre-commit-config.yaml

@@ -17,15 +17,20 @@ repos:
   - repo: https://github.com/astral-sh/ruff-pre-commit
     rev: v0.16.1
     hooks:
-      - id: ruff
+      - id: ruff-check
         args: ["--fix"]
       - id: ruff-format
   - repo: https://github.com/pre-commit/mirrors-mypy
-    rev: v1.14.1
+    rev: v2.3.0
     hooks:
       - id: mypy
         additional_dependencies:
+          - faiss-cpu
+          - matplotlib
           - pynndescent
+          - pytest
+          - scikit-learn
+          - scipy-stubs
   - repo: https://github.com/pre-commit/mirrors-prettier
     rev: v4.0.0-alpha.8
     hooks:

.readthedocs.yaml

@@ -0,0 +1,18 @@
+# https://docs.readthedocs.io/en/stable/config-file/v2.html
+version: 2
+build:
+  os: ubuntu-26.04
+  tools:
+    python: "3.14"
+  jobs:
+    post_checkout:
+      # unshallow so version can be derived from tag
+      - git fetch --unshallow || true
+    create_environment:
+      - asdf plugin add uv
+      - asdf install uv latest
+      - asdf global uv latest
+    build:
+      html:
+        - uvx hatch run docs:build
+        - mv docs/_build $READTHEDOCS_OUTPUT

.readthedocs.yml

@@ -19,17 +19,16 @@
 from sklearn_ann.cluster.rnn_dbscan import simple_rnn_dbscan_pipeline
 
 if TYPE_CHECKING:
+    from collections.abc import Callable
     from typing import Any
 
-    from sklearn.utils import Bunch
-
 
 # #############################################################################
 # Generate sample data
-def fetch_mnist() -> Bunch:
+def fetch_mnist() -> tuple[Any, Any]:
     print("Downloading mnist_784")
-    mnist = fetch_openml("mnist_784")
-    return mnist.data / 255, mnist.target
+    data, target = fetch_openml("mnist_784", return_X_y=True)
+    return data / 255, target
 
 
 memory = Memory("./mnist")
@@ -38,7 +37,7 @@ def fetch_mnist() -> Bunch:
 
 
 def run_rnn_dbscan(
-    neighbor_transformer: object, n_neighbors: int, **kwargs: Any
+    neighbor_transformer: Callable[..., Any], n_neighbors: int, **kwargs: Any
 ) -> None:
     # #############################################################################
     # Compute RnnDBSCAN

examples/rnn_dbscan_big.py

@@ -8,6 +8,7 @@
 Mostly copypasted from sklearn's DBSCAN example.
 
 """
+
 from __future__ import annotations
 
 import numpy as np
@@ -35,7 +36,7 @@
 
 # Number of clusters in labels, ignoring noise if present.
 n_clusters_ = len(set(labels)) - (1 if -1 in labels else 0)
-n_noise_ = list(labels).count(-1)
+n_noise_ = int((labels == -1).sum())
 
 print(f"""\
 Estimated number of clusters: {n_clusters_}

examples/rnn_dbscan_simple.py

@@ -9,26 +9,13 @@ license = "BSD-3-Clause"
 urls.Source = "https://github.com/scikit-learn-contrib/sklearn-ann"
 urls.Documentation = "https://sklearn-ann.readthedocs.io/"
 dynamic = ["version", "readme"]
-requires-python = ">=3.9"
+requires-python = ">=3.10"
 dependencies = [
     "scikit-learn>=1.6.0",
     "scipy>=1.11.1,<2.0.0",
 ]
 
 [project.optional-dependencies]
-tests = [
-    "pytest>=6.2.1",
-    "pytest-cov>=2.10.1",
-]
-docs = [
-    "sphinx>=7",
-    "sphinx-gallery>=0.8.2",
-    "sphinx-book-theme>=1.1.0",
-    "sphinx-issues>=1.2.0",
-    "numpydoc>=1.1.0",
-    "matplotlib>=3.3.3",
-    "scanpydoc",
-]
 annoy = [
     "annoy>=1.17.0,<2.0.0",
 ]
@@ -46,6 +33,21 @@ annlibs = [
     "sklearn-ann[annoy,faiss,pynndescent,nmslib]",
 ]
 
+[dependency-groups]
+tests = [
+    "pytest>=6.2.1",
+    "pytest-cov>=2.10.1",
+]
+docs = [
+    "sphinx>=7",
+    "sphinx-gallery>=0.8.2",
+    "sphinx-book-theme>=1.1.0",
+    "sphinx-issues>=1.2.0",
+    "numpydoc>=1.1.0",
+    "matplotlib>=3.3.3",
+    "scanpydoc",
+]
+
 [tool.hatch.version]
 source = "vcs"
 
@@ -92,23 +94,33 @@ strict = true
 
 [tool.hatch.envs.docs]
 installer = "uv"
-features = ["docs", "annlibs"]
+features = ["annlibs"]
+dependency-groups = ["docs"]
 scripts.build = "sphinx-build -M html docs docs/_build"
 
 [tool.hatch.envs.hatch-test]
 default-args = []
-features = ["tests", "annlibs"]
+features = ["annlibs"]
+dependency-groups = ["tests"]
 
 [tool.hatch.build.targets.wheel]
 packages = ["src/sklearn_ann"]
 
 [tool.mypy]
-python_version = "3.11"
+python_version = "3.12"
 mypy_path = ["src", "tests"]
 strict = true
 explicit_package_bases = true # pytest doesn’t do __init__.py
 no_implicit_optional = true
 disallow_untyped_decorators = false # e.g. pytest.mark.parametrize
+follow_untyped_imports = true
+# follow_untyped_imports makes these readable, but not annotated
+untyped_calls_exclude = ["sklearn", "joblib", "pynndescent"]
+
+[[tool.mypy.overrides]]
+# pynndescent has no __all__, so its re-exports are implicit
+module = ["pynndescent"]
+no_implicit_reexport = false
 
 [build-system]
 requires = ["hatchling", "hatch-vcs", "hatch-fancy-pypi-readme"]

pyproject.toml

@@ -4,33 +4,35 @@
 from typing import TYPE_CHECKING, cast
 
 import numpy as np
-from scipy.sparse import csr_matrix
 from sklearn.base import BaseEstimator, ClusterMixin
 from sklearn.neighbors import KNeighborsTransformer
-from sklearn.utils import Tags
 from sklearn.utils.validation import validate_data
 
 from ..utils import get_sparse_row
 
 if TYPE_CHECKING:
-    from collections.abc import Iterator
-    from typing import Literal, Self
+    from collections.abc import Callable, Iterator
+    from typing import Any, Literal, Self
 
     from numpy.typing import NDArray
+    from scipy.sparse import csr_matrix
     from sklearn.pipeline import Pipeline
+    from sklearn.utils import Tags
 
 
 UNCLASSIFIED = -2
 NOISE = -1
 
 
-def join(it1, it2):
+def join(
+    it1: Iterator[tuple[int, float]], it2: Iterator[tuple[int, float]]
+) -> Iterator[tuple[int, float]]:
     cur_it1 = next(it1, None)
     cur_it2 = next(it2, None)
     while 1:
-        if cur_it1 is None and cur_it2 is None:
-            break
-        elif cur_it1 is None:
+        if cur_it1 is None:
+            if cur_it2 is None:
+                break
             yield cur_it2
             cur_it2 = next(it2, None)
         elif cur_it2 is None:
@@ -66,7 +68,10 @@ def neighborhood(
 
 
 def rnn_dbscan_inner(
-    is_core: NDArray[np.bool_], knns: csr_matrix, rev_knns: csr_matrix, labels
+    is_core: NDArray[np.bool_],
+    knns: csr_matrix,
+    rev_knns: csr_matrix,
+    labels: NDArray[np.int32],
 ) -> list[float]:
     cluster = 0
     cur_dens = 0.0
@@ -78,7 +83,7 @@ def rnn_dbscan_inner(
                 labels[x_idx] = cluster
                 # TODO: Make this inner bit faster - can just assume
                 # sorted an keep sorted
-                seeds = deque()
+                seeds: deque[int] = deque()
                 for neighbor_idx, dist in neighborhood(is_core, knns, rev_knns, x_idx):
                     labels[neighbor_idx] = cluster
                     if dist > cur_dens:
@@ -166,37 +171,39 @@ def __init__(
         self.keep_knns = keep_knns
 
     def fit(self, X: NDArray[np.float64] | csr_matrix, y: None = None) -> Self:
-        X = cast(csr_matrix, validate_data(self, X, accept_sparse="csr"))
+        X = validate_data(self, X, accept_sparse="csr")
         if self.input_guarantee == "none":
             algorithm = KNeighborsTransformer(n_neighbors=self.n_neighbors)
-            X = algorithm.fit_transform(X)
+            knns = cast("csr_matrix", algorithm.fit_transform(X))
         elif self.input_guarantee == "kneighbors":
-            pass
+            knns = cast("csr_matrix", X)
         else:
             raise ValueError(
                 "Expected input_guarantee to be one of 'none', 'kneighbors'"
             )
 
-        XT = cast(csr_matrix, X.transpose().tocsr(copy=True))
+        rev_knns = knns.transpose().tocsr(copy=True)
         if self.keep_knns:
-            self.knns_ = X
-            self.rev_knns_ = XT
+            self.knns_ = knns
+            self.rev_knns_ = rev_knns
 
         # Initially, all samples are unclassified.
-        labels = np.full(X.shape[0], UNCLASSIFIED, dtype=np.int32)
+        labels = np.full(knns.shape[0], UNCLASSIFIED, dtype=np.int32)
 
         # A list of all core samples found. -1 is to account for diagonal.
-        core_samples = XT.getnnz(1) - 1 >= self.n_neighbors
+        core_samples = rev_knns.getnnz(1) - 1 >= self.n_neighbors
 
-        dens = rnn_dbscan_inner(core_samples, X, XT, labels)
+        dens = rnn_dbscan_inner(core_samples, knns, rev_knns, labels)
 
         self.core_sample_indices_ = core_samples.nonzero()
         self.labels_ = labels
         self.dens_ = dens
 
         return self
 
-    def fit_predict(self, X, y=None) -> NDArray[np.int32]:
+    def fit_predict(  # type: ignore[override]
+        self, X: NDArray[np.float64] | csr_matrix, y: None = None
+    ) -> NDArray[np.int32]:
         self.fit(X, y=y)
         return self.labels_
 
@@ -205,13 +212,13 @@ def drop_knns(self) -> None:
         del self.rev_knns_
 
     def __sklearn_tags__(self) -> Tags:
-        tags = cast(Tags, super().__sklearn_tags__())
+        tags = cast("Tags", super().__sklearn_tags__())  # type: ignore[no-untyped-call]
         tags.input_tags.sparse = True
         return tags
 
 
 def simple_rnn_dbscan_pipeline(
-    neighbor_transformer: object,
+    neighbor_transformer: Callable[..., Any],
     n_neighbors: int,
     *,
     n_jobs: int | None = None,
@@ -236,12 +243,17 @@ class implementing KNeighborsTransformer interface
     """
     from sklearn.pipeline import make_pipeline
 
-    return make_pipeline(
-        neighbor_transformer(n_neighbors=n_neighbors, n_jobs=n_jobs, **kwargs),
-        RnnDBSCAN(
-            n_neighbors=n_neighbors,
-            input_guarantee="kneighbors",
-            n_jobs=n_jobs,
-            keep_knns=keep_knns,
+    return cast(
+        "Pipeline",
+        make_pipeline(
+            neighbor_transformer(
+                n_neighbors=n_neighbors, n_jobs=n_jobs, input_guarantee=input_guarantee
+            ),
+            RnnDBSCAN(
+                n_neighbors=n_neighbors,
+                input_guarantee="kneighbors",
+                n_jobs=n_jobs,
+                keep_knns=keep_knns,
+            ),
         ),
     )