library packages

.venv/lib/python3.12/site-packages/sklearn/__check_build/__init__.py

@@ -0,0 +1,51 @@
+"""Module to give helpful messages to the user that did not
+compile scikit-learn properly.
+"""
+
+import os
+
+INPLACE_MSG = """
+It appears that you are importing a local scikit-learn source tree. For
+this, you need to have an inplace install. Maybe you are in the source
+directory and you need to try from another location."""
+
+STANDARD_MSG = """
+If you have used an installer, please check that it is suited for your
+Python version, your operating system and your platform."""
+
+
+def raise_build_error(e):
+    # Raise a comprehensible error and list the contents of the
+    # directory to help debugging on the mailing list.
+    local_dir = os.path.split(__file__)[0]
+    msg = STANDARD_MSG
+    if local_dir == "sklearn/__check_build":
+        # Picking up the local install: this will work only if the
+        # install is an 'inplace build'
+        msg = INPLACE_MSG
+    dir_content = list()
+    for i, filename in enumerate(os.listdir(local_dir)):
+        if (i + 1) % 3:
+            dir_content.append(filename.ljust(26))
+        else:
+            dir_content.append(filename + "\n")
+    raise ImportError(
+        """%s
+___________________________________________________________________________
+Contents of %s:
+%s
+___________________________________________________________________________
+It seems that scikit-learn has not been built correctly.
+
+If you have installed scikit-learn from source, please do not forget
+to build the package before using it: run `python setup.py install` or
+`make` in the source directory.
+%s"""
+        % (e, local_dir, "".join(dir_content).strip(), msg)
+    )
+
+
+try:
+    from ._check_build import check_build  # noqa
+except ImportError as e:
+    raise_build_error(e)

.venv/lib/python3.12/site-packages/sklearn/__check_build/_check_build.pyx

@@ -0,0 +1,2 @@
+def check_build():
+    return

.venv/lib/python3.12/site-packages/sklearn/__check_build/meson.build

@@ -0,0 +1,7 @@
+py.extension_module(
+  '_check_build',
+  '_check_build.pyx',
+  cython_args: cython_args,
+  install: true,
+  subdir: 'sklearn/__check_build',
+)

.venv/lib/python3.12/site-packages/sklearn/__init__.py

@@ -0,0 +1,154 @@
+"""Configure global settings and get information about the working environment."""
+
+# Machine learning module for Python
+# ==================================
+#
+# sklearn is a Python module integrating classical machine
+# learning algorithms in the tightly-knit world of scientific Python
+# packages (numpy, scipy, matplotlib).
+#
+# It aims to provide simple and efficient solutions to learning problems
+# that are accessible to everybody and reusable in various contexts:
+# machine-learning as a versatile tool for science and engineering.
+#
+# See https://scikit-learn.org for complete documentation.
+
+import logging
+import os
+import random
+import sys
+
+from ._config import config_context, get_config, set_config
+
+logger = logging.getLogger(__name__)
+
+
+# PEP0440 compatible formatted version, see:
+# https://www.python.org/dev/peps/pep-0440/
+#
+# Generic release markers:
+#   X.Y.0   # For first release after an increment in Y
+#   X.Y.Z   # For bugfix releases
+#
+# Admissible pre-release markers:
+#   X.Y.ZaN   # Alpha release
+#   X.Y.ZbN   # Beta release
+#   X.Y.ZrcN  # Release Candidate
+#   X.Y.Z     # Final release
+#
+# Dev branch marker is: 'X.Y.dev' or 'X.Y.devN' where N is an integer.
+# 'X.Y.dev0' is the canonical version of 'X.Y.dev'
+#
+__version__ = "1.5.2"
+
+
+# On OSX, we can get a runtime error due to multiple OpenMP libraries loaded
+# simultaneously. This can happen for instance when calling BLAS inside a
+# prange. Setting the following environment variable allows multiple OpenMP
+# libraries to be loaded. It should not degrade performances since we manually
+# take care of potential over-subcription performance issues, in sections of
+# the code where nested OpenMP loops can happen, by dynamically reconfiguring
+# the inner OpenMP runtime to temporarily disable it while under the scope of
+# the outer OpenMP parallel section.
+os.environ.setdefault("KMP_DUPLICATE_LIB_OK", "True")
+
+# Workaround issue discovered in intel-openmp 2019.5:
+# https://github.com/ContinuumIO/anaconda-issues/issues/11294
+os.environ.setdefault("KMP_INIT_AT_FORK", "FALSE")
+
+try:
+    # This variable is injected in the __builtins__ by the build
+    # process. It is used to enable importing subpackages of sklearn when
+    # the binaries are not built
+    # mypy error: Cannot determine type of '__SKLEARN_SETUP__'
+    __SKLEARN_SETUP__  # type: ignore
+except NameError:
+    __SKLEARN_SETUP__ = False
+
+if __SKLEARN_SETUP__:
+    sys.stderr.write("Partial import of sklearn during the build process.\n")
+    # We are not importing the rest of scikit-learn during the build
+    # process, as it may not be compiled yet
+else:
+    # `_distributor_init` allows distributors to run custom init code.
+    # For instance, for the Windows wheel, this is used to pre-load the
+    # vcomp shared library runtime for OpenMP embedded in the sklearn/.libs
+    # sub-folder.
+    # It is necessary to do this prior to importing show_versions as the
+    # later is linked to the OpenMP runtime to make it possible to introspect
+    # it and importing it first would fail if the OpenMP dll cannot be found.
+    from . import (
+        __check_build,  # noqa: F401
+        _distributor_init,  # noqa: F401
+    )
+    from .base import clone
+    from .utils._show_versions import show_versions
+
+    __all__ = [
+        "calibration",
+        "cluster",
+        "covariance",
+        "cross_decomposition",
+        "datasets",
+        "decomposition",
+        "dummy",
+        "ensemble",
+        "exceptions",
+        "experimental",
+        "externals",
+        "feature_extraction",
+        "feature_selection",
+        "gaussian_process",
+        "inspection",
+        "isotonic",
+        "kernel_approximation",
+        "kernel_ridge",
+        "linear_model",
+        "manifold",
+        "metrics",
+        "mixture",
+        "model_selection",
+        "multiclass",
+        "multioutput",
+        "naive_bayes",
+        "neighbors",
+        "neural_network",
+        "pipeline",
+        "preprocessing",
+        "random_projection",
+        "semi_supervised",
+        "svm",
+        "tree",
+        "discriminant_analysis",
+        "impute",
+        "compose",
+        # Non-modules:
+        "clone",
+        "get_config",
+        "set_config",
+        "config_context",
+        "show_versions",
+    ]
+
+    _BUILT_WITH_MESON = False
+    try:
+        import sklearn._built_with_meson  # noqa: F401
+
+        _BUILT_WITH_MESON = True
+    except ModuleNotFoundError:
+        pass
+
+
+def setup_module(module):
+    """Fixture for the tests to assure globally controllable seeding of RNGs"""
+
+    import numpy as np
+
+    # Check if a random seed exists in the environment, if not create one.
+    _random_seed = os.environ.get("SKLEARN_SEED", None)
+    if _random_seed is None:
+        _random_seed = np.random.uniform() * np.iinfo(np.int32).max
+    _random_seed = int(_random_seed)
+    print("I: Seeding RNGs with %r" % _random_seed)
+    np.random.seed(_random_seed)
+    random.seed(_random_seed)

.venv/lib/python3.12/site-packages/sklearn/_build_utils/__init__.py

@@ -0,0 +1,116 @@
+"""
+Utilities useful during the build.
+"""
+
+# author: Andy Mueller, Gael Varoquaux
+# license: BSD
+
+
+import contextlib
+import os
+
+import sklearn
+
+from .._min_dependencies import CYTHON_MIN_VERSION
+from ..externals._packaging.version import parse
+from .openmp_helpers import check_openmp_support
+from .pre_build_helpers import basic_check_build
+
+DEFAULT_ROOT = "sklearn"
+
+
+def _check_cython_version():
+    message = (
+        "Please install Cython with a version >= {0} in order "
+        "to build a scikit-learn from source."
+    ).format(CYTHON_MIN_VERSION)
+    try:
+        import Cython
+    except ModuleNotFoundError as e:
+        # Re-raise with more informative error message instead:
+        raise ModuleNotFoundError(message) from e
+
+    if parse(Cython.__version__) < parse(CYTHON_MIN_VERSION):
+        message += " The current version of Cython is {} installed in {}.".format(
+            Cython.__version__, Cython.__path__
+        )
+        raise ValueError(message)
+
+
+def cythonize_extensions(extension):
+    """Check that a recent Cython is available and cythonize extensions"""
+    _check_cython_version()
+    from Cython.Build import cythonize
+
+    # Fast fail before cythonization if compiler fails compiling basic test
+    # code even without OpenMP
+    basic_check_build()
+
+    # check simple compilation with OpenMP. If it fails scikit-learn will be
+    # built without OpenMP and the test test_openmp_supported in the test suite
+    # will fail.
+    # `check_openmp_support` compiles a small test program to see if the
+    # compilers are properly configured to build with OpenMP. This is expensive
+    # and we only want to call this function once.
+    # The result of this check is cached as a private attribute on the sklearn
+    # module (only at build-time) to be used in the build_ext subclass defined
+    # in the top-level setup.py file to actually build the compiled extensions
+    # with OpenMP flags if needed.
+    sklearn._OPENMP_SUPPORTED = check_openmp_support()
+
+    n_jobs = 1
+    with contextlib.suppress(ImportError):
+        import joblib
+
+        n_jobs = joblib.cpu_count()
+
+    # Additional checks for Cython
+    cython_enable_debug_directives = (
+        os.environ.get("SKLEARN_ENABLE_DEBUG_CYTHON_DIRECTIVES", "0") != "0"
+    )
+
+    compiler_directives = {
+        "language_level": 3,
+        "boundscheck": cython_enable_debug_directives,
+        "wraparound": False,
+        "initializedcheck": False,
+        "nonecheck": False,
+        "cdivision": True,
+        "profile": False,
+    }
+
+    return cythonize(
+        extension,
+        nthreads=n_jobs,
+        compiler_directives=compiler_directives,
+        annotate=False,
+    )
+
+
+def gen_from_templates(templates):
+    """Generate cython files from a list of templates"""
+    # Lazy import because cython is not a runtime dependency.
+    from Cython import Tempita
+
+    for template in templates:
+        outfile = template.replace(".tp", "")
+
+        # if the template is not updated, no need to output the cython file
+        if not (
+            os.path.exists(outfile)
+            and os.stat(template).st_mtime < os.stat(outfile).st_mtime
+        ):
+            with open(template, "r") as f:
+                tmpl = f.read()
+
+            tmpl_ = Tempita.sub(tmpl)
+
+            warn_msg = (
+                "# WARNING: Do not edit this file directly.\n"
+                f"# It is automatically generated from {template!r}.\n"
+                "# Changes must be made there.\n\n"
+            )
+
+            with open(outfile, "w") as f:
+                f.write(warn_msg)
+                f.write(tmpl_)

.venv/lib/python3.12/site-packages/sklearn/_build_utils/openmp_helpers.py

@@ -0,0 +1,127 @@
+"""Helpers for OpenMP support during the build."""
+
+# This code is adapted for a large part from the astropy openmp helpers, which
+# can be found at: https://github.com/astropy/extension-helpers/blob/master/extension_helpers/_openmp_helpers.py  # noqa
+
+
+import os
+import sys
+import textwrap
+import warnings
+
+from .pre_build_helpers import compile_test_program
+
+
+def get_openmp_flag():
+    if sys.platform == "win32":
+        return ["/openmp"]
+    elif sys.platform == "darwin" and "openmp" in os.getenv("CPPFLAGS", ""):
+        # -fopenmp can't be passed as compile flag when using Apple-clang.
+        # OpenMP support has to be enabled during preprocessing.
+        #
+        # For example, our macOS wheel build jobs use the following environment
+        # variables to build with Apple-clang and the brew installed "libomp":
+        #
+        # export CPPFLAGS="$CPPFLAGS -Xpreprocessor -fopenmp"
+        # export CFLAGS="$CFLAGS -I/usr/local/opt/libomp/include"
+        # export CXXFLAGS="$CXXFLAGS -I/usr/local/opt/libomp/include"
+        # export LDFLAGS="$LDFLAGS -Wl,-rpath,/usr/local/opt/libomp/lib
+        #                          -L/usr/local/opt/libomp/lib -lomp"
+        return []
+    # Default flag for GCC and clang:
+    return ["-fopenmp"]
+
+
+def check_openmp_support():
+    """Check whether OpenMP test code can be compiled and run"""
+    if "PYODIDE" in os.environ:
+        # Pyodide doesn't support OpenMP
+        return False
+
+    code = textwrap.dedent(
+        """\
+        #include <omp.h>
+        #include <stdio.h>
+        int main(void) {
+        #pragma omp parallel
+        printf("nthreads=%d\\n", omp_get_num_threads());
+        return 0;
+        }
+        """
+    )
+
+    extra_preargs = os.getenv("LDFLAGS", None)
+    if extra_preargs is not None:
+        extra_preargs = extra_preargs.strip().split(" ")
+        # FIXME: temporary fix to link against system libraries on linux
+        # "-Wl,--sysroot=/" should be removed
+        extra_preargs = [
+            flag
+            for flag in extra_preargs
+            if flag.startswith(("-L", "-Wl,-rpath", "-l", "-Wl,--sysroot=/"))
+        ]
+
+    extra_postargs = get_openmp_flag()
+
+    openmp_exception = None
+    try:
+        output = compile_test_program(
+            code, extra_preargs=extra_preargs, extra_postargs=extra_postargs
+        )
+
+        if output and "nthreads=" in output[0]:
+            nthreads = int(output[0].strip().split("=")[1])
+            openmp_supported = len(output) == nthreads
+        elif "PYTHON_CROSSENV" in os.environ:
+            # Since we can't run the test program when cross-compiling
+            # assume that openmp is supported if the program can be
+            # compiled.
+            openmp_supported = True
+        else:
+            openmp_supported = False
+
+    except Exception as exception:
+        # We could be more specific and only catch: CompileError, LinkError,
+        # and subprocess.CalledProcessError.
+        # setuptools introduced CompileError and LinkError, but that requires
+        # version 61.1. Even the latest version of Ubuntu (22.04LTS) only
+        # ships with 59.6. So for now we catch all exceptions and reraise a
+        # generic exception with the original error message instead:
+        openmp_supported = False
+        openmp_exception = exception
+
+    if not openmp_supported:
+        if os.getenv("SKLEARN_FAIL_NO_OPENMP"):
+            raise Exception(
+                "Failed to build scikit-learn with OpenMP support"
+            ) from openmp_exception
+        else:
+            message = textwrap.dedent(
+                """
+
+                                ***********
+                                * WARNING *
+                                ***********
+
+                It seems that scikit-learn cannot be built with OpenMP.
+
+                - Make sure you have followed the installation instructions:
+
+                    https://scikit-learn.org/dev/developers/advanced_installation.html
+
+                - If your compiler supports OpenMP but you still see this
+                  message, please submit a bug report at:
+
+                    https://github.com/scikit-learn/scikit-learn/issues
+
+                - The build will continue with OpenMP-based parallelism
+                  disabled. Note however that some estimators will run in
+                  sequential mode instead of leveraging thread-based
+                  parallelism.
+
+                                    ***
+                """
+            )
+            warnings.warn(message)
+
+    return openmp_supported

.venv/lib/python3.12/site-packages/sklearn/_build_utils/pre_build_helpers.py

@@ -0,0 +1,75 @@
+"""Helpers to check build environment before actual build of scikit-learn"""
+
+import glob
+import os
+import subprocess
+import sys
+import tempfile
+import textwrap
+
+from setuptools.command.build_ext import customize_compiler, new_compiler
+
+
+def compile_test_program(code, extra_preargs=None, extra_postargs=None):
+    """Check that some C code can be compiled and run"""
+    ccompiler = new_compiler()
+    customize_compiler(ccompiler)
+
+    start_dir = os.path.abspath(".")
+
+    with tempfile.TemporaryDirectory() as tmp_dir:
+        try:
+            os.chdir(tmp_dir)
+
+            # Write test program
+            with open("test_program.c", "w") as f:
+                f.write(code)
+
+            os.mkdir("objects")
+
+            # Compile, test program
+            ccompiler.compile(
+                ["test_program.c"], output_dir="objects", extra_postargs=extra_postargs
+            )
+
+            # Link test program
+            objects = glob.glob(os.path.join("objects", "*" + ccompiler.obj_extension))
+            ccompiler.link_executable(
+                objects,
+                "test_program",
+                extra_preargs=extra_preargs,
+                extra_postargs=extra_postargs,
+            )
+
+            if "PYTHON_CROSSENV" not in os.environ:
+                # Run test program if not cross compiling
+                # will raise a CalledProcessError if return code was non-zero
+                output = subprocess.check_output("./test_program")
+                output = output.decode(sys.stdout.encoding or "utf-8").splitlines()
+            else:
+                # Return an empty output if we are cross compiling
+                # as we cannot run the test_program
+                output = []
+        except Exception:
+            raise
+        finally:
+            os.chdir(start_dir)
+
+    return output
+
+
+def basic_check_build():
+    """Check basic compilation and linking of C code"""
+    if "PYODIDE" in os.environ:
+        # The following check won't work in pyodide
+        return
+
+    code = textwrap.dedent(
+        """\
+        #include <stdio.h>
+        int main(void) {
+        return 0;
+        }
+        """
+    )
+    compile_test_program(code)

.venv/lib/python3.12/site-packages/sklearn/_build_utils/tempita.py

@@ -0,0 +1,57 @@
+import argparse
+import os
+
+from Cython import Tempita as tempita
+
+# XXX: If this import ever fails (does it really?), vendor either
+# cython.tempita or numpy/npy_tempita.
+
+
+def process_tempita(fromfile, outfile=None):
+    """Process tempita templated file and write out the result.
+
+    The template file is expected to end in `.c.tp` or `.pyx.tp`:
+    E.g. processing `template.c.in` generates `template.c`.
+
+    """
+    with open(fromfile, "r", encoding="utf-8") as f:
+        template_content = f.read()
+
+    template = tempita.Template(template_content)
+    content = template.substitute()
+
+    with open(outfile, "w", encoding="utf-8") as f:
+        f.write(content)
+
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("infile", type=str, help="Path to the input file")
+    parser.add_argument("-o", "--outdir", type=str, help="Path to the output directory")
+    parser.add_argument(
+        "-i",
+        "--ignore",
+        type=str,
+        help=(
+            "An ignored input - may be useful to add a "
+            "dependency between custom targets"
+        ),
+    )
+    args = parser.parse_args()
+
+    if not args.infile.endswith(".tp"):
+        raise ValueError(f"Unexpected extension: {args.infile}")
+
+    if not args.outdir:
+        raise ValueError("Missing `--outdir` argument to tempita.py")
+
+    outdir_abs = os.path.join(os.getcwd(), args.outdir)
+    outfile = os.path.join(
+        outdir_abs, os.path.splitext(os.path.split(args.infile)[1])[0]
+    )
+
+    process_tempita(args.infile, outfile)
+
+
+if __name__ == "__main__":
+    main()

.venv/lib/python3.12/site-packages/sklearn/_build_utils/version.py

@@ -0,0 +1,13 @@
+#!/usr/bin/env python3
+"""Extract version number from __init__.py"""
+
+import os
+
+sklearn_init = os.path.join(os.path.dirname(__file__), "../__init__.py")
+
+data = open(sklearn_init).readlines()
+version_line = next(line for line in data if line.startswith("__version__"))
+
+version = version_line.strip().split(" = ")[1].replace('"', "").replace("'", "")
+
+print(version)

.venv/lib/python3.12/site-packages/sklearn/_config.py

@@ -0,0 +1,373 @@
+"""Global configuration state and functions for management"""
+
+import os
+import threading
+from contextlib import contextmanager as contextmanager
+
+_global_config = {
+    "assume_finite": bool(os.environ.get("SKLEARN_ASSUME_FINITE", False)),
+    "working_memory": int(os.environ.get("SKLEARN_WORKING_MEMORY", 1024)),
+    "print_changed_only": True,
+    "display": "diagram",
+    "pairwise_dist_chunk_size": int(
+        os.environ.get("SKLEARN_PAIRWISE_DIST_CHUNK_SIZE", 256)
+    ),
+    "enable_cython_pairwise_dist": True,
+    "array_api_dispatch": False,
+    "transform_output": "default",
+    "enable_metadata_routing": False,
+    "skip_parameter_validation": False,
+}
+_threadlocal = threading.local()
+
+
+def _get_threadlocal_config():
+    """Get a threadlocal **mutable** configuration. If the configuration
+    does not exist, copy the default global configuration."""
+    if not hasattr(_threadlocal, "global_config"):
+        _threadlocal.global_config = _global_config.copy()
+    return _threadlocal.global_config
+
+
+def get_config():
+    """Retrieve current values for configuration set by :func:`set_config`.
+
+    Returns
+    -------
+    config : dict
+        Keys are parameter names that can be passed to :func:`set_config`.
+
+    See Also
+    --------
+    config_context : Context manager for global scikit-learn configuration.
+    set_config : Set global scikit-learn configuration.
+
+    Examples
+    --------
+    >>> import sklearn
+    >>> config = sklearn.get_config()
+    >>> config.keys()
+    dict_keys([...])
+    """
+    # Return a copy of the threadlocal configuration so that users will
+    # not be able to modify the configuration with the returned dict.
+    return _get_threadlocal_config().copy()
+
+
+def set_config(
+    assume_finite=None,
+    working_memory=None,
+    print_changed_only=None,
+    display=None,
+    pairwise_dist_chunk_size=None,
+    enable_cython_pairwise_dist=None,
+    array_api_dispatch=None,
+    transform_output=None,
+    enable_metadata_routing=None,
+    skip_parameter_validation=None,
+):
+    """Set global scikit-learn configuration.
+
+    .. versionadded:: 0.19
+
+    Parameters
+    ----------
+    assume_finite : bool, default=None
+        If True, validation for finiteness will be skipped,
+        saving time, but leading to potential crashes. If
+        False, validation for finiteness will be performed,
+        avoiding error.  Global default: False.
+
+        .. versionadded:: 0.19
+
+    working_memory : int, default=None
+        If set, scikit-learn will attempt to limit the size of temporary arrays
+        to this number of MiB (per job when parallelised), often saving both
+        computation time and memory on expensive operations that can be
+        performed in chunks. Global default: 1024.
+
+        .. versionadded:: 0.20
+
+    print_changed_only : bool, default=None
+        If True, only the parameters that were set to non-default
+        values will be printed when printing an estimator. For example,
+        ``print(SVC())`` while True will only print 'SVC()' while the default
+        behaviour would be to print 'SVC(C=1.0, cache_size=200, ...)' with
+        all the non-changed parameters.
+
+        .. versionadded:: 0.21
+
+    display : {'text', 'diagram'}, default=None
+        If 'diagram', estimators will be displayed as a diagram in a Jupyter
+        lab or notebook context. If 'text', estimators will be displayed as
+        text. Default is 'diagram'.
+
+        .. versionadded:: 0.23
+
+    pairwise_dist_chunk_size : int, default=None
+        The number of row vectors per chunk for the accelerated pairwise-
+        distances reduction backend. Default is 256 (suitable for most of
+        modern laptops' caches and architectures).
+
+        Intended for easier benchmarking and testing of scikit-learn internals.
+        End users are not expected to benefit from customizing this configuration
+        setting.
+
+        .. versionadded:: 1.1
+
+    enable_cython_pairwise_dist : bool, default=None
+        Use the accelerated pairwise-distances reduction backend when
+        possible. Global default: True.
+
+        Intended for easier benchmarking and testing of scikit-learn internals.
+        End users are not expected to benefit from customizing this configuration
+        setting.
+
+        .. versionadded:: 1.1
+
+    array_api_dispatch : bool, default=None
+        Use Array API dispatching when inputs follow the Array API standard.
+        Default is False.
+
+        See the :ref:`User Guide <array_api>` for more details.
+
+        .. versionadded:: 1.2
+
+    transform_output : str, default=None
+        Configure output of `transform` and `fit_transform`.
+
+        See :ref:`sphx_glr_auto_examples_miscellaneous_plot_set_output.py`
+        for an example on how to use the API.
+
+        - `"default"`: Default output format of a transformer
+        - `"pandas"`: DataFrame output
+        - `"polars"`: Polars output
+        - `None`: Transform configuration is unchanged
+
+        .. versionadded:: 1.2
+        .. versionadded:: 1.4
+            `"polars"` option was added.
+
+    enable_metadata_routing : bool, default=None
+        Enable metadata routing. By default this feature is disabled.
+
+        Refer to :ref:`metadata routing user guide <metadata_routing>` for more
+        details.
+
+        - `True`: Metadata routing is enabled
+        - `False`: Metadata routing is disabled, use the old syntax.
+        - `None`: Configuration is unchanged
+
+        .. versionadded:: 1.3
+
+    skip_parameter_validation : bool, default=None
+        If `True`, disable the validation of the hyper-parameters' types and values in
+        the fit method of estimators and for arguments passed to public helper
+        functions. It can save time in some situations but can lead to low level
+        crashes and exceptions with confusing error messages.
+
+        Note that for data parameters, such as `X` and `y`, only type validation is
+        skipped but validation with `check_array` will continue to run.
+
+        .. versionadded:: 1.3
+
+    See Also
+    --------
+    config_context : Context manager for global scikit-learn configuration.
+    get_config : Retrieve current values of the global configuration.
+
+    Examples
+    --------
+    >>> from sklearn import set_config
+    >>> set_config(display='diagram')  # doctest: +SKIP
+    """
+    local_config = _get_threadlocal_config()
+
+    if assume_finite is not None:
+        local_config["assume_finite"] = assume_finite
+    if working_memory is not None:
+        local_config["working_memory"] = working_memory
+    if print_changed_only is not None:
+        local_config["print_changed_only"] = print_changed_only
+    if display is not None:
+        local_config["display"] = display
+    if pairwise_dist_chunk_size is not None:
+        local_config["pairwise_dist_chunk_size"] = pairwise_dist_chunk_size
+    if enable_cython_pairwise_dist is not None:
+        local_config["enable_cython_pairwise_dist"] = enable_cython_pairwise_dist
+    if array_api_dispatch is not None:
+        from .utils._array_api import _check_array_api_dispatch
+
+        _check_array_api_dispatch(array_api_dispatch)
+        local_config["array_api_dispatch"] = array_api_dispatch
+    if transform_output is not None:
+        local_config["transform_output"] = transform_output
+    if enable_metadata_routing is not None:
+        local_config["enable_metadata_routing"] = enable_metadata_routing
+    if skip_parameter_validation is not None:
+        local_config["skip_parameter_validation"] = skip_parameter_validation
+
+
+@contextmanager
+def config_context(
+    *,
+    assume_finite=None,
+    working_memory=None,
+    print_changed_only=None,
+    display=None,
+    pairwise_dist_chunk_size=None,
+    enable_cython_pairwise_dist=None,
+    array_api_dispatch=None,
+    transform_output=None,
+    enable_metadata_routing=None,
+    skip_parameter_validation=None,
+):
+    """Context manager for global scikit-learn configuration.
+
+    Parameters
+    ----------
+    assume_finite : bool, default=None
+        If True, validation for finiteness will be skipped,
+        saving time, but leading to potential crashes. If
+        False, validation for finiteness will be performed,
+        avoiding error. If None, the existing value won't change.
+        The default value is False.
+
+    working_memory : int, default=None
+        If set, scikit-learn will attempt to limit the size of temporary arrays
+        to this number of MiB (per job when parallelised), often saving both
+        computation time and memory on expensive operations that can be
+        performed in chunks. If None, the existing value won't change.
+        The default value is 1024.
+
+    print_changed_only : bool, default=None
+        If True, only the parameters that were set to non-default
+        values will be printed when printing an estimator. For example,
+        ``print(SVC())`` while True will only print 'SVC()', but would print
+        'SVC(C=1.0, cache_size=200, ...)' with all the non-changed parameters
+        when False. If None, the existing value won't change.
+        The default value is True.
+
+        .. versionchanged:: 0.23
+           Default changed from False to True.
+
+    display : {'text', 'diagram'}, default=None
+        If 'diagram', estimators will be displayed as a diagram in a Jupyter
+        lab or notebook context. If 'text', estimators will be displayed as
+        text. If None, the existing value won't change.
+        The default value is 'diagram'.
+
+        .. versionadded:: 0.23
+
+    pairwise_dist_chunk_size : int, default=None
+        The number of row vectors per chunk for the accelerated pairwise-
+        distances reduction backend. Default is 256 (suitable for most of
+        modern laptops' caches and architectures).
+
+        Intended for easier benchmarking and testing of scikit-learn internals.
+        End users are not expected to benefit from customizing this configuration
+        setting.
+
+        .. versionadded:: 1.1
+
+    enable_cython_pairwise_dist : bool, default=None
+        Use the accelerated pairwise-distances reduction backend when
+        possible. Global default: True.
+
+        Intended for easier benchmarking and testing of scikit-learn internals.
+        End users are not expected to benefit from customizing this configuration
+        setting.
+
+        .. versionadded:: 1.1
+
+    array_api_dispatch : bool, default=None
+        Use Array API dispatching when inputs follow the Array API standard.
+        Default is False.
+
+        See the :ref:`User Guide <array_api>` for more details.
+
+        .. versionadded:: 1.2
+
+    transform_output : str, default=None
+        Configure output of `transform` and `fit_transform`.
+
+        See :ref:`sphx_glr_auto_examples_miscellaneous_plot_set_output.py`
+        for an example on how to use the API.
+
+        - `"default"`: Default output format of a transformer
+        - `"pandas"`: DataFrame output
+        - `"polars"`: Polars output
+        - `None`: Transform configuration is unchanged
+
+        .. versionadded:: 1.2
+        .. versionadded:: 1.4
+            `"polars"` option was added.
+
+    enable_metadata_routing : bool, default=None
+        Enable metadata routing. By default this feature is disabled.
+
+        Refer to :ref:`metadata routing user guide <metadata_routing>` for more
+        details.
+
+        - `True`: Metadata routing is enabled
+        - `False`: Metadata routing is disabled, use the old syntax.
+        - `None`: Configuration is unchanged
+
+        .. versionadded:: 1.3
+
+    skip_parameter_validation : bool, default=None
+        If `True`, disable the validation of the hyper-parameters' types and values in
+        the fit method of estimators and for arguments passed to public helper
+        functions. It can save time in some situations but can lead to low level
+        crashes and exceptions with confusing error messages.
+
+        Note that for data parameters, such as `X` and `y`, only type validation is
+        skipped but validation with `check_array` will continue to run.
+
+        .. versionadded:: 1.3
+
+    Yields
+    ------
+    None.
+
+    See Also
+    --------
+    set_config : Set global scikit-learn configuration.
+    get_config : Retrieve current values of the global configuration.
+
+    Notes
+    -----
+    All settings, not just those presently modified, will be returned to
+    their previous values when the context manager is exited.
+
+    Examples
+    --------
+    >>> import sklearn
+    >>> from sklearn.utils.validation import assert_all_finite
+    >>> with sklearn.config_context(assume_finite=True):
+    ...     assert_all_finite([float('nan')])
+    >>> with sklearn.config_context(assume_finite=True):
+    ...     with sklearn.config_context(assume_finite=False):
+    ...         assert_all_finite([float('nan')])
+    Traceback (most recent call last):
+    ...
+    ValueError: Input contains NaN...
+    """
+    old_config = get_config()
+    set_config(
+        assume_finite=assume_finite,
+        working_memory=working_memory,
+        print_changed_only=print_changed_only,
+        display=display,
+        pairwise_dist_chunk_size=pairwise_dist_chunk_size,
+        enable_cython_pairwise_dist=enable_cython_pairwise_dist,
+        array_api_dispatch=array_api_dispatch,
+        transform_output=transform_output,
+        enable_metadata_routing=enable_metadata_routing,
+        skip_parameter_validation=skip_parameter_validation,
+    )
+
+    try:
+        yield
+    finally:
+        set_config(**old_config)

.venv/lib/python3.12/site-packages/sklearn/_distributor_init.py

@@ -0,0 +1,10 @@
+"""Distributor init file
+
+Distributors: you can add custom code here to support particular distributions
+of scikit-learn.
+
+For example, this is a good place to put any checks for hardware requirements.
+
+The scikit-learn standard source distribution will not put code in this file,
+so you can safely replace this file with your own version.
+"""

.venv/lib/python3.12/site-packages/sklearn/_isotonic.pyx

@@ -0,0 +1,115 @@
+# Author: Nelle Varoquaux, Andrew Tulloch, Antony Lee
+
+# Uses the pool adjacent violators algorithm (PAVA), with the
+# enhancement of searching for the longest decreasing subsequence to
+# pool at each step.
+
+import numpy as np
+from cython cimport floating
+
+
+def _inplace_contiguous_isotonic_regression(floating[::1] y, floating[::1] w):
+    cdef:
+        Py_ssize_t n = y.shape[0], i, k
+        floating prev_y, sum_wy, sum_w
+        Py_ssize_t[::1] target = np.arange(n, dtype=np.intp)
+
+    # target describes a list of blocks.  At any time, if [i..j] (inclusive) is
+    # an active block, then target[i] := j and target[j] := i.
+
+    # For "active" indices (block starts):
+    # w[i] := sum{w_orig[j], j=[i..target[i]]}
+    # y[i] := sum{y_orig[j]*w_orig[j], j=[i..target[i]]} / w[i]
+
+    with nogil:
+        i = 0
+        while i < n:
+            k = target[i] + 1
+            if k == n:
+                break
+            if y[i] < y[k]:
+                i = k
+                continue
+            sum_wy = w[i] * y[i]
+            sum_w = w[i]
+            while True:
+                # We are within a decreasing subsequence.
+                prev_y = y[k]
+                sum_wy += w[k] * y[k]
+                sum_w += w[k]
+                k = target[k] + 1
+                if k == n or prev_y < y[k]:
+                    # Non-singleton decreasing subsequence is finished,
+                    # update first entry.
+                    y[i] = sum_wy / sum_w
+                    w[i] = sum_w
+                    target[i] = k - 1
+                    target[k - 1] = i
+                    if i > 0:
+                        # Backtrack if we can.  This makes the algorithm
+                        # single-pass and ensures O(n) complexity.
+                        i = target[i - 1]
+                    # Otherwise, restart from the same point.
+                    break
+        # Reconstruct the solution.
+        i = 0
+        while i < n:
+            k = target[i] + 1
+            y[i + 1 : k] = y[i]
+            i = k
+
+
+def _make_unique(const floating[::1] X,
+                 const floating[::1] y,
+                 const floating[::1] sample_weights):
+    """Average targets for duplicate X, drop duplicates.
+
+    Aggregates duplicate X values into a single X value where
+    the target y is a (sample_weighted) average of the individual
+    targets.
+
+    Assumes that X is ordered, so that all duplicates follow each other.
+    """
+    unique_values = len(np.unique(X))
+
+    if floating is float:
+        dtype = np.float32
+    else:
+        dtype = np.float64
+
+    cdef floating[::1] y_out = np.empty(unique_values, dtype=dtype)
+    cdef floating[::1] x_out = np.empty_like(y_out)
+    cdef floating[::1] weights_out = np.empty_like(y_out)
+
+    cdef floating current_x = X[0]
+    cdef floating current_y = 0
+    cdef floating current_weight = 0
+    cdef int i = 0
+    cdef int j
+    cdef floating x
+    cdef int n_samples = len(X)
+    cdef floating eps = np.finfo(dtype).resolution
+
+    for j in range(n_samples):
+        x = X[j]
+        if x - current_x >= eps:
+            # next unique value
+            x_out[i] = current_x
+            weights_out[i] = current_weight
+            y_out[i] = current_y / current_weight
+            i += 1
+            current_x = x
+            current_weight = sample_weights[j]
+            current_y = y[j] * sample_weights[j]
+        else:
+            current_weight += sample_weights[j]
+            current_y += y[j] * sample_weights[j]
+
+    x_out[i] = current_x
+    weights_out[i] = current_weight
+    y_out[i] = current_y / current_weight
+    return(
+        np.asarray(x_out[:i+1]),
+        np.asarray(y_out[:i+1]),
+        np.asarray(weights_out[:i+1]),
+    )

.venv/lib/python3.12/site-packages/sklearn/_loss/__init__.py

@@ -0,0 +1,30 @@
+"""
+The :mod:`sklearn._loss` module includes loss function classes suitable for
+fitting classification and regression tasks.
+"""
+
+from .loss import (
+    AbsoluteError,
+    HalfBinomialLoss,
+    HalfGammaLoss,
+    HalfMultinomialLoss,
+    HalfPoissonLoss,
+    HalfSquaredError,
+    HalfTweedieLoss,
+    HalfTweedieLossIdentity,
+    HuberLoss,
+    PinballLoss,
+)
+
+__all__ = [
+    "HalfSquaredError",
+    "AbsoluteError",
+    "PinballLoss",
+    "HuberLoss",
+    "HalfPoissonLoss",
+    "HalfGammaLoss",
+    "HalfTweedieLoss",
+    "HalfTweedieLossIdentity",
+    "HalfBinomialLoss",
+    "HalfMultinomialLoss",
+]

.venv/lib/python3.12/site-packages/sklearn/_loss/_loss.pxd

@@ -0,0 +1,91 @@
+# Fused types for input like y_true, raw_prediction, sample_weights.
+ctypedef fused floating_in:
+    double
+    float
+
+
+# Fused types for output like gradient and hessian
+# We use a different fused types for input (floating_in) and output (floating_out), such
+# that input and output can have different dtypes in the same function call. A single
+# fused type can only take on one single value (type) for all arguments in one function
+# call.
+ctypedef fused floating_out:
+    double
+    float
+
+
+# Struct to return 2 doubles
+ctypedef struct double_pair:
+    double val1
+    double val2
+
+
+# C base class for loss functions
+cdef class CyLossFunction:
+    cdef double cy_loss(self, double y_true, double raw_prediction) noexcept nogil
+    cdef double cy_gradient(self, double y_true, double raw_prediction) noexcept nogil
+    cdef double_pair cy_grad_hess(self, double y_true, double raw_prediction) noexcept nogil
+
+
+cdef class CyHalfSquaredError(CyLossFunction):
+    cdef double cy_loss(self, double y_true, double raw_prediction) noexcept nogil
+    cdef double cy_gradient(self, double y_true, double raw_prediction) noexcept nogil
+    cdef double_pair cy_grad_hess(self, double y_true, double raw_prediction) noexcept nogil
+
+
+cdef class CyAbsoluteError(CyLossFunction):
+    cdef double cy_loss(self, double y_true, double raw_prediction) noexcept nogil
+    cdef double cy_gradient(self, double y_true, double raw_prediction) noexcept nogil
+    cdef double_pair cy_grad_hess(self, double y_true, double raw_prediction) noexcept nogil
+
+
+cdef class CyPinballLoss(CyLossFunction):
+    cdef readonly double quantile  # readonly makes it accessible from Python
+    cdef double cy_loss(self, double y_true, double raw_prediction) noexcept nogil
+    cdef double cy_gradient(self, double y_true, double raw_prediction) noexcept nogil
+    cdef double_pair cy_grad_hess(self, double y_true, double raw_prediction) noexcept nogil
+
+
+cdef class CyHuberLoss(CyLossFunction):
+    cdef public double delta  # public makes it accessible from Python
+    cdef double cy_loss(self, double y_true, double raw_prediction) noexcept nogil
+    cdef double cy_gradient(self, double y_true, double raw_prediction) noexcept nogil
+    cdef double_pair cy_grad_hess(self, double y_true, double raw_prediction) noexcept nogil
+
+
+cdef class CyHalfPoissonLoss(CyLossFunction):
+    cdef double cy_loss(self, double y_true, double raw_prediction) noexcept nogil
+    cdef double cy_gradient(self, double y_true, double raw_prediction) noexcept nogil
+    cdef double_pair cy_grad_hess(self, double y_true, double raw_prediction) noexcept nogil
+
+
+cdef class CyHalfGammaLoss(CyLossFunction):
+    cdef double cy_loss(self, double y_true, double raw_prediction) noexcept nogil
+    cdef double cy_gradient(self, double y_true, double raw_prediction) noexcept nogil
+    cdef double_pair cy_grad_hess(self, double y_true, double raw_prediction) noexcept nogil
+
+
+cdef class CyHalfTweedieLoss(CyLossFunction):
+    cdef readonly double power  # readonly makes it accessible from Python
+    cdef double cy_loss(self, double y_true, double raw_prediction) noexcept nogil
+    cdef double cy_gradient(self, double y_true, double raw_prediction) noexcept nogil
+    cdef double_pair cy_grad_hess(self, double y_true, double raw_prediction) noexcept nogil
+
+
+cdef class CyHalfTweedieLossIdentity(CyLossFunction):
+    cdef readonly double power  # readonly makes it accessible from Python
+    cdef double cy_loss(self, double y_true, double raw_prediction) noexcept nogil
+    cdef double cy_gradient(self, double y_true, double raw_prediction) noexcept nogil
+    cdef double_pair cy_grad_hess(self, double y_true, double raw_prediction) noexcept nogil
+
+
+cdef class CyHalfBinomialLoss(CyLossFunction):
+    cdef double cy_loss(self, double y_true, double raw_prediction) noexcept nogil
+    cdef double cy_gradient(self, double y_true, double raw_prediction) noexcept nogil
+    cdef double_pair cy_grad_hess(self, double y_true, double raw_prediction) noexcept nogil
+
+
+cdef class CyExponentialLoss(CyLossFunction):
+    cdef double cy_loss(self, double y_true, double raw_prediction) noexcept nogil
+    cdef double cy_gradient(self, double y_true, double raw_prediction) noexcept nogil
+    cdef double_pair cy_grad_hess(self, double y_true, double raw_prediction) noexcept nogil

.venv/lib/python3.12/site-packages/sklearn/_loss/link.py

@@ -0,0 +1,281 @@
+"""
+Module contains classes for invertible (and differentiable) link functions.
+"""
+
+# Author: Christian Lorentzen <lorentzen.ch@gmail.com>
+
+from abc import ABC, abstractmethod
+from dataclasses import dataclass
+
+import numpy as np
+from scipy.special import expit, logit
+from scipy.stats import gmean
+
+from ..utils.extmath import softmax
+
+
+@dataclass
+class Interval:
+    low: float
+    high: float
+    low_inclusive: bool
+    high_inclusive: bool
+
+    def __post_init__(self):
+        """Check that low <= high"""
+        if self.low > self.high:
+            raise ValueError(
+                f"One must have low <= high; got low={self.low}, high={self.high}."
+            )
+
+    def includes(self, x):
+        """Test whether all values of x are in interval range.
+
+        Parameters
+        ----------
+        x : ndarray
+            Array whose elements are tested to be in interval range.
+
+        Returns
+        -------
+        result : bool
+        """
+        if self.low_inclusive:
+            low = np.greater_equal(x, self.low)
+        else:
+            low = np.greater(x, self.low)
+
+        if not np.all(low):
+            return False
+
+        if self.high_inclusive:
+            high = np.less_equal(x, self.high)
+        else:
+            high = np.less(x, self.high)
+
+        # Note: np.all returns numpy.bool_
+        return bool(np.all(high))
+
+
+def _inclusive_low_high(interval, dtype=np.float64):
+    """Generate values low and high to be within the interval range.
+
+    This is used in tests only.
+
+    Returns
+    -------
+    low, high : tuple
+        The returned values low and high lie within the interval.
+    """
+    eps = 10 * np.finfo(dtype).eps
+    if interval.low == -np.inf:
+        low = -1e10
+    elif interval.low < 0:
+        low = interval.low * (1 - eps) + eps
+    else:
+        low = interval.low * (1 + eps) + eps
+
+    if interval.high == np.inf:
+        high = 1e10
+    elif interval.high < 0:
+        high = interval.high * (1 + eps) - eps
+    else:
+        high = interval.high * (1 - eps) - eps
+
+    return low, high
+
+
+class BaseLink(ABC):
+    """Abstract base class for differentiable, invertible link functions.
+
+    Convention:
+        - link function g: raw_prediction = g(y_pred)
+        - inverse link h: y_pred = h(raw_prediction)
+
+    For (generalized) linear models, `raw_prediction = X @ coef` is the so
+    called linear predictor, and `y_pred = h(raw_prediction)` is the predicted
+    conditional (on X) expected value of the target `y_true`.
+
+    The methods are not implemented as staticmethods in case a link function needs
+    parameters.
+    """
+
+    is_multiclass = False  # used for testing only
+
+    # Usually, raw_prediction may be any real number and y_pred is an open
+    # interval.
+    # interval_raw_prediction = Interval(-np.inf, np.inf, False, False)
+    interval_y_pred = Interval(-np.inf, np.inf, False, False)
+
+    @abstractmethod
+    def link(self, y_pred, out=None):
+        """Compute the link function g(y_pred).
+
+        The link function maps (predicted) target values to raw predictions,
+        i.e. `g(y_pred) = raw_prediction`.
+
+        Parameters
+        ----------
+        y_pred : array
+            Predicted target values.
+        out : array
+            A location into which the result is stored. If provided, it must
+            have a shape that the inputs broadcast to. If not provided or None,
+            a freshly-allocated array is returned.
+
+        Returns
+        -------
+        out : array
+            Output array, element-wise link function.
+        """
+
+    @abstractmethod
+    def inverse(self, raw_prediction, out=None):
+        """Compute the inverse link function h(raw_prediction).
+
+        The inverse link function maps raw predictions to predicted target
+        values, i.e. `h(raw_prediction) = y_pred`.
+
+        Parameters
+        ----------
+        raw_prediction : array
+            Raw prediction values (in link space).
+        out : array
+            A location into which the result is stored. If provided, it must
+            have a shape that the inputs broadcast to. If not provided or None,
+            a freshly-allocated array is returned.
+
+        Returns
+        -------
+        out : array
+            Output array, element-wise inverse link function.
+        """
+
+
+class IdentityLink(BaseLink):
+    """The identity link function g(x)=x."""
+
+    def link(self, y_pred, out=None):
+        if out is not None:
+            np.copyto(out, y_pred)
+            return out
+        else:
+            return y_pred
+
+    inverse = link
+
+
+class LogLink(BaseLink):
+    """The log link function g(x)=log(x)."""
+
+    interval_y_pred = Interval(0, np.inf, False, False)
+
+    def link(self, y_pred, out=None):
+        return np.log(y_pred, out=out)
+
+    def inverse(self, raw_prediction, out=None):
+        return np.exp(raw_prediction, out=out)
+
+
+class LogitLink(BaseLink):
+    """The logit link function g(x)=logit(x)."""
+
+    interval_y_pred = Interval(0, 1, False, False)
+
+    def link(self, y_pred, out=None):
+        return logit(y_pred, out=out)
+
+    def inverse(self, raw_prediction, out=None):
+        return expit(raw_prediction, out=out)
+
+
+class HalfLogitLink(BaseLink):
+    """Half the logit link function g(x)=1/2 * logit(x).
+
+    Used for the exponential loss.
+    """
+
+    interval_y_pred = Interval(0, 1, False, False)
+
+    def link(self, y_pred, out=None):
+        out = logit(y_pred, out=out)
+        out *= 0.5
+        return out
+
+    def inverse(self, raw_prediction, out=None):
+        return expit(2 * raw_prediction, out)
+
+
+class MultinomialLogit(BaseLink):
+    """The symmetric multinomial logit function.
+
+    Convention:
+        - y_pred.shape = raw_prediction.shape = (n_samples, n_classes)
+
+    Notes:
+        - The inverse link h is the softmax function.
+        - The sum is over the second axis, i.e. axis=1 (n_classes).
+
+    We have to choose additional constraints in order to make
+
+        y_pred[k] = exp(raw_pred[k]) / sum(exp(raw_pred[k]), k=0..n_classes-1)
+
+    for n_classes classes identifiable and invertible.
+    We choose the symmetric side constraint where the geometric mean response
+    is set as reference category, see [2]:
+
+    The symmetric multinomial logit link function for a single data point is
+    then defined as
+
+        raw_prediction[k] = g(y_pred[k]) = log(y_pred[k]/gmean(y_pred))
+        = log(y_pred[k]) - mean(log(y_pred)).
+
+    Note that this is equivalent to the definition in [1] and implies mean
+    centered raw predictions:
+
+        sum(raw_prediction[k], k=0..n_classes-1) = 0.
+
+    For linear models with raw_prediction = X @ coef, this corresponds to
+    sum(coef[k], k=0..n_classes-1) = 0, i.e. the sum over classes for every
+    feature is zero.
+
+    Reference
+    ---------
+    .. [1] Friedman, Jerome; Hastie, Trevor; Tibshirani, Robert. "Additive
+        logistic regression: a statistical view of boosting" Ann. Statist.
+        28 (2000), no. 2, 337--407. doi:10.1214/aos/1016218223.
+        https://projecteuclid.org/euclid.aos/1016218223
+
+    .. [2] Zahid, Faisal Maqbool and Gerhard Tutz. "Ridge estimation for
+        multinomial logit models with symmetric side constraints."
+        Computational Statistics 28 (2013): 1017-1034.
+        http://epub.ub.uni-muenchen.de/11001/1/tr067.pdf
+    """
+
+    is_multiclass = True
+    interval_y_pred = Interval(0, 1, False, False)
+
+    def symmetrize_raw_prediction(self, raw_prediction):
+        return raw_prediction - np.mean(raw_prediction, axis=1)[:, np.newaxis]
+
+    def link(self, y_pred, out=None):
+        # geometric mean as reference category
+        gm = gmean(y_pred, axis=1)
+        return np.log(y_pred / gm[:, np.newaxis], out=out)
+
+    def inverse(self, raw_prediction, out=None):
+        if out is None:
+            return softmax(raw_prediction, copy=True)
+        else:
+            np.copyto(out, raw_prediction)
+            softmax(out, copy=False)
+            return out
+
+
+_LINKS = {
+    "identity": IdentityLink,
+    "log": LogLink,
+    "logit": LogitLink,
+    "half_logit": HalfLogitLink,
+    "multinomial_logit": MultinomialLogit,
+}

.venv/lib/python3.12/site-packages/sklearn/_loss/meson.build

@@ -0,0 +1,24 @@
+# .pyx is generated, so this is needed to make Cython compilation work
+_loss_cython_tree = [
+  fs.copyfile('_loss.pxd')
+]
+
+_loss_pyx = custom_target(
+  '_loss_pyx',
+  output: '_loss.pyx',
+  input: '_loss.pyx.tp',
+  command: [py, tempita, '@INPUT@', '-o', '@OUTDIR@'],
+  # TODO in principle this should go in py.exension_module below. This is
+  # temporary work-around for dependency issue with .pyx.tp files. For more
+  # details, see https://github.com/mesonbuild/meson/issues/13212
+  depends: _loss_cython_tree,
+)
+
+py.extension_module(
+  '_loss',
+  _loss_pyx,
+  dependencies: [openmp_dep],
+  cython_args: cython_args,
+  install: true,
+  subdir: 'sklearn/_loss',
+)

.venv/lib/python3.12/site-packages/sklearn/_loss/tests/test_link.py

@@ -0,0 +1,111 @@
+import numpy as np
+import pytest
+from numpy.testing import assert_allclose, assert_array_equal
+
+from sklearn._loss.link import (
+    _LINKS,
+    HalfLogitLink,
+    Interval,
+    MultinomialLogit,
+    _inclusive_low_high,
+)
+
+LINK_FUNCTIONS = list(_LINKS.values())
+
+
+def test_interval_raises():
+    """Test that interval with low > high raises ValueError."""
+    with pytest.raises(
+        ValueError, match="One must have low <= high; got low=1, high=0."
+    ):
+        Interval(1, 0, False, False)
+
+
+@pytest.mark.parametrize(
+    "interval",
+    [
+        Interval(0, 1, False, False),
+        Interval(0, 1, False, True),
+        Interval(0, 1, True, False),
+        Interval(0, 1, True, True),
+        Interval(-np.inf, np.inf, False, False),
+        Interval(-np.inf, np.inf, False, True),
+        Interval(-np.inf, np.inf, True, False),
+        Interval(-np.inf, np.inf, True, True),
+        Interval(-10, -1, False, False),
+        Interval(-10, -1, False, True),
+        Interval(-10, -1, True, False),
+        Interval(-10, -1, True, True),
+    ],
+)
+def test_is_in_range(interval):
+    # make sure low and high are always within the interval, used for linspace
+    low, high = _inclusive_low_high(interval)
+
+    x = np.linspace(low, high, num=10)
+    assert interval.includes(x)
+
+    # x contains lower bound
+    assert interval.includes(np.r_[x, interval.low]) == interval.low_inclusive
+
+    # x contains upper bound
+    assert interval.includes(np.r_[x, interval.high]) == interval.high_inclusive
+
+    # x contains upper and lower bound
+    assert interval.includes(np.r_[x, interval.low, interval.high]) == (
+        interval.low_inclusive and interval.high_inclusive
+    )
+
+
+@pytest.mark.parametrize("link", LINK_FUNCTIONS)
+def test_link_inverse_identity(link, global_random_seed):
+    # Test that link of inverse gives identity.
+    rng = np.random.RandomState(global_random_seed)
+    link = link()
+    n_samples, n_classes = 100, None
+    # The values for `raw_prediction` are limited from -20 to 20 because in the
+    # class `LogitLink` the term `expit(x)` comes very close to 1 for large
+    # positive x and therefore loses precision.
+    if link.is_multiclass:
+        n_classes = 10
+        raw_prediction = rng.uniform(low=-20, high=20, size=(n_samples, n_classes))
+        if isinstance(link, MultinomialLogit):
+            raw_prediction = link.symmetrize_raw_prediction(raw_prediction)
+    elif isinstance(link, HalfLogitLink):
+        raw_prediction = rng.uniform(low=-10, high=10, size=(n_samples))
+    else:
+        raw_prediction = rng.uniform(low=-20, high=20, size=(n_samples))
+
+    assert_allclose(link.link(link.inverse(raw_prediction)), raw_prediction)
+    y_pred = link.inverse(raw_prediction)
+    assert_allclose(link.inverse(link.link(y_pred)), y_pred)
+
+
+@pytest.mark.parametrize("link", LINK_FUNCTIONS)
+def test_link_out_argument(link):
+    # Test that out argument gets assigned the result.
+    rng = np.random.RandomState(42)
+    link = link()
+    n_samples, n_classes = 100, None
+    if link.is_multiclass:
+        n_classes = 10
+        raw_prediction = rng.normal(loc=0, scale=10, size=(n_samples, n_classes))
+        if isinstance(link, MultinomialLogit):
+            raw_prediction = link.symmetrize_raw_prediction(raw_prediction)
+    else:
+        # So far, the valid interval of raw_prediction is (-inf, inf) and
+        # we do not need to distinguish.
+        raw_prediction = rng.uniform(low=-10, high=10, size=(n_samples))
+
+    y_pred = link.inverse(raw_prediction, out=None)
+    out = np.empty_like(raw_prediction)
+    y_pred_2 = link.inverse(raw_prediction, out=out)
+    assert_allclose(y_pred, out)
+    assert_array_equal(out, y_pred_2)
+    assert np.shares_memory(out, y_pred_2)
+
+    out = np.empty_like(y_pred)
+    raw_prediction_2 = link.link(y_pred, out=out)
+    assert_allclose(raw_prediction, out)
+    assert_array_equal(out, raw_prediction_2)
+    assert np.shares_memory(out, raw_prediction_2)

.venv/lib/python3.12/site-packages/sklearn/_min_dependencies.py

@@ -0,0 +1,71 @@
+"""All minimum dependencies for scikit-learn."""
+
+import argparse
+from collections import defaultdict
+
+# scipy and cython should by in sync with pyproject.toml
+NUMPY_MIN_VERSION = "1.19.5"
+SCIPY_MIN_VERSION = "1.6.0"
+JOBLIB_MIN_VERSION = "1.2.0"
+THREADPOOLCTL_MIN_VERSION = "3.1.0"
+PYTEST_MIN_VERSION = "7.1.2"
+CYTHON_MIN_VERSION = "3.0.10"
+
+
+# 'build' and 'install' is included to have structured metadata for CI.
+# It will NOT be included in setup's extras_require
+# The values are (version_spec, comma separated tags)
+dependent_packages = {
+    "numpy": (NUMPY_MIN_VERSION, "build, install"),
+    "scipy": (SCIPY_MIN_VERSION, "build, install"),
+    "joblib": (JOBLIB_MIN_VERSION, "install"),
+    "threadpoolctl": (THREADPOOLCTL_MIN_VERSION, "install"),
+    "cython": (CYTHON_MIN_VERSION, "build"),
+    "meson-python": ("0.16.0", "build"),
+    "matplotlib": ("3.3.4", "benchmark, docs, examples, tests"),
+    "scikit-image": ("0.17.2", "docs, examples, tests"),
+    "pandas": ("1.1.5", "benchmark, docs, examples, tests"),
+    "seaborn": ("0.9.0", "docs, examples"),
+    "memory_profiler": ("0.57.0", "benchmark, docs"),
+    "pytest": (PYTEST_MIN_VERSION, "tests"),
+    "pytest-cov": ("2.9.0", "tests"),
+    "ruff": ("0.2.1", "tests"),
+    "black": ("24.3.0", "tests"),
+    "mypy": ("1.9", "tests"),
+    "pyamg": ("4.0.0", "tests"),
+    "polars": ("0.20.30", "docs, tests"),
+    "pyarrow": ("12.0.0", "tests"),
+    "sphinx": ("7.3.7", "docs"),
+    "sphinx-copybutton": ("0.5.2", "docs"),
+    "sphinx-gallery": ("0.16.0", "docs"),
+    "numpydoc": ("1.2.0", "docs, tests"),
+    "Pillow": ("7.1.2", "docs"),
+    "pooch": ("1.6.0", "docs, examples, tests"),
+    "sphinx-prompt": ("1.4.0", "docs"),
+    "sphinxext-opengraph": ("0.9.1", "docs"),
+    "plotly": ("5.14.0", "docs, examples"),
+    "sphinxcontrib-sass": ("0.3.4", "docs"),
+    "sphinx-remove-toctrees": ("1.0.0.post1", "docs"),
+    "sphinx-design": ("0.6.0", "docs"),
+    "pydata-sphinx-theme": ("0.15.3", "docs"),
+    # XXX: Pin conda-lock to the latest released version (needs manual update
+    # from time to time)
+    "conda-lock": ("2.5.6", "maintenance"),
+}
+
+
+# create inverse mapping for setuptools
+tag_to_packages: dict = defaultdict(list)
+for package, (min_version, extras) in dependent_packages.items():
+    for extra in extras.split(", "):
+        tag_to_packages[extra].append("{}>={}".format(package, min_version))
+
+
+# Used by CI to get the min dependencies
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description="Get min dependencies for a package")
+
+    parser.add_argument("package", choices=dependent_packages)
+    args = parser.parse_args()
+    min_version = dependent_packages[args.package][0]
+    print(min_version)

.venv/lib/python3.12/site-packages/sklearn/cluster/__init__.py

@@ -0,0 +1,53 @@
+"""Popular unsupervised clustering algorithms."""
+
+from ._affinity_propagation import AffinityPropagation, affinity_propagation
+from ._agglomerative import (
+    AgglomerativeClustering,
+    FeatureAgglomeration,
+    linkage_tree,
+    ward_tree,
+)
+from ._bicluster import SpectralBiclustering, SpectralCoclustering
+from ._birch import Birch
+from ._bisect_k_means import BisectingKMeans
+from ._dbscan import DBSCAN, dbscan
+from ._hdbscan.hdbscan import HDBSCAN
+from ._kmeans import KMeans, MiniBatchKMeans, k_means, kmeans_plusplus
+from ._mean_shift import MeanShift, estimate_bandwidth, get_bin_seeds, mean_shift
+from ._optics import (
+    OPTICS,
+    cluster_optics_dbscan,
+    cluster_optics_xi,
+    compute_optics_graph,
+)
+from ._spectral import SpectralClustering, spectral_clustering
+
+__all__ = [
+    "AffinityPropagation",
+    "AgglomerativeClustering",
+    "Birch",
+    "DBSCAN",
+    "OPTICS",
+    "cluster_optics_dbscan",
+    "cluster_optics_xi",
+    "compute_optics_graph",
+    "KMeans",
+    "BisectingKMeans",
+    "FeatureAgglomeration",
+    "MeanShift",
+    "MiniBatchKMeans",
+    "SpectralClustering",
+    "affinity_propagation",
+    "dbscan",
+    "estimate_bandwidth",
+    "get_bin_seeds",
+    "k_means",
+    "kmeans_plusplus",
+    "linkage_tree",
+    "mean_shift",
+    "spectral_clustering",
+    "ward_tree",
+    "SpectralBiclustering",
+    "SpectralCoclustering",
+    "HDBSCAN",
+]

.venv/lib/python3.12/site-packages/sklearn/cluster/_affinity_propagation.py

@@ -0,0 +1,601 @@
+"""Affinity Propagation clustering algorithm."""
+
+# Author: Alexandre Gramfort alexandre.gramfort@inria.fr
+#        Gael Varoquaux gael.varoquaux@normalesup.org
+
+# License: BSD 3 clause
+
+import warnings
+from numbers import Integral, Real
+
+import numpy as np
+
+from .._config import config_context
+from ..base import BaseEstimator, ClusterMixin, _fit_context
+from ..exceptions import ConvergenceWarning
+from ..metrics import euclidean_distances, pairwise_distances_argmin
+from ..utils import check_random_state
+from ..utils._param_validation import Interval, StrOptions, validate_params
+from ..utils.validation import check_is_fitted
+
+
+def _equal_similarities_and_preferences(S, preference):
+    def all_equal_preferences():
+        return np.all(preference == preference.flat[0])
+
+    def all_equal_similarities():
+        # Create mask to ignore diagonal of S
+        mask = np.ones(S.shape, dtype=bool)
+        np.fill_diagonal(mask, 0)
+
+        return np.all(S[mask].flat == S[mask].flat[0])
+
+    return all_equal_preferences() and all_equal_similarities()
+
+
+def _affinity_propagation(
+    S,
+    *,
+    preference,
+    convergence_iter,
+    max_iter,
+    damping,
+    verbose,
+    return_n_iter,
+    random_state,
+):
+    """Main affinity propagation algorithm."""
+    n_samples = S.shape[0]
+    if n_samples == 1 or _equal_similarities_and_preferences(S, preference):
+        # It makes no sense to run the algorithm in this case, so return 1 or
+        # n_samples clusters, depending on preferences
+        warnings.warn(
+            "All samples have mutually equal similarities. "
+            "Returning arbitrary cluster center(s)."
+        )
+        if preference.flat[0] > S.flat[n_samples - 1]:
+            return (
+                (np.arange(n_samples), np.arange(n_samples), 0)
+                if return_n_iter
+                else (np.arange(n_samples), np.arange(n_samples))
+            )
+        else:
+            return (
+                (np.array([0]), np.array([0] * n_samples), 0)
+                if return_n_iter
+                else (np.array([0]), np.array([0] * n_samples))
+            )
+
+    # Place preference on the diagonal of S
+    S.flat[:: (n_samples + 1)] = preference
+
+    A = np.zeros((n_samples, n_samples))
+    R = np.zeros((n_samples, n_samples))  # Initialize messages
+    # Intermediate results
+    tmp = np.zeros((n_samples, n_samples))
+
+    # Remove degeneracies
+    S += (
+        np.finfo(S.dtype).eps * S + np.finfo(S.dtype).tiny * 100
+    ) * random_state.standard_normal(size=(n_samples, n_samples))
+
+    # Execute parallel affinity propagation updates
+    e = np.zeros((n_samples, convergence_iter))
+
+    ind = np.arange(n_samples)
+
+    for it in range(max_iter):
+        # tmp = A + S; compute responsibilities
+        np.add(A, S, tmp)
+        I = np.argmax(tmp, axis=1)
+        Y = tmp[ind, I]  # np.max(A + S, axis=1)
+        tmp[ind, I] = -np.inf
+        Y2 = np.max(tmp, axis=1)
+
+        # tmp = Rnew
+        np.subtract(S, Y[:, None], tmp)
+        tmp[ind, I] = S[ind, I] - Y2
+
+        # Damping
+        tmp *= 1 - damping
+        R *= damping
+        R += tmp
+
+        # tmp = Rp; compute availabilities
+        np.maximum(R, 0, tmp)
+        tmp.flat[:: n_samples + 1] = R.flat[:: n_samples + 1]
+
+        # tmp = -Anew
+        tmp -= np.sum(tmp, axis=0)
+        dA = np.diag(tmp).copy()
+        tmp.clip(0, np.inf, tmp)
+        tmp.flat[:: n_samples + 1] = dA
+
+        # Damping
+        tmp *= 1 - damping
+        A *= damping
+        A -= tmp
+
+        # Check for convergence
+        E = (np.diag(A) + np.diag(R)) > 0
+        e[:, it % convergence_iter] = E
+        K = np.sum(E, axis=0)
+
+        if it >= convergence_iter:
+            se = np.sum(e, axis=1)
+            unconverged = np.sum((se == convergence_iter) + (se == 0)) != n_samples
+            if (not unconverged and (K > 0)) or (it == max_iter):
+                never_converged = False
+                if verbose:
+                    print("Converged after %d iterations." % it)
+                break
+    else:
+        never_converged = True
+        if verbose:
+            print("Did not converge")
+
+    I = np.flatnonzero(E)
+    K = I.size  # Identify exemplars
+
+    if K > 0:
+        if never_converged:
+            warnings.warn(
+                (
+                    "Affinity propagation did not converge, this model "
+                    "may return degenerate cluster centers and labels."
+                ),
+                ConvergenceWarning,
+            )
+        c = np.argmax(S[:, I], axis=1)
+        c[I] = np.arange(K)  # Identify clusters
+        # Refine the final set of exemplars and clusters and return results
+        for k in range(K):
+            ii = np.where(c == k)[0]
+            j = np.argmax(np.sum(S[ii[:, np.newaxis], ii], axis=0))
+            I[k] = ii[j]
+
+        c = np.argmax(S[:, I], axis=1)
+        c[I] = np.arange(K)
+        labels = I[c]
+        # Reduce labels to a sorted, gapless, list
+        cluster_centers_indices = np.unique(labels)
+        labels = np.searchsorted(cluster_centers_indices, labels)
+    else:
+        warnings.warn(
+            (
+                "Affinity propagation did not converge and this model "
+                "will not have any cluster centers."
+            ),
+            ConvergenceWarning,
+        )
+        labels = np.array([-1] * n_samples)
+        cluster_centers_indices = []
+
+    if return_n_iter:
+        return cluster_centers_indices, labels, it + 1
+    else:
+        return cluster_centers_indices, labels
+
+
+###############################################################################
+# Public API
+
+
+@validate_params(
+    {
+        "S": ["array-like"],
+        "return_n_iter": ["boolean"],
+    },
+    prefer_skip_nested_validation=False,
+)
+def affinity_propagation(
+    S,
+    *,
+    preference=None,
+    convergence_iter=15,
+    max_iter=200,
+    damping=0.5,
+    copy=True,
+    verbose=False,
+    return_n_iter=False,
+    random_state=None,
+):
+    """Perform Affinity Propagation Clustering of data.
+
+    Read more in the :ref:`User Guide <affinity_propagation>`.
+
+    Parameters
+    ----------
+    S : array-like of shape (n_samples, n_samples)
+        Matrix of similarities between points.
+
+    preference : array-like of shape (n_samples,) or float, default=None
+        Preferences for each point - points with larger values of
+        preferences are more likely to be chosen as exemplars. The number of
+        exemplars, i.e. of clusters, is influenced by the input preferences
+        value. If the preferences are not passed as arguments, they will be
+        set to the median of the input similarities (resulting in a moderate
+        number of clusters). For a smaller amount of clusters, this can be set
+        to the minimum value of the similarities.
+
+    convergence_iter : int, default=15
+        Number of iterations with no change in the number
+        of estimated clusters that stops the convergence.
+
+    max_iter : int, default=200
+        Maximum number of iterations.
+
+    damping : float, default=0.5
+        Damping factor between 0.5 and 1.
+
+    copy : bool, default=True
+        If copy is False, the affinity matrix is modified inplace by the
+        algorithm, for memory efficiency.
+
+    verbose : bool, default=False
+        The verbosity level.
+
+    return_n_iter : bool, default=False
+        Whether or not to return the number of iterations.
+
+    random_state : int, RandomState instance or None, default=None
+        Pseudo-random number generator to control the starting state.
+        Use an int for reproducible results across function calls.
+        See the :term:`Glossary <random_state>`.
+
+        .. versionadded:: 0.23
+            this parameter was previously hardcoded as 0.
+
+    Returns
+    -------
+    cluster_centers_indices : ndarray of shape (n_clusters,)
+        Index of clusters centers.
+
+    labels : ndarray of shape (n_samples,)
+        Cluster labels for each point.
+
+    n_iter : int
+        Number of iterations run. Returned only if `return_n_iter` is
+        set to True.
+
+    Notes
+    -----
+    For an example, see :ref:`examples/cluster/plot_affinity_propagation.py
+    <sphx_glr_auto_examples_cluster_plot_affinity_propagation.py>`.
+
+    When the algorithm does not converge, it will still return a arrays of
+    ``cluster_center_indices`` and labels if there are any exemplars/clusters,
+    however they may be degenerate and should be used with caution.
+
+    When all training samples have equal similarities and equal preferences,
+    the assignment of cluster centers and labels depends on the preference.
+    If the preference is smaller than the similarities, a single cluster center
+    and label ``0`` for every sample will be returned. Otherwise, every
+    training sample becomes its own cluster center and is assigned a unique
+    label.
+
+    References
+    ----------
+    Brendan J. Frey and Delbert Dueck, "Clustering by Passing Messages
+    Between Data Points", Science Feb. 2007
+
+    Examples
+    --------
+    >>> import numpy as np
+    >>> from sklearn.cluster import affinity_propagation
+    >>> from sklearn.metrics.pairwise import euclidean_distances
+    >>> X = np.array([[1, 2], [1, 4], [1, 0],
+    ...               [4, 2], [4, 4], [4, 0]])
+    >>> S = -euclidean_distances(X, squared=True)
+    >>> cluster_centers_indices, labels = affinity_propagation(S, random_state=0)
+    >>> cluster_centers_indices
+    array([0, 3])
+    >>> labels
+    array([0, 0, 0, 1, 1, 1])
+    """
+    estimator = AffinityPropagation(
+        damping=damping,
+        max_iter=max_iter,
+        convergence_iter=convergence_iter,
+        copy=copy,
+        preference=preference,
+        affinity="precomputed",
+        verbose=verbose,
+        random_state=random_state,
+    ).fit(S)
+
+    if return_n_iter:
+        return estimator.cluster_centers_indices_, estimator.labels_, estimator.n_iter_
+    return estimator.cluster_centers_indices_, estimator.labels_
+
+
+class AffinityPropagation(ClusterMixin, BaseEstimator):
+    """Perform Affinity Propagation Clustering of data.
+
+    Read more in the :ref:`User Guide <affinity_propagation>`.
+
+    Parameters
+    ----------
+    damping : float, default=0.5
+        Damping factor in the range `[0.5, 1.0)` is the extent to
+        which the current value is maintained relative to
+        incoming values (weighted 1 - damping). This in order
+        to avoid numerical oscillations when updating these
+        values (messages).
+
+    max_iter : int, default=200
+        Maximum number of iterations.
+
+    convergence_iter : int, default=15
+        Number of iterations with no change in the number
+        of estimated clusters that stops the convergence.
+
+    copy : bool, default=True
+        Make a copy of input data.
+
+    preference : array-like of shape (n_samples,) or float, default=None
+        Preferences for each point - points with larger values of
+        preferences are more likely to be chosen as exemplars. The number
+        of exemplars, ie of clusters, is influenced by the input
+        preferences value. If the preferences are not passed as arguments,
+        they will be set to the median of the input similarities.
+
+    affinity : {'euclidean', 'precomputed'}, default='euclidean'
+        Which affinity to use. At the moment 'precomputed' and
+        ``euclidean`` are supported. 'euclidean' uses the
+        negative squared euclidean distance between points.
+
+    verbose : bool, default=False
+        Whether to be verbose.
+
+    random_state : int, RandomState instance or None, default=None
+        Pseudo-random number generator to control the starting state.
+        Use an int for reproducible results across function calls.
+        See the :term:`Glossary <random_state>`.
+
+        .. versionadded:: 0.23
+            this parameter was previously hardcoded as 0.
+
+    Attributes
+    ----------
+    cluster_centers_indices_ : ndarray of shape (n_clusters,)
+        Indices of cluster centers.
+
+    cluster_centers_ : ndarray of shape (n_clusters, n_features)
+        Cluster centers (if affinity != ``precomputed``).
+
+    labels_ : ndarray of shape (n_samples,)
+        Labels of each point.
+
+    affinity_matrix_ : ndarray of shape (n_samples, n_samples)
+        Stores the affinity matrix used in ``fit``.
+
+    n_iter_ : int
+        Number of iterations taken to converge.
+
+    n_features_in_ : int
+        Number of features seen during :term:`fit`.
+
+        .. versionadded:: 0.24
+
+    feature_names_in_ : ndarray of shape (`n_features_in_`,)
+        Names of features seen during :term:`fit`. Defined only when `X`
+        has feature names that are all strings.
+
+        .. versionadded:: 1.0
+
+    See Also
+    --------
+    AgglomerativeClustering : Recursively merges the pair of
+        clusters that minimally increases a given linkage distance.
+    FeatureAgglomeration : Similar to AgglomerativeClustering,
+        but recursively merges features instead of samples.
+    KMeans : K-Means clustering.
+    MiniBatchKMeans : Mini-Batch K-Means clustering.
+    MeanShift : Mean shift clustering using a flat kernel.
+    SpectralClustering : Apply clustering to a projection
+        of the normalized Laplacian.
+
+    Notes
+    -----
+    For an example, see :ref:`examples/cluster/plot_affinity_propagation.py
+    <sphx_glr_auto_examples_cluster_plot_affinity_propagation.py>`.
+
+    The algorithmic complexity of affinity propagation is quadratic
+    in the number of points.
+
+    When the algorithm does not converge, it will still return a arrays of
+    ``cluster_center_indices`` and labels if there are any exemplars/clusters,
+    however they may be degenerate and should be used with caution.
+
+    When ``fit`` does not converge, ``cluster_centers_`` is still populated
+    however it may be degenerate. In such a case, proceed with caution.
+    If ``fit`` does not converge and fails to produce any ``cluster_centers_``
+    then ``predict`` will label every sample as ``-1``.
+
+    When all training samples have equal similarities and equal preferences,
+    the assignment of cluster centers and labels depends on the preference.
+    If the preference is smaller than the similarities, ``fit`` will result in
+    a single cluster center and label ``0`` for every sample. Otherwise, every
+    training sample becomes its own cluster center and is assigned a unique
+    label.
+
+    References
+    ----------
+
+    Brendan J. Frey and Delbert Dueck, "Clustering by Passing Messages
+    Between Data Points", Science Feb. 2007
+
+    Examples
+    --------
+    >>> from sklearn.cluster import AffinityPropagation
+    >>> import numpy as np
+    >>> X = np.array([[1, 2], [1, 4], [1, 0],
+    ...               [4, 2], [4, 4], [4, 0]])
+    >>> clustering = AffinityPropagation(random_state=5).fit(X)
+    >>> clustering
+    AffinityPropagation(random_state=5)
+    >>> clustering.labels_
+    array([0, 0, 0, 1, 1, 1])
+    >>> clustering.predict([[0, 0], [4, 4]])
+    array([0, 1])
+    >>> clustering.cluster_centers_
+    array([[1, 2],
+           [4, 2]])
+    """
+
+    _parameter_constraints: dict = {
+        "damping": [Interval(Real, 0.5, 1.0, closed="left")],
+        "max_iter": [Interval(Integral, 1, None, closed="left")],
+        "convergence_iter": [Interval(Integral, 1, None, closed="left")],
+        "copy": ["boolean"],
+        "preference": [
+            "array-like",
+            Interval(Real, None, None, closed="neither"),
+            None,
+        ],
+        "affinity": [StrOptions({"euclidean", "precomputed"})],
+        "verbose": ["verbose"],
+        "random_state": ["random_state"],
+    }
+
+    def __init__(
+        self,
+        *,
+        damping=0.5,
+        max_iter=200,
+        convergence_iter=15,
+        copy=True,
+        preference=None,
+        affinity="euclidean",
+        verbose=False,
+        random_state=None,
+    ):
+        self.damping = damping
+        self.max_iter = max_iter
+        self.convergence_iter = convergence_iter
+        self.copy = copy
+        self.verbose = verbose
+        self.preference = preference
+        self.affinity = affinity
+        self.random_state = random_state
+
+    def _more_tags(self):
+        return {"pairwise": self.affinity == "precomputed"}
+
+    @_fit_context(prefer_skip_nested_validation=True)
+    def fit(self, X, y=None):
+        """Fit the clustering from features, or affinity matrix.
+
+        Parameters
+        ----------
+        X : {array-like, sparse matrix} of shape (n_samples, n_features), or \
+                array-like of shape (n_samples, n_samples)
+            Training instances to cluster, or similarities / affinities between
+            instances if ``affinity='precomputed'``. If a sparse feature matrix
+            is provided, it will be converted into a sparse ``csr_matrix``.
+
+        y : Ignored
+            Not used, present here for API consistency by convention.
+
+        Returns
+        -------
+        self
+            Returns the instance itself.
+        """
+        if self.affinity == "precomputed":
+            X = self._validate_data(X, copy=self.copy, force_writeable=True)
+            self.affinity_matrix_ = X
+        else:  # self.affinity == "euclidean"
+            X = self._validate_data(X, accept_sparse="csr")
+            self.affinity_matrix_ = -euclidean_distances(X, squared=True)
+
+        if self.affinity_matrix_.shape[0] != self.affinity_matrix_.shape[1]:
+            raise ValueError(
+                "The matrix of similarities must be a square array. "
+                f"Got {self.affinity_matrix_.shape} instead."
+            )
+
+        if self.preference is None:
+            preference = np.median(self.affinity_matrix_)
+        else:
+            preference = self.preference
+        preference = np.asarray(preference)
+
+        random_state = check_random_state(self.random_state)
+
+        (
+            self.cluster_centers_indices_,
+            self.labels_,
+            self.n_iter_,
+        ) = _affinity_propagation(
+            self.affinity_matrix_,
+            max_iter=self.max_iter,
+            convergence_iter=self.convergence_iter,
+            preference=preference,
+            damping=self.damping,
+            verbose=self.verbose,
+            return_n_iter=True,
+            random_state=random_state,
+        )
+
+        if self.affinity != "precomputed":
+            self.cluster_centers_ = X[self.cluster_centers_indices_].copy()
+
+        return self
+
+    def predict(self, X):
+        """Predict the closest cluster each sample in X belongs to.
+
+        Parameters
+        ----------
+        X : {array-like, sparse matrix} of shape (n_samples, n_features)
+            New data to predict. If a sparse matrix is provided, it will be
+            converted into a sparse ``csr_matrix``.
+
+        Returns
+        -------
+        labels : ndarray of shape (n_samples,)
+            Cluster labels.
+        """
+        check_is_fitted(self)
+        X = self._validate_data(X, reset=False, accept_sparse="csr")
+        if not hasattr(self, "cluster_centers_"):
+            raise ValueError(
+                "Predict method is not supported when affinity='precomputed'."
+            )
+
+        if self.cluster_centers_.shape[0] > 0:
+            with config_context(assume_finite=True):
+                return pairwise_distances_argmin(X, self.cluster_centers_)
+        else:
+            warnings.warn(
+                (
+                    "This model does not have any cluster centers "
+                    "because affinity propagation did not converge. "
+                    "Labeling every sample as '-1'."
+                ),
+                ConvergenceWarning,
+            )
+            return np.array([-1] * X.shape[0])
+
+    def fit_predict(self, X, y=None):
+        """Fit clustering from features/affinity matrix; return cluster labels.
+
+        Parameters
+        ----------
+        X : {array-like, sparse matrix} of shape (n_samples, n_features), or \
+                array-like of shape (n_samples, n_samples)
+            Training instances to cluster, or similarities / affinities between
+            instances if ``affinity='precomputed'``. If a sparse feature matrix
+            is provided, it will be converted into a sparse ``csr_matrix``.
+
+        y : Ignored
+            Not used, present here for API consistency by convention.
+
+        Returns
+        -------
+        labels : ndarray of shape (n_samples,)
+            Cluster labels.
+        """
+        return super().fit_predict(X, y)

.venv/lib/python3.12/site-packages/sklearn/cluster/_bicluster.py

@@ -0,0 +1,627 @@
+"""Spectral biclustering algorithms."""
+
+# Authors : Kemal Eren
+# License: BSD 3 clause
+
+from abc import ABCMeta, abstractmethod
+from numbers import Integral
+
+import numpy as np
+from scipy.linalg import norm
+from scipy.sparse import dia_matrix, issparse
+from scipy.sparse.linalg import eigsh, svds
+
+from ..base import BaseEstimator, BiclusterMixin, _fit_context
+from ..utils import check_random_state, check_scalar
+from ..utils._param_validation import Interval, StrOptions
+from ..utils.extmath import make_nonnegative, randomized_svd, safe_sparse_dot
+from ..utils.validation import assert_all_finite
+from ._kmeans import KMeans, MiniBatchKMeans
+
+__all__ = ["SpectralCoclustering", "SpectralBiclustering"]
+
+
+def _scale_normalize(X):
+    """Normalize ``X`` by scaling rows and columns independently.
+
+    Returns the normalized matrix and the row and column scaling
+    factors.
+    """
+    X = make_nonnegative(X)
+    row_diag = np.asarray(1.0 / np.sqrt(X.sum(axis=1))).squeeze()
+    col_diag = np.asarray(1.0 / np.sqrt(X.sum(axis=0))).squeeze()
+    row_diag = np.where(np.isnan(row_diag), 0, row_diag)
+    col_diag = np.where(np.isnan(col_diag), 0, col_diag)
+    if issparse(X):
+        n_rows, n_cols = X.shape
+        r = dia_matrix((row_diag, [0]), shape=(n_rows, n_rows))
+        c = dia_matrix((col_diag, [0]), shape=(n_cols, n_cols))
+        an = r * X * c
+    else:
+        an = row_diag[:, np.newaxis] * X * col_diag
+    return an, row_diag, col_diag
+
+
+def _bistochastic_normalize(X, max_iter=1000, tol=1e-5):
+    """Normalize rows and columns of ``X`` simultaneously so that all
+    rows sum to one constant and all columns sum to a different
+    constant.
+    """
+    # According to paper, this can also be done more efficiently with
+    # deviation reduction and balancing algorithms.
+    X = make_nonnegative(X)
+    X_scaled = X
+    for _ in range(max_iter):
+        X_new, _, _ = _scale_normalize(X_scaled)
+        if issparse(X):
+            dist = norm(X_scaled.data - X.data)
+        else:
+            dist = norm(X_scaled - X_new)
+        X_scaled = X_new
+        if dist is not None and dist < tol:
+            break
+    return X_scaled
+
+
+def _log_normalize(X):
+    """Normalize ``X`` according to Kluger's log-interactions scheme."""
+    X = make_nonnegative(X, min_value=1)
+    if issparse(X):
+        raise ValueError(
+            "Cannot compute log of a sparse matrix,"
+            " because log(x) diverges to -infinity as x"
+            " goes to 0."
+        )
+    L = np.log(X)
+    row_avg = L.mean(axis=1)[:, np.newaxis]
+    col_avg = L.mean(axis=0)
+    avg = L.mean()
+    return L - row_avg - col_avg + avg
+
+
+class BaseSpectral(BiclusterMixin, BaseEstimator, metaclass=ABCMeta):
+    """Base class for spectral biclustering."""
+
+    _parameter_constraints: dict = {
+        "svd_method": [StrOptions({"randomized", "arpack"})],
+        "n_svd_vecs": [Interval(Integral, 0, None, closed="left"), None],
+        "mini_batch": ["boolean"],
+        "init": [StrOptions({"k-means++", "random"}), np.ndarray],
+        "n_init": [Interval(Integral, 1, None, closed="left")],
+        "random_state": ["random_state"],
+    }
+
+    @abstractmethod
+    def __init__(
+        self,
+        n_clusters=3,
+        svd_method="randomized",
+        n_svd_vecs=None,
+        mini_batch=False,
+        init="k-means++",
+        n_init=10,
+        random_state=None,
+    ):
+        self.n_clusters = n_clusters
+        self.svd_method = svd_method
+        self.n_svd_vecs = n_svd_vecs
+        self.mini_batch = mini_batch
+        self.init = init
+        self.n_init = n_init
+        self.random_state = random_state
+
+    @abstractmethod
+    def _check_parameters(self, n_samples):
+        """Validate parameters depending on the input data."""
+
+    @_fit_context(prefer_skip_nested_validation=True)
+    def fit(self, X, y=None):
+        """Create a biclustering for X.
+
+        Parameters
+        ----------
+        X : array-like of shape (n_samples, n_features)
+            Training data.
+
+        y : Ignored
+            Not used, present for API consistency by convention.
+
+        Returns
+        -------
+        self : object
+            SpectralBiclustering instance.
+        """
+        X = self._validate_data(X, accept_sparse="csr", dtype=np.float64)
+        self._check_parameters(X.shape[0])
+        self._fit(X)
+        return self
+
+    def _svd(self, array, n_components, n_discard):
+        """Returns first `n_components` left and right singular
+        vectors u and v, discarding the first `n_discard`.
+        """
+        if self.svd_method == "randomized":
+            kwargs = {}
+            if self.n_svd_vecs is not None:
+                kwargs["n_oversamples"] = self.n_svd_vecs
+            u, _, vt = randomized_svd(
+                array, n_components, random_state=self.random_state, **kwargs
+            )
+
+        elif self.svd_method == "arpack":
+            u, _, vt = svds(array, k=n_components, ncv=self.n_svd_vecs)
+            if np.any(np.isnan(vt)):
+                # some eigenvalues of A * A.T are negative, causing
+                # sqrt() to be np.nan. This causes some vectors in vt
+                # to be np.nan.
+                A = safe_sparse_dot(array.T, array)
+                random_state = check_random_state(self.random_state)
+                # initialize with [-1,1] as in ARPACK
+                v0 = random_state.uniform(-1, 1, A.shape[0])
+                _, v = eigsh(A, ncv=self.n_svd_vecs, v0=v0)
+                vt = v.T
+            if np.any(np.isnan(u)):
+                A = safe_sparse_dot(array, array.T)
+                random_state = check_random_state(self.random_state)
+                # initialize with [-1,1] as in ARPACK
+                v0 = random_state.uniform(-1, 1, A.shape[0])
+                _, u = eigsh(A, ncv=self.n_svd_vecs, v0=v0)
+
+        assert_all_finite(u)
+        assert_all_finite(vt)
+        u = u[:, n_discard:]
+        vt = vt[n_discard:]
+        return u, vt.T
+
+    def _k_means(self, data, n_clusters):
+        if self.mini_batch:
+            model = MiniBatchKMeans(
+                n_clusters,
+                init=self.init,
+                n_init=self.n_init,
+                random_state=self.random_state,
+            )
+        else:
+            model = KMeans(
+                n_clusters,
+                init=self.init,
+                n_init=self.n_init,
+                random_state=self.random_state,
+            )
+        model.fit(data)
+        centroid = model.cluster_centers_
+        labels = model.labels_
+        return centroid, labels
+
+    def _more_tags(self):
+        return {
+            "_xfail_checks": {
+                "check_estimators_dtypes": "raises nan error",
+                "check_fit2d_1sample": "_scale_normalize fails",
+                "check_fit2d_1feature": "raises apply_along_axis error",
+                "check_estimator_sparse_matrix": "does not fail gracefully",
+                "check_estimator_sparse_array": "does not fail gracefully",
+                "check_methods_subset_invariance": "empty array passed inside",
+                "check_dont_overwrite_parameters": "empty array passed inside",
+                "check_fit2d_predict1d": "empty array passed inside",
+            }
+        }
+
+
+class SpectralCoclustering(BaseSpectral):
+    """Spectral Co-Clustering algorithm (Dhillon, 2001).
+
+    Clusters rows and columns of an array `X` to solve the relaxed
+    normalized cut of the bipartite graph created from `X` as follows:
+    the edge between row vertex `i` and column vertex `j` has weight
+    `X[i, j]`.
+
+    The resulting bicluster structure is block-diagonal, since each
+    row and each column belongs to exactly one bicluster.
+
+    Supports sparse matrices, as long as they are nonnegative.
+
+    Read more in the :ref:`User Guide <spectral_coclustering>`.
+
+    Parameters
+    ----------
+    n_clusters : int, default=3
+        The number of biclusters to find.
+
+    svd_method : {'randomized', 'arpack'}, default='randomized'
+        Selects the algorithm for finding singular vectors. May be
+        'randomized' or 'arpack'. If 'randomized', use
+        :func:`sklearn.utils.extmath.randomized_svd`, which may be faster
+        for large matrices. If 'arpack', use
+        :func:`scipy.sparse.linalg.svds`, which is more accurate, but
+        possibly slower in some cases.
+
+    n_svd_vecs : int, default=None
+        Number of vectors to use in calculating the SVD. Corresponds
+        to `ncv` when `svd_method=arpack` and `n_oversamples` when
+        `svd_method` is 'randomized`.
+
+    mini_batch : bool, default=False
+        Whether to use mini-batch k-means, which is faster but may get
+        different results.
+
+    init : {'k-means++', 'random'}, or ndarray of shape \
+            (n_clusters, n_features), default='k-means++'
+        Method for initialization of k-means algorithm; defaults to
+        'k-means++'.
+
+    n_init : int, default=10
+        Number of random initializations that are tried with the
+        k-means algorithm.
+
+        If mini-batch k-means is used, the best initialization is
+        chosen and the algorithm runs once. Otherwise, the algorithm
+        is run for each initialization and the best solution chosen.
+
+    random_state : int, RandomState instance, default=None
+        Used for randomizing the singular value decomposition and the k-means
+        initialization. Use an int to make the randomness deterministic.
+        See :term:`Glossary <random_state>`.
+
+    Attributes
+    ----------
+    rows_ : array-like of shape (n_row_clusters, n_rows)
+        Results of the clustering. `rows[i, r]` is True if
+        cluster `i` contains row `r`. Available only after calling ``fit``.
+
+    columns_ : array-like of shape (n_column_clusters, n_columns)
+        Results of the clustering, like `rows`.
+
+    row_labels_ : array-like of shape (n_rows,)
+        The bicluster label of each row.
+
+    column_labels_ : array-like of shape (n_cols,)
+        The bicluster label of each column.
+
+    biclusters_ : tuple of two ndarrays
+        The tuple contains the `rows_` and `columns_` arrays.
+
+    n_features_in_ : int
+        Number of features seen during :term:`fit`.
+
+        .. versionadded:: 0.24
+
+    feature_names_in_ : ndarray of shape (`n_features_in_`,)
+        Names of features seen during :term:`fit`. Defined only when `X`
+        has feature names that are all strings.
+
+        .. versionadded:: 1.0
+
+    See Also
+    --------
+    SpectralBiclustering : Partitions rows and columns under the assumption
+        that the data has an underlying checkerboard structure.
+
+    References
+    ----------
+    * :doi:`Dhillon, Inderjit S, 2001. Co-clustering documents and words using
+      bipartite spectral graph partitioning.
+      <10.1145/502512.502550>`
+
+    Examples
+    --------
+    >>> from sklearn.cluster import SpectralCoclustering
+    >>> import numpy as np
+    >>> X = np.array([[1, 1], [2, 1], [1, 0],
+    ...               [4, 7], [3, 5], [3, 6]])
+    >>> clustering = SpectralCoclustering(n_clusters=2, random_state=0).fit(X)
+    >>> clustering.row_labels_ #doctest: +SKIP
+    array([0, 1, 1, 0, 0, 0], dtype=int32)
+    >>> clustering.column_labels_ #doctest: +SKIP
+    array([0, 0], dtype=int32)
+    >>> clustering
+    SpectralCoclustering(n_clusters=2, random_state=0)
+    """
+
+    _parameter_constraints: dict = {
+        **BaseSpectral._parameter_constraints,
+        "n_clusters": [Interval(Integral, 1, None, closed="left")],
+    }
+
+    def __init__(
+        self,
+        n_clusters=3,
+        *,
+        svd_method="randomized",
+        n_svd_vecs=None,
+        mini_batch=False,
+        init="k-means++",
+        n_init=10,
+        random_state=None,
+    ):
+        super().__init__(
+            n_clusters, svd_method, n_svd_vecs, mini_batch, init, n_init, random_state
+        )
+
+    def _check_parameters(self, n_samples):
+        if self.n_clusters > n_samples:
+            raise ValueError(
+                f"n_clusters should be <= n_samples={n_samples}. Got"
+                f" {self.n_clusters} instead."
+            )
+
+    def _fit(self, X):
+        normalized_data, row_diag, col_diag = _scale_normalize(X)
+        n_sv = 1 + int(np.ceil(np.log2(self.n_clusters)))
+        u, v = self._svd(normalized_data, n_sv, n_discard=1)
+        z = np.vstack((row_diag[:, np.newaxis] * u, col_diag[:, np.newaxis] * v))
+
+        _, labels = self._k_means(z, self.n_clusters)
+
+        n_rows = X.shape[0]
+        self.row_labels_ = labels[:n_rows]
+        self.column_labels_ = labels[n_rows:]
+
+        self.rows_ = np.vstack([self.row_labels_ == c for c in range(self.n_clusters)])
+        self.columns_ = np.vstack(
+            [self.column_labels_ == c for c in range(self.n_clusters)]
+        )
+
+
+class SpectralBiclustering(BaseSpectral):
+    """Spectral biclustering (Kluger, 2003).
+
+    Partitions rows and columns under the assumption that the data has
+    an underlying checkerboard structure. For instance, if there are
+    two row partitions and three column partitions, each row will
+    belong to three biclusters, and each column will belong to two
+    biclusters. The outer product of the corresponding row and column
+    label vectors gives this checkerboard structure.
+
+    Read more in the :ref:`User Guide <spectral_biclustering>`.
+
+    Parameters
+    ----------
+    n_clusters : int or tuple (n_row_clusters, n_column_clusters), default=3
+        The number of row and column clusters in the checkerboard
+        structure.
+
+    method : {'bistochastic', 'scale', 'log'}, default='bistochastic'
+        Method of normalizing and converting singular vectors into
+        biclusters. May be one of 'scale', 'bistochastic', or 'log'.
+        The authors recommend using 'log'. If the data is sparse,
+        however, log normalization will not work, which is why the
+        default is 'bistochastic'.
+
+        .. warning::
+           if `method='log'`, the data must not be sparse.
+
+    n_components : int, default=6
+        Number of singular vectors to check.
+
+    n_best : int, default=3
+        Number of best singular vectors to which to project the data
+        for clustering.
+
+    svd_method : {'randomized', 'arpack'}, default='randomized'
+        Selects the algorithm for finding singular vectors. May be
+        'randomized' or 'arpack'. If 'randomized', uses
+        :func:`~sklearn.utils.extmath.randomized_svd`, which may be faster
+        for large matrices. If 'arpack', uses
+        `scipy.sparse.linalg.svds`, which is more accurate, but
+        possibly slower in some cases.
+
+    n_svd_vecs : int, default=None
+        Number of vectors to use in calculating the SVD. Corresponds
+        to `ncv` when `svd_method=arpack` and `n_oversamples` when
+        `svd_method` is 'randomized`.
+
+    mini_batch : bool, default=False
+        Whether to use mini-batch k-means, which is faster but may get
+        different results.
+
+    init : {'k-means++', 'random'} or ndarray of shape (n_clusters, n_features), \
+            default='k-means++'
+        Method for initialization of k-means algorithm; defaults to
+        'k-means++'.
+
+    n_init : int, default=10
+        Number of random initializations that are tried with the
+        k-means algorithm.
+
+        If mini-batch k-means is used, the best initialization is
+        chosen and the algorithm runs once. Otherwise, the algorithm
+        is run for each initialization and the best solution chosen.
+
+    random_state : int, RandomState instance, default=None
+        Used for randomizing the singular value decomposition and the k-means
+        initialization. Use an int to make the randomness deterministic.
+        See :term:`Glossary <random_state>`.
+
+    Attributes
+    ----------
+    rows_ : array-like of shape (n_row_clusters, n_rows)
+        Results of the clustering. `rows[i, r]` is True if
+        cluster `i` contains row `r`. Available only after calling ``fit``.
+
+    columns_ : array-like of shape (n_column_clusters, n_columns)
+        Results of the clustering, like `rows`.
+
+    row_labels_ : array-like of shape (n_rows,)
+        Row partition labels.
+
+    column_labels_ : array-like of shape (n_cols,)
+        Column partition labels.
+
+    biclusters_ : tuple of two ndarrays
+        The tuple contains the `rows_` and `columns_` arrays.
+
+    n_features_in_ : int
+        Number of features seen during :term:`fit`.
+
+        .. versionadded:: 0.24
+
+    feature_names_in_ : ndarray of shape (`n_features_in_`,)
+        Names of features seen during :term:`fit`. Defined only when `X`
+        has feature names that are all strings.
+
+        .. versionadded:: 1.0
+
+    See Also
+    --------
+    SpectralCoclustering : Spectral Co-Clustering algorithm (Dhillon, 2001).
+
+    References
+    ----------
+
+    * :doi:`Kluger, Yuval, et. al., 2003. Spectral biclustering of microarray
+      data: coclustering genes and conditions.
+      <10.1101/gr.648603>`
+
+    Examples
+    --------
+    >>> from sklearn.cluster import SpectralBiclustering
+    >>> import numpy as np
+    >>> X = np.array([[1, 1], [2, 1], [1, 0],
+    ...               [4, 7], [3, 5], [3, 6]])
+    >>> clustering = SpectralBiclustering(n_clusters=2, random_state=0).fit(X)
+    >>> clustering.row_labels_
+    array([1, 1, 1, 0, 0, 0], dtype=int32)
+    >>> clustering.column_labels_
+    array([1, 0], dtype=int32)
+    >>> clustering
+    SpectralBiclustering(n_clusters=2, random_state=0)
+
+    For a more detailed example, see
+    :ref:`sphx_glr_auto_examples_bicluster_plot_spectral_biclustering.py`
+    """
+
+    _parameter_constraints: dict = {
+        **BaseSpectral._parameter_constraints,
+        "n_clusters": [Interval(Integral, 1, None, closed="left"), tuple],
+        "method": [StrOptions({"bistochastic", "scale", "log"})],
+        "n_components": [Interval(Integral, 1, None, closed="left")],
+        "n_best": [Interval(Integral, 1, None, closed="left")],
+    }
+
+    def __init__(
+        self,
+        n_clusters=3,
+        *,
+        method="bistochastic",
+        n_components=6,
+        n_best=3,
+        svd_method="randomized",
+        n_svd_vecs=None,
+        mini_batch=False,
+        init="k-means++",
+        n_init=10,
+        random_state=None,
+    ):
+        super().__init__(
+            n_clusters, svd_method, n_svd_vecs, mini_batch, init, n_init, random_state
+        )
+        self.method = method
+        self.n_components = n_components
+        self.n_best = n_best
+
+    def _check_parameters(self, n_samples):
+        if isinstance(self.n_clusters, Integral):
+            if self.n_clusters > n_samples:
+                raise ValueError(
+                    f"n_clusters should be <= n_samples={n_samples}. Got"
+                    f" {self.n_clusters} instead."
+                )
+        else:  # tuple
+            try:
+                n_row_clusters, n_column_clusters = self.n_clusters
+                check_scalar(
+                    n_row_clusters,
+                    "n_row_clusters",
+                    target_type=Integral,
+                    min_val=1,
+                    max_val=n_samples,
+                )
+                check_scalar(
+                    n_column_clusters,
+                    "n_column_clusters",
+                    target_type=Integral,
+                    min_val=1,
+                    max_val=n_samples,
+                )
+            except (ValueError, TypeError) as e:
+                raise ValueError(
+                    "Incorrect parameter n_clusters has value:"
+                    f" {self.n_clusters}. It should either be a single integer"
+                    " or an iterable with two integers:"
+                    " (n_row_clusters, n_column_clusters)"
+                    " And the values are should be in the"
+                    " range: (1, n_samples)"
+                ) from e
+
+        if self.n_best > self.n_components:
+            raise ValueError(
+                f"n_best={self.n_best} must be <= n_components={self.n_components}."
+            )
+
+    def _fit(self, X):
+        n_sv = self.n_components
+        if self.method == "bistochastic":
+            normalized_data = _bistochastic_normalize(X)
+            n_sv += 1
+        elif self.method == "scale":
+            normalized_data, _, _ = _scale_normalize(X)
+            n_sv += 1
+        elif self.method == "log":
+            normalized_data = _log_normalize(X)
+        n_discard = 0 if self.method == "log" else 1
+        u, v = self._svd(normalized_data, n_sv, n_discard)
+        ut = u.T
+        vt = v.T
+
+        try:
+            n_row_clusters, n_col_clusters = self.n_clusters
+        except TypeError:
+            n_row_clusters = n_col_clusters = self.n_clusters
+
+        best_ut = self._fit_best_piecewise(ut, self.n_best, n_row_clusters)
+
+        best_vt = self._fit_best_piecewise(vt, self.n_best, n_col_clusters)
+
+        self.row_labels_ = self._project_and_cluster(X, best_vt.T, n_row_clusters)
+
+        self.column_labels_ = self._project_and_cluster(X.T, best_ut.T, n_col_clusters)
+
+        self.rows_ = np.vstack(
+            [
+                self.row_labels_ == label
+                for label in range(n_row_clusters)
+                for _ in range(n_col_clusters)
+            ]
+        )
+        self.columns_ = np.vstack(
+            [
+                self.column_labels_ == label
+                for _ in range(n_row_clusters)
+                for label in range(n_col_clusters)
+            ]
+        )
+
+    def _fit_best_piecewise(self, vectors, n_best, n_clusters):
+        """Find the ``n_best`` vectors that are best approximated by piecewise
+        constant vectors.
+
+        The piecewise vectors are found by k-means; the best is chosen
+        according to Euclidean distance.
+
+        """
+
+        def make_piecewise(v):
+            centroid, labels = self._k_means(v.reshape(-1, 1), n_clusters)
+            return centroid[labels].ravel()
+
+        piecewise_vectors = np.apply_along_axis(make_piecewise, axis=1, arr=vectors)
+        dists = np.apply_along_axis(norm, axis=1, arr=(vectors - piecewise_vectors))
+        result = vectors[np.argsort(dists)[:n_best]]
+        return result
+
+    def _project_and_cluster(self, data, vectors, n_clusters):
+        """Project ``data`` to ``vectors`` and cluster the result."""
+        projected = safe_sparse_dot(data, vectors)
+        _, labels = self._k_means(projected, n_clusters)
+        return labels

.venv/lib/python3.12/site-packages/sklearn/cluster/_birch.py

@@ -0,0 +1,741 @@
+# Authors: Manoj Kumar <manojkumarsivaraj334@gmail.com>
+#          Alexandre Gramfort <alexandre.gramfort@telecom-paristech.fr>
+#          Joel Nothman <joel.nothman@gmail.com>
+# License: BSD 3 clause
+
+import warnings
+from math import sqrt
+from numbers import Integral, Real
+
+import numpy as np
+from scipy import sparse
+
+from .._config import config_context
+from ..base import (
+    BaseEstimator,
+    ClassNamePrefixFeaturesOutMixin,
+    ClusterMixin,
+    TransformerMixin,
+    _fit_context,
+)
+from ..exceptions import ConvergenceWarning
+from ..metrics import pairwise_distances_argmin
+from ..metrics.pairwise import euclidean_distances
+from ..utils._param_validation import Interval
+from ..utils.extmath import row_norms
+from ..utils.validation import check_is_fitted
+from . import AgglomerativeClustering
+
+
+def _iterate_sparse_X(X):
+    """This little hack returns a densified row when iterating over a sparse
+    matrix, instead of constructing a sparse matrix for every row that is
+    expensive.
+    """
+    n_samples = X.shape[0]
+    X_indices = X.indices
+    X_data = X.data
+    X_indptr = X.indptr
+
+    for i in range(n_samples):
+        row = np.zeros(X.shape[1])
+        startptr, endptr = X_indptr[i], X_indptr[i + 1]
+        nonzero_indices = X_indices[startptr:endptr]
+        row[nonzero_indices] = X_data[startptr:endptr]
+        yield row
+
+
+def _split_node(node, threshold, branching_factor):
+    """The node has to be split if there is no place for a new subcluster
+    in the node.
+    1. Two empty nodes and two empty subclusters are initialized.
+    2. The pair of distant subclusters are found.
+    3. The properties of the empty subclusters and nodes are updated
+       according to the nearest distance between the subclusters to the
+       pair of distant subclusters.
+    4. The two nodes are set as children to the two subclusters.
+    """
+    new_subcluster1 = _CFSubcluster()
+    new_subcluster2 = _CFSubcluster()
+    new_node1 = _CFNode(
+        threshold=threshold,
+        branching_factor=branching_factor,
+        is_leaf=node.is_leaf,
+        n_features=node.n_features,
+        dtype=node.init_centroids_.dtype,
+    )
+    new_node2 = _CFNode(
+        threshold=threshold,
+        branching_factor=branching_factor,
+        is_leaf=node.is_leaf,
+        n_features=node.n_features,
+        dtype=node.init_centroids_.dtype,
+    )
+    new_subcluster1.child_ = new_node1
+    new_subcluster2.child_ = new_node2
+
+    if node.is_leaf:
+        if node.prev_leaf_ is not None:
+            node.prev_leaf_.next_leaf_ = new_node1
+        new_node1.prev_leaf_ = node.prev_leaf_
+        new_node1.next_leaf_ = new_node2
+        new_node2.prev_leaf_ = new_node1
+        new_node2.next_leaf_ = node.next_leaf_
+        if node.next_leaf_ is not None:
+            node.next_leaf_.prev_leaf_ = new_node2
+
+    dist = euclidean_distances(
+        node.centroids_, Y_norm_squared=node.squared_norm_, squared=True
+    )
+    n_clusters = dist.shape[0]
+
+    farthest_idx = np.unravel_index(dist.argmax(), (n_clusters, n_clusters))
+    node1_dist, node2_dist = dist[(farthest_idx,)]
+
+    node1_closer = node1_dist < node2_dist
+    # make sure node1 is closest to itself even if all distances are equal.
+    # This can only happen when all node.centroids_ are duplicates leading to all
+    # distances between centroids being zero.
+    node1_closer[farthest_idx[0]] = True
+
+    for idx, subcluster in enumerate(node.subclusters_):
+        if node1_closer[idx]:
+            new_node1.append_subcluster(subcluster)
+            new_subcluster1.update(subcluster)
+        else:
+            new_node2.append_subcluster(subcluster)
+            new_subcluster2.update(subcluster)
+    return new_subcluster1, new_subcluster2
+
+
+class _CFNode:
+    """Each node in a CFTree is called a CFNode.
+
+    The CFNode can have a maximum of branching_factor
+    number of CFSubclusters.
+
+    Parameters
+    ----------
+    threshold : float
+        Threshold needed for a new subcluster to enter a CFSubcluster.
+
+    branching_factor : int
+        Maximum number of CF subclusters in each node.
+
+    is_leaf : bool
+        We need to know if the CFNode is a leaf or not, in order to
+        retrieve the final subclusters.
+
+    n_features : int
+        The number of features.
+
+    Attributes
+    ----------
+    subclusters_ : list
+        List of subclusters for a particular CFNode.
+
+    prev_leaf_ : _CFNode
+        Useful only if is_leaf is True.
+
+    next_leaf_ : _CFNode
+        next_leaf. Useful only if is_leaf is True.
+        the final subclusters.
+
+    init_centroids_ : ndarray of shape (branching_factor + 1, n_features)
+        Manipulate ``init_centroids_`` throughout rather than centroids_ since
+        the centroids are just a view of the ``init_centroids_`` .
+
+    init_sq_norm_ : ndarray of shape (branching_factor + 1,)
+        manipulate init_sq_norm_ throughout. similar to ``init_centroids_``.
+
+    centroids_ : ndarray of shape (branching_factor + 1, n_features)
+        View of ``init_centroids_``.
+
+    squared_norm_ : ndarray of shape (branching_factor + 1,)
+        View of ``init_sq_norm_``.
+
+    """
+
+    def __init__(self, *, threshold, branching_factor, is_leaf, n_features, dtype):
+        self.threshold = threshold
+        self.branching_factor = branching_factor
+        self.is_leaf = is_leaf
+        self.n_features = n_features
+
+        # The list of subclusters, centroids and squared norms
+        # to manipulate throughout.
+        self.subclusters_ = []
+        self.init_centroids_ = np.zeros((branching_factor + 1, n_features), dtype=dtype)
+        self.init_sq_norm_ = np.zeros((branching_factor + 1), dtype)
+        self.squared_norm_ = []
+        self.prev_leaf_ = None
+        self.next_leaf_ = None
+
+    def append_subcluster(self, subcluster):
+        n_samples = len(self.subclusters_)
+        self.subclusters_.append(subcluster)
+        self.init_centroids_[n_samples] = subcluster.centroid_
+        self.init_sq_norm_[n_samples] = subcluster.sq_norm_
+
+        # Keep centroids and squared norm as views. In this way
+        # if we change init_centroids and init_sq_norm_, it is
+        # sufficient,
+        self.centroids_ = self.init_centroids_[: n_samples + 1, :]
+        self.squared_norm_ = self.init_sq_norm_[: n_samples + 1]
+
+    def update_split_subclusters(self, subcluster, new_subcluster1, new_subcluster2):
+        """Remove a subcluster from a node and update it with the
+        split subclusters.
+        """
+        ind = self.subclusters_.index(subcluster)
+        self.subclusters_[ind] = new_subcluster1
+        self.init_centroids_[ind] = new_subcluster1.centroid_
+        self.init_sq_norm_[ind] = new_subcluster1.sq_norm_
+        self.append_subcluster(new_subcluster2)
+
+    def insert_cf_subcluster(self, subcluster):
+        """Insert a new subcluster into the node."""
+        if not self.subclusters_:
+            self.append_subcluster(subcluster)
+            return False
+
+        threshold = self.threshold
+        branching_factor = self.branching_factor
+        # We need to find the closest subcluster among all the
+        # subclusters so that we can insert our new subcluster.
+        dist_matrix = np.dot(self.centroids_, subcluster.centroid_)
+        dist_matrix *= -2.0
+        dist_matrix += self.squared_norm_
+        closest_index = np.argmin(dist_matrix)
+        closest_subcluster = self.subclusters_[closest_index]
+
+        # If the subcluster has a child, we need a recursive strategy.
+        if closest_subcluster.child_ is not None:
+            split_child = closest_subcluster.child_.insert_cf_subcluster(subcluster)
+
+            if not split_child:
+                # If it is determined that the child need not be split, we
+                # can just update the closest_subcluster
+                closest_subcluster.update(subcluster)
+                self.init_centroids_[closest_index] = self.subclusters_[
+                    closest_index
+                ].centroid_
+                self.init_sq_norm_[closest_index] = self.subclusters_[
+                    closest_index
+                ].sq_norm_
+                return False
+
+            # things not too good. we need to redistribute the subclusters in
+            # our child node, and add a new subcluster in the parent
+            # subcluster to accommodate the new child.
+            else:
+                new_subcluster1, new_subcluster2 = _split_node(
+                    closest_subcluster.child_,
+                    threshold,
+                    branching_factor,
+                )
+                self.update_split_subclusters(
+                    closest_subcluster, new_subcluster1, new_subcluster2
+                )
+
+                if len(self.subclusters_) > self.branching_factor:
+                    return True
+                return False
+
+        # good to go!
+        else:
+            merged = closest_subcluster.merge_subcluster(subcluster, self.threshold)
+            if merged:
+                self.init_centroids_[closest_index] = closest_subcluster.centroid_
+                self.init_sq_norm_[closest_index] = closest_subcluster.sq_norm_
+                return False
+
+            # not close to any other subclusters, and we still
+            # have space, so add.
+            elif len(self.subclusters_) < self.branching_factor:
+                self.append_subcluster(subcluster)
+                return False
+
+            # We do not have enough space nor is it closer to an
+            # other subcluster. We need to split.
+            else:
+                self.append_subcluster(subcluster)
+                return True
+
+
+class _CFSubcluster:
+    """Each subcluster in a CFNode is called a CFSubcluster.
+
+    A CFSubcluster can have a CFNode has its child.
+
+    Parameters
+    ----------
+    linear_sum : ndarray of shape (n_features,), default=None
+        Sample. This is kept optional to allow initialization of empty
+        subclusters.
+
+    Attributes
+    ----------
+    n_samples_ : int
+        Number of samples that belong to each subcluster.
+
+    linear_sum_ : ndarray
+        Linear sum of all the samples in a subcluster. Prevents holding
+        all sample data in memory.
+
+    squared_sum_ : float
+        Sum of the squared l2 norms of all samples belonging to a subcluster.
+
+    centroid_ : ndarray of shape (branching_factor + 1, n_features)
+        Centroid of the subcluster. Prevent recomputing of centroids when
+        ``CFNode.centroids_`` is called.
+
+    child_ : _CFNode
+        Child Node of the subcluster. Once a given _CFNode is set as the child
+        of the _CFNode, it is set to ``self.child_``.
+
+    sq_norm_ : ndarray of shape (branching_factor + 1,)
+        Squared norm of the subcluster. Used to prevent recomputing when
+        pairwise minimum distances are computed.
+    """
+
+    def __init__(self, *, linear_sum=None):
+        if linear_sum is None:
+            self.n_samples_ = 0
+            self.squared_sum_ = 0.0
+            self.centroid_ = self.linear_sum_ = 0
+        else:
+            self.n_samples_ = 1
+            self.centroid_ = self.linear_sum_ = linear_sum
+            self.squared_sum_ = self.sq_norm_ = np.dot(
+                self.linear_sum_, self.linear_sum_
+            )
+        self.child_ = None
+
+    def update(self, subcluster):
+        self.n_samples_ += subcluster.n_samples_
+        self.linear_sum_ += subcluster.linear_sum_
+        self.squared_sum_ += subcluster.squared_sum_
+        self.centroid_ = self.linear_sum_ / self.n_samples_
+        self.sq_norm_ = np.dot(self.centroid_, self.centroid_)
+
+    def merge_subcluster(self, nominee_cluster, threshold):
+        """Check if a cluster is worthy enough to be merged. If
+        yes then merge.
+        """
+        new_ss = self.squared_sum_ + nominee_cluster.squared_sum_
+        new_ls = self.linear_sum_ + nominee_cluster.linear_sum_
+        new_n = self.n_samples_ + nominee_cluster.n_samples_
+        new_centroid = (1 / new_n) * new_ls
+        new_sq_norm = np.dot(new_centroid, new_centroid)
+
+        # The squared radius of the cluster is defined:
+        #   r^2  = sum_i ||x_i - c||^2 / n
+        # with x_i the n points assigned to the cluster and c its centroid:
+        #   c = sum_i x_i / n
+        # This can be expanded to:
+        #   r^2 = sum_i ||x_i||^2 / n - 2 < sum_i x_i / n, c> + n ||c||^2 / n
+        # and therefore simplifies to:
+        #   r^2 = sum_i ||x_i||^2 / n - ||c||^2
+        sq_radius = new_ss / new_n - new_sq_norm
+
+        if sq_radius <= threshold**2:
+            (
+                self.n_samples_,
+                self.linear_sum_,
+                self.squared_sum_,
+                self.centroid_,
+                self.sq_norm_,
+            ) = (new_n, new_ls, new_ss, new_centroid, new_sq_norm)
+            return True
+        return False
+
+    @property
+    def radius(self):
+        """Return radius of the subcluster"""
+        # Because of numerical issues, this could become negative
+        sq_radius = self.squared_sum_ / self.n_samples_ - self.sq_norm_
+        return sqrt(max(0, sq_radius))
+
+
+class Birch(
+    ClassNamePrefixFeaturesOutMixin, ClusterMixin, TransformerMixin, BaseEstimator
+):
+    """Implements the BIRCH clustering algorithm.
+
+    It is a memory-efficient, online-learning algorithm provided as an
+    alternative to :class:`MiniBatchKMeans`. It constructs a tree
+    data structure with the cluster centroids being read off the leaf.
+    These can be either the final cluster centroids or can be provided as input
+    to another clustering algorithm such as :class:`AgglomerativeClustering`.
+
+    Read more in the :ref:`User Guide <birch>`.
+
+    .. versionadded:: 0.16
+
+    Parameters
+    ----------
+    threshold : float, default=0.5
+        The radius of the subcluster obtained by merging a new sample and the
+        closest subcluster should be lesser than the threshold. Otherwise a new
+        subcluster is started. Setting this value to be very low promotes
+        splitting and vice-versa.
+
+    branching_factor : int, default=50
+        Maximum number of CF subclusters in each node. If a new samples enters
+        such that the number of subclusters exceed the branching_factor then
+        that node is split into two nodes with the subclusters redistributed
+        in each. The parent subcluster of that node is removed and two new
+        subclusters are added as parents of the 2 split nodes.
+
+    n_clusters : int, instance of sklearn.cluster model or None, default=3
+        Number of clusters after the final clustering step, which treats the
+        subclusters from the leaves as new samples.
+
+        - `None` : the final clustering step is not performed and the
+          subclusters are returned as they are.
+
+        - :mod:`sklearn.cluster` Estimator : If a model is provided, the model
+          is fit treating the subclusters as new samples and the initial data
+          is mapped to the label of the closest subcluster.
+
+        - `int` : the model fit is :class:`AgglomerativeClustering` with
+          `n_clusters` set to be equal to the int.
+
+    compute_labels : bool, default=True
+        Whether or not to compute labels for each fit.
+
+    copy : bool, default=True
+        Whether or not to make a copy of the given data. If set to False,
+        the initial data will be overwritten.
+
+    Attributes
+    ----------
+    root_ : _CFNode
+        Root of the CFTree.
+
+    dummy_leaf_ : _CFNode
+        Start pointer to all the leaves.
+
+    subcluster_centers_ : ndarray
+        Centroids of all subclusters read directly from the leaves.
+
+    subcluster_labels_ : ndarray
+        Labels assigned to the centroids of the subclusters after
+        they are clustered globally.
+
+    labels_ : ndarray of shape (n_samples,)
+        Array of labels assigned to the input data.
+        if partial_fit is used instead of fit, they are assigned to the
+        last batch of data.
+
+    n_features_in_ : int
+        Number of features seen during :term:`fit`.
+
+        .. versionadded:: 0.24
+
+    feature_names_in_ : ndarray of shape (`n_features_in_`,)
+        Names of features seen during :term:`fit`. Defined only when `X`
+        has feature names that are all strings.
+
+        .. versionadded:: 1.0
+
+    See Also
+    --------
+    MiniBatchKMeans : Alternative implementation that does incremental updates
+        of the centers' positions using mini-batches.
+
+    Notes
+    -----
+    The tree data structure consists of nodes with each node consisting of
+    a number of subclusters. The maximum number of subclusters in a node
+    is determined by the branching factor. Each subcluster maintains a
+    linear sum, squared sum and the number of samples in that subcluster.
+    In addition, each subcluster can also have a node as its child, if the
+    subcluster is not a member of a leaf node.
+
+    For a new point entering the root, it is merged with the subcluster closest
+    to it and the linear sum, squared sum and the number of samples of that
+    subcluster are updated. This is done recursively till the properties of
+    the leaf node are updated.
+
+    References
+    ----------
+    * Tian Zhang, Raghu Ramakrishnan, Maron Livny
+      BIRCH: An efficient data clustering method for large databases.
+      https://www.cs.sfu.ca/CourseCentral/459/han/papers/zhang96.pdf
+
+    * Roberto Perdisci
+      JBirch - Java implementation of BIRCH clustering algorithm
+      https://code.google.com/archive/p/jbirch
+
+    Examples
+    --------
+    >>> from sklearn.cluster import Birch
+    >>> X = [[0, 1], [0.3, 1], [-0.3, 1], [0, -1], [0.3, -1], [-0.3, -1]]
+    >>> brc = Birch(n_clusters=None)
+    >>> brc.fit(X)
+    Birch(n_clusters=None)
+    >>> brc.predict(X)
+    array([0, 0, 0, 1, 1, 1])
+    """
+
+    _parameter_constraints: dict = {
+        "threshold": [Interval(Real, 0.0, None, closed="neither")],
+        "branching_factor": [Interval(Integral, 1, None, closed="neither")],
+        "n_clusters": [None, ClusterMixin, Interval(Integral, 1, None, closed="left")],
+        "compute_labels": ["boolean"],
+        "copy": ["boolean"],
+    }
+
+    def __init__(
+        self,
+        *,
+        threshold=0.5,
+        branching_factor=50,
+        n_clusters=3,
+        compute_labels=True,
+        copy=True,
+    ):
+        self.threshold = threshold
+        self.branching_factor = branching_factor
+        self.n_clusters = n_clusters
+        self.compute_labels = compute_labels
+        self.copy = copy
+
+    @_fit_context(prefer_skip_nested_validation=True)
+    def fit(self, X, y=None):
+        """
+        Build a CF Tree for the input data.
+
+        Parameters
+        ----------
+        X : {array-like, sparse matrix} of shape (n_samples, n_features)
+            Input data.
+
+        y : Ignored
+            Not used, present here for API consistency by convention.
+
+        Returns
+        -------
+        self
+            Fitted estimator.
+        """
+        return self._fit(X, partial=False)
+
+    def _fit(self, X, partial):
+        has_root = getattr(self, "root_", None)
+        first_call = not (partial and has_root)
+
+        X = self._validate_data(
+            X,
+            accept_sparse="csr",
+            copy=self.copy,
+            reset=first_call,
+            dtype=[np.float64, np.float32],
+        )
+        threshold = self.threshold
+        branching_factor = self.branching_factor
+
+        n_samples, n_features = X.shape
+
+        # If partial_fit is called for the first time or fit is called, we
+        # start a new tree.
+        if first_call:
+            # The first root is the leaf. Manipulate this object throughout.
+            self.root_ = _CFNode(
+                threshold=threshold,
+                branching_factor=branching_factor,
+                is_leaf=True,
+                n_features=n_features,
+                dtype=X.dtype,
+            )
+
+            # To enable getting back subclusters.
+            self.dummy_leaf_ = _CFNode(
+                threshold=threshold,
+                branching_factor=branching_factor,
+                is_leaf=True,
+                n_features=n_features,
+                dtype=X.dtype,
+            )
+            self.dummy_leaf_.next_leaf_ = self.root_
+            self.root_.prev_leaf_ = self.dummy_leaf_
+
+        # Cannot vectorize. Enough to convince to use cython.
+        if not sparse.issparse(X):
+            iter_func = iter
+        else:
+            iter_func = _iterate_sparse_X
+
+        for sample in iter_func(X):
+            subcluster = _CFSubcluster(linear_sum=sample)
+            split = self.root_.insert_cf_subcluster(subcluster)
+
+            if split:
+                new_subcluster1, new_subcluster2 = _split_node(
+                    self.root_, threshold, branching_factor
+                )
+                del self.root_
+                self.root_ = _CFNode(
+                    threshold=threshold,
+                    branching_factor=branching_factor,
+                    is_leaf=False,
+                    n_features=n_features,
+                    dtype=X.dtype,
+                )
+                self.root_.append_subcluster(new_subcluster1)
+                self.root_.append_subcluster(new_subcluster2)
+
+        centroids = np.concatenate([leaf.centroids_ for leaf in self._get_leaves()])
+        self.subcluster_centers_ = centroids
+        self._n_features_out = self.subcluster_centers_.shape[0]
+
+        self._global_clustering(X)
+        return self
+
+    def _get_leaves(self):
+        """
+        Retrieve the leaves of the CF Node.
+
+        Returns
+        -------
+        leaves : list of shape (n_leaves,)
+            List of the leaf nodes.
+        """
+        leaf_ptr = self.dummy_leaf_.next_leaf_
+        leaves = []
+        while leaf_ptr is not None:
+            leaves.append(leaf_ptr)
+            leaf_ptr = leaf_ptr.next_leaf_
+        return leaves
+
+    @_fit_context(prefer_skip_nested_validation=True)
+    def partial_fit(self, X=None, y=None):
+        """
+        Online learning. Prevents rebuilding of CFTree from scratch.
+
+        Parameters
+        ----------
+        X : {array-like, sparse matrix} of shape (n_samples, n_features), \
+            default=None
+            Input data. If X is not provided, only the global clustering
+            step is done.
+
+        y : Ignored
+            Not used, present here for API consistency by convention.
+
+        Returns
+        -------
+        self
+            Fitted estimator.
+        """
+        if X is None:
+            # Perform just the final global clustering step.
+            self._global_clustering()
+            return self
+        else:
+            return self._fit(X, partial=True)
+
+    def _check_fit(self, X):
+        check_is_fitted(self)
+
+        if (
+            hasattr(self, "subcluster_centers_")
+            and X.shape[1] != self.subcluster_centers_.shape[1]
+        ):
+            raise ValueError(
+                "Training data and predicted data do not have same number of features."
+            )
+
+    def predict(self, X):
+        """
+        Predict data using the ``centroids_`` of subclusters.
+
+        Avoid computation of the row norms of X.
+
+        Parameters
+        ----------
+        X : {array-like, sparse matrix} of shape (n_samples, n_features)
+            Input data.
+
+        Returns
+        -------
+        labels : ndarray of shape(n_samples,)
+            Labelled data.
+        """
+        check_is_fitted(self)
+        X = self._validate_data(X, accept_sparse="csr", reset=False)
+        return self._predict(X)
+
+    def _predict(self, X):
+        """Predict data using the ``centroids_`` of subclusters."""
+        kwargs = {"Y_norm_squared": self._subcluster_norms}
+
+        with config_context(assume_finite=True):
+            argmin = pairwise_distances_argmin(
+                X, self.subcluster_centers_, metric_kwargs=kwargs
+            )
+        return self.subcluster_labels_[argmin]
+
+    def transform(self, X):
+        """
+        Transform X into subcluster centroids dimension.
+
+        Each dimension represents the distance from the sample point to each
+        cluster centroid.
+
+        Parameters
+        ----------
+        X : {array-like, sparse matrix} of shape (n_samples, n_features)
+            Input data.
+
+        Returns
+        -------
+        X_trans : {array-like, sparse matrix} of shape (n_samples, n_clusters)
+            Transformed data.
+        """
+        check_is_fitted(self)
+        X = self._validate_data(X, accept_sparse="csr", reset=False)
+        with config_context(assume_finite=True):
+            return euclidean_distances(X, self.subcluster_centers_)
+
+    def _global_clustering(self, X=None):
+        """
+        Global clustering for the subclusters obtained after fitting
+        """
+        clusterer = self.n_clusters
+        centroids = self.subcluster_centers_
+        compute_labels = (X is not None) and self.compute_labels
+
+        # Preprocessing for the global clustering.
+        not_enough_centroids = False
+        if isinstance(clusterer, Integral):
+            clusterer = AgglomerativeClustering(n_clusters=self.n_clusters)
+            # There is no need to perform the global clustering step.
+            if len(centroids) < self.n_clusters:
+                not_enough_centroids = True
+
+        # To use in predict to avoid recalculation.
+        self._subcluster_norms = row_norms(self.subcluster_centers_, squared=True)
+
+        if clusterer is None or not_enough_centroids:
+            self.subcluster_labels_ = np.arange(len(centroids))
+            if not_enough_centroids:
+                warnings.warn(
+                    "Number of subclusters found (%d) by BIRCH is less "
+                    "than (%d). Decrease the threshold."
+                    % (len(centroids), self.n_clusters),
+                    ConvergenceWarning,
+                )
+        else:
+            # The global clustering step that clusters the subclusters of
+            # the leaves. It assumes the centroids of the subclusters as
+            # samples and finds the final centroids.
+            self.subcluster_labels_ = clusterer.fit_predict(self.subcluster_centers_)
+
+        if compute_labels:
+            self.labels_ = self._predict(X)
+
+    def _more_tags(self):
+        return {"preserves_dtype": [np.float64, np.float32]}

.venv/lib/python3.12/site-packages/sklearn/cluster/_bisect_k_means.py

@@ -0,0 +1,530 @@
+"""Bisecting K-means clustering."""
+
+# Author: Michal Krawczyk <mkrwczyk.1@gmail.com>
+
+import warnings
+
+import numpy as np
+import scipy.sparse as sp
+
+from ..base import _fit_context
+from ..utils._openmp_helpers import _openmp_effective_n_threads
+from ..utils._param_validation import Integral, Interval, StrOptions
+from ..utils.extmath import row_norms
+from ..utils.validation import _check_sample_weight, check_is_fitted, check_random_state
+from ._k_means_common import _inertia_dense, _inertia_sparse
+from ._kmeans import (
+    _BaseKMeans,
+    _kmeans_single_elkan,
+    _kmeans_single_lloyd,
+    _labels_inertia_threadpool_limit,
+)
+
+
+class _BisectingTree:
+    """Tree structure representing the hierarchical clusters of BisectingKMeans."""
+
+    def __init__(self, center, indices, score):
+        """Create a new cluster node in the tree.
+
+        The node holds the center of this cluster and the indices of the data points
+        that belong to it.
+        """
+        self.center = center
+        self.indices = indices
+        self.score = score
+
+        self.left = None
+        self.right = None
+
+    def split(self, labels, centers, scores):
+        """Split the cluster node into two subclusters."""
+        self.left = _BisectingTree(
+            indices=self.indices[labels == 0], center=centers[0], score=scores[0]
+        )
+        self.right = _BisectingTree(
+            indices=self.indices[labels == 1], center=centers[1], score=scores[1]
+        )
+
+        # reset the indices attribute to save memory
+        self.indices = None
+
+    def get_cluster_to_bisect(self):
+        """Return the cluster node to bisect next.
+
+        It's based on the score of the cluster, which can be either the number of
+        data points assigned to that cluster or the inertia of that cluster
+        (see `bisecting_strategy` for details).
+        """
+        max_score = None
+
+        for cluster_leaf in self.iter_leaves():
+            if max_score is None or cluster_leaf.score > max_score:
+                max_score = cluster_leaf.score
+                best_cluster_leaf = cluster_leaf
+
+        return best_cluster_leaf
+
+    def iter_leaves(self):
+        """Iterate over all the cluster leaves in the tree."""
+        if self.left is None:
+            yield self
+        else:
+            yield from self.left.iter_leaves()
+            yield from self.right.iter_leaves()
+
+
+class BisectingKMeans(_BaseKMeans):
+    """Bisecting K-Means clustering.
+
+    Read more in the :ref:`User Guide <bisect_k_means>`.
+
+    .. versionadded:: 1.1
+
+    Parameters
+    ----------
+    n_clusters : int, default=8
+        The number of clusters to form as well as the number of
+        centroids to generate.
+
+    init : {'k-means++', 'random'} or callable, default='random'
+        Method for initialization:
+
+        'k-means++' : selects initial cluster centers for k-mean
+        clustering in a smart way to speed up convergence. See section
+        Notes in k_init for more details.
+
+        'random': choose `n_clusters` observations (rows) at random from data
+        for the initial centroids.
+
+        If a callable is passed, it should take arguments X, n_clusters and a
+        random state and return an initialization.
+
+    n_init : int, default=1
+        Number of time the inner k-means algorithm will be run with different
+        centroid seeds in each bisection.
+        That will result producing for each bisection best output of n_init
+        consecutive runs in terms of inertia.
+
+    random_state : int, RandomState instance or None, default=None
+        Determines random number generation for centroid initialization
+        in inner K-Means. Use an int to make the randomness deterministic.
+        See :term:`Glossary <random_state>`.
+
+    max_iter : int, default=300
+        Maximum number of iterations of the inner k-means algorithm at each
+        bisection.
+
+    verbose : int, default=0
+        Verbosity mode.
+
+    tol : float, default=1e-4
+        Relative tolerance with regards to Frobenius norm of the difference
+        in the cluster centers of two consecutive iterations  to declare
+        convergence. Used in inner k-means algorithm at each bisection to pick
+        best possible clusters.
+
+    copy_x : bool, default=True
+        When pre-computing distances it is more numerically accurate to center
+        the data first. If copy_x is True (default), then the original data is
+        not modified. If False, the original data is modified, and put back
+        before the function returns, but small numerical differences may be
+        introduced by subtracting and then adding the data mean. Note that if
+        the original data is not C-contiguous, a copy will be made even if
+        copy_x is False. If the original data is sparse, but not in CSR format,
+        a copy will be made even if copy_x is False.
+
+    algorithm : {"lloyd", "elkan"}, default="lloyd"
+        Inner K-means algorithm used in bisection.
+        The classical EM-style algorithm is `"lloyd"`.
+        The `"elkan"` variation can be more efficient on some datasets with
+        well-defined clusters, by using the triangle inequality. However it's
+        more memory intensive due to the allocation of an extra array of shape
+        `(n_samples, n_clusters)`.
+
+    bisecting_strategy : {"biggest_inertia", "largest_cluster"},\
+            default="biggest_inertia"
+        Defines how bisection should be performed:
+
+         - "biggest_inertia" means that BisectingKMeans will always check
+            all calculated cluster for cluster with biggest SSE
+            (Sum of squared errors) and bisect it. This approach concentrates on
+            precision, but may be costly in terms of execution time (especially for
+            larger amount of data points).
+
+         - "largest_cluster" - BisectingKMeans will always split cluster with
+            largest amount of points assigned to it from all clusters
+            previously calculated. That should work faster than picking by SSE
+            ('biggest_inertia') and may produce similar results in most cases.
+
+    Attributes
+    ----------
+    cluster_centers_ : ndarray of shape (n_clusters, n_features)
+        Coordinates of cluster centers. If the algorithm stops before fully
+        converging (see ``tol`` and ``max_iter``), these will not be
+        consistent with ``labels_``.
+
+    labels_ : ndarray of shape (n_samples,)
+        Labels of each point.
+
+    inertia_ : float
+        Sum of squared distances of samples to their closest cluster center,
+        weighted by the sample weights if provided.
+
+    n_features_in_ : int
+        Number of features seen during :term:`fit`.
+
+    feature_names_in_ : ndarray of shape (`n_features_in_`,)
+        Names of features seen during :term:`fit`. Defined only when `X`
+        has feature names that are all strings.
+
+    See Also
+    --------
+    KMeans : Original implementation of K-Means algorithm.
+
+    Notes
+    -----
+    It might be inefficient when n_cluster is less than 3, due to unnecessary
+    calculations for that case.
+
+    Examples
+    --------
+    >>> from sklearn.cluster import BisectingKMeans
+    >>> import numpy as np
+    >>> X = np.array([[1, 1], [10, 1], [3, 1],
+    ...               [10, 0], [2, 1], [10, 2],
+    ...               [10, 8], [10, 9], [10, 10]])
+    >>> bisect_means = BisectingKMeans(n_clusters=3, random_state=0).fit(X)
+    >>> bisect_means.labels_
+    array([0, 2, 0, 2, 0, 2, 1, 1, 1], dtype=int32)
+    >>> bisect_means.predict([[0, 0], [12, 3]])
+    array([0, 2], dtype=int32)
+    >>> bisect_means.cluster_centers_
+    array([[ 2., 1.],
+           [10., 9.],
+           [10., 1.]])
+    """
+
+    _parameter_constraints: dict = {
+        **_BaseKMeans._parameter_constraints,
+        "init": [StrOptions({"k-means++", "random"}), callable],
+        "n_init": [Interval(Integral, 1, None, closed="left")],
+        "copy_x": ["boolean"],
+        "algorithm": [StrOptions({"lloyd", "elkan"})],
+        "bisecting_strategy": [StrOptions({"biggest_inertia", "largest_cluster"})],
+    }
+
+    def __init__(
+        self,
+        n_clusters=8,
+        *,
+        init="random",
+        n_init=1,
+        random_state=None,
+        max_iter=300,
+        verbose=0,
+        tol=1e-4,
+        copy_x=True,
+        algorithm="lloyd",
+        bisecting_strategy="biggest_inertia",
+    ):
+        super().__init__(
+            n_clusters=n_clusters,
+            init=init,
+            max_iter=max_iter,
+            verbose=verbose,
+            random_state=random_state,
+            tol=tol,
+            n_init=n_init,
+        )
+
+        self.copy_x = copy_x
+        self.algorithm = algorithm
+        self.bisecting_strategy = bisecting_strategy
+
+    def _warn_mkl_vcomp(self, n_active_threads):
+        """Warn when vcomp and mkl are both present"""
+        warnings.warn(
+            "BisectingKMeans is known to have a memory leak on Windows "
+            "with MKL, when there are less chunks than available "
+            "threads. You can avoid it by setting the environment"
+            f" variable OMP_NUM_THREADS={n_active_threads}."
+        )
+
+    def _inertia_per_cluster(self, X, centers, labels, sample_weight):
+        """Calculate the sum of squared errors (inertia) per cluster.
+
+        Parameters
+        ----------
+        X : {ndarray, csr_matrix} of shape (n_samples, n_features)
+            The input samples.
+
+        centers : ndarray of shape (n_clusters=2, n_features)
+            The cluster centers.
+
+        labels : ndarray of shape (n_samples,)
+            Index of the cluster each sample belongs to.
+
+        sample_weight : ndarray of shape (n_samples,)
+            The weights for each observation in X.
+
+        Returns
+        -------
+        inertia_per_cluster : ndarray of shape (n_clusters=2,)
+            Sum of squared errors (inertia) for each cluster.
+        """
+        n_clusters = centers.shape[0]  # = 2 since centers comes from a bisection
+        _inertia = _inertia_sparse if sp.issparse(X) else _inertia_dense
+
+        inertia_per_cluster = np.empty(n_clusters)
+        for label in range(n_clusters):
+            inertia_per_cluster[label] = _inertia(
+                X, sample_weight, centers, labels, self._n_threads, single_label=label
+            )
+
+        return inertia_per_cluster
+
+    def _bisect(self, X, x_squared_norms, sample_weight, cluster_to_bisect):
+        """Split a cluster into 2 subsclusters.
+
+        Parameters
+        ----------
+        X : {ndarray, csr_matrix} of shape (n_samples, n_features)
+            Training instances to cluster.
+
+        x_squared_norms : ndarray of shape (n_samples,)
+            Squared euclidean norm of each data point.
+
+        sample_weight : ndarray of shape (n_samples,)
+            The weights for each observation in X.
+
+        cluster_to_bisect : _BisectingTree node object
+            The cluster node to split.
+        """
+        X = X[cluster_to_bisect.indices]
+        x_squared_norms = x_squared_norms[cluster_to_bisect.indices]
+        sample_weight = sample_weight[cluster_to_bisect.indices]
+
+        best_inertia = None
+
+        # Split samples in X into 2 clusters.
+        # Repeating `n_init` times to obtain best clusters
+        for _ in range(self.n_init):
+            centers_init = self._init_centroids(
+                X,
+                x_squared_norms=x_squared_norms,
+                init=self.init,
+                random_state=self._random_state,
+                n_centroids=2,
+                sample_weight=sample_weight,
+            )
+
+            labels, inertia, centers, _ = self._kmeans_single(
+                X,
+                sample_weight,
+                centers_init,
+                max_iter=self.max_iter,
+                verbose=self.verbose,
+                tol=self.tol,
+                n_threads=self._n_threads,
+            )
+
+            # allow small tolerance on the inertia to accommodate for
+            # non-deterministic rounding errors due to parallel computation
+            if best_inertia is None or inertia < best_inertia * (1 - 1e-6):
+                best_labels = labels
+                best_centers = centers
+                best_inertia = inertia
+
+        if self.verbose:
+            print(f"New centroids from bisection: {best_centers}")
+
+        if self.bisecting_strategy == "biggest_inertia":
+            scores = self._inertia_per_cluster(
+                X, best_centers, best_labels, sample_weight
+            )
+        else:  # bisecting_strategy == "largest_cluster"
+            # Using minlength to make sure that we have the counts for both labels even
+            # if all samples are labelled 0.
+            scores = np.bincount(best_labels, minlength=2)
+
+        cluster_to_bisect.split(best_labels, best_centers, scores)
+
+    @_fit_context(prefer_skip_nested_validation=True)
+    def fit(self, X, y=None, sample_weight=None):
+        """Compute bisecting k-means clustering.
+
+        Parameters
+        ----------
+        X : {array-like, sparse matrix} of shape (n_samples, n_features)
+
+            Training instances to cluster.
+
+            .. note:: The data will be converted to C ordering,
+                which will cause a memory copy
+                if the given data is not C-contiguous.
+
+        y : Ignored
+            Not used, present here for API consistency by convention.
+
+        sample_weight : array-like of shape (n_samples,), default=None
+            The weights for each observation in X. If None, all observations
+            are assigned equal weight. `sample_weight` is not used during
+            initialization if `init` is a callable.
+
+        Returns
+        -------
+        self
+            Fitted estimator.
+        """
+        X = self._validate_data(
+            X,
+            accept_sparse="csr",
+            dtype=[np.float64, np.float32],
+            order="C",
+            copy=self.copy_x,
+            accept_large_sparse=False,
+        )
+
+        self._check_params_vs_input(X)
+
+        self._random_state = check_random_state(self.random_state)
+        sample_weight = _check_sample_weight(sample_weight, X, dtype=X.dtype)
+        self._n_threads = _openmp_effective_n_threads()
+
+        if self.algorithm == "lloyd" or self.n_clusters == 1:
+            self._kmeans_single = _kmeans_single_lloyd
+            self._check_mkl_vcomp(X, X.shape[0])
+        else:
+            self._kmeans_single = _kmeans_single_elkan
+
+        # Subtract of mean of X for more accurate distance computations
+        if not sp.issparse(X):
+            self._X_mean = X.mean(axis=0)
+            X -= self._X_mean
+
+        # Initialize the hierarchical clusters tree
+        self._bisecting_tree = _BisectingTree(
+            indices=np.arange(X.shape[0]),
+            center=X.mean(axis=0),
+            score=0,
+        )
+
+        x_squared_norms = row_norms(X, squared=True)
+
+        for _ in range(self.n_clusters - 1):
+            # Chose cluster to bisect
+            cluster_to_bisect = self._bisecting_tree.get_cluster_to_bisect()
+
+            # Split this cluster into 2 subclusters
+            self._bisect(X, x_squared_norms, sample_weight, cluster_to_bisect)
+
+        # Aggregate final labels and centers from the bisecting tree
+        self.labels_ = np.full(X.shape[0], -1, dtype=np.int32)
+        self.cluster_centers_ = np.empty((self.n_clusters, X.shape[1]), dtype=X.dtype)
+
+        for i, cluster_node in enumerate(self._bisecting_tree.iter_leaves()):
+            self.labels_[cluster_node.indices] = i
+            self.cluster_centers_[i] = cluster_node.center
+            cluster_node.label = i  # label final clusters for future prediction
+            cluster_node.indices = None  # release memory
+
+        # Restore original data
+        if not sp.issparse(X):
+            X += self._X_mean
+            self.cluster_centers_ += self._X_mean
+
+        _inertia = _inertia_sparse if sp.issparse(X) else _inertia_dense
+        self.inertia_ = _inertia(
+            X, sample_weight, self.cluster_centers_, self.labels_, self._n_threads
+        )
+
+        self._n_features_out = self.cluster_centers_.shape[0]
+
+        return self
+
+    def predict(self, X):
+        """Predict which cluster each sample in X belongs to.
+
+        Prediction is made by going down the hierarchical tree
+        in searching of closest leaf cluster.
+
+        In the vector quantization literature, `cluster_centers_` is called
+        the code book and each value returned by `predict` is the index of
+        the closest code in the code book.
+
+        Parameters
+        ----------
+        X : {array-like, sparse matrix} of shape (n_samples, n_features)
+            New data to predict.
+
+        Returns
+        -------
+        labels : ndarray of shape (n_samples,)
+            Index of the cluster each sample belongs to.
+        """
+        check_is_fitted(self)
+
+        X = self._check_test_data(X)
+        x_squared_norms = row_norms(X, squared=True)
+
+        # sample weights are unused but necessary in cython helpers
+        sample_weight = np.ones_like(x_squared_norms)
+
+        labels = self._predict_recursive(X, sample_weight, self._bisecting_tree)
+
+        return labels
+
+    def _predict_recursive(self, X, sample_weight, cluster_node):
+        """Predict recursively by going down the hierarchical tree.
+
+        Parameters
+        ----------
+        X : {ndarray, csr_matrix} of shape (n_samples, n_features)
+            The data points, currently assigned to `cluster_node`, to predict between
+            the subclusters of this node.
+
+        sample_weight : ndarray of shape (n_samples,)
+            The weights for each observation in X.
+
+        cluster_node : _BisectingTree node object
+            The cluster node of the hierarchical tree.
+
+        Returns
+        -------
+        labels : ndarray of shape (n_samples,)
+            Index of the cluster each sample belongs to.
+        """
+        if cluster_node.left is None:
+            # This cluster has no subcluster. Labels are just the label of the cluster.
+            return np.full(X.shape[0], cluster_node.label, dtype=np.int32)
+
+        # Determine if data points belong to the left or right subcluster
+        centers = np.vstack((cluster_node.left.center, cluster_node.right.center))
+        if hasattr(self, "_X_mean"):
+            centers += self._X_mean
+
+        cluster_labels = _labels_inertia_threadpool_limit(
+            X,
+            sample_weight,
+            centers,
+            self._n_threads,
+            return_inertia=False,
+        )
+        mask = cluster_labels == 0
+
+        # Compute the labels for each subset of the data points.
+        labels = np.full(X.shape[0], -1, dtype=np.int32)
+
+        labels[mask] = self._predict_recursive(
+            X[mask], sample_weight[mask], cluster_node.left
+        )
+
+        labels[~mask] = self._predict_recursive(
+            X[~mask], sample_weight[~mask], cluster_node.right
+        )
+
+        return labels
+
+    def _more_tags(self):
+        return {"preserves_dtype": [np.float64, np.float32]}

.venv/lib/python3.12/site-packages/sklearn/cluster/_dbscan.py

@@ -0,0 +1,478 @@
+"""
+DBSCAN: Density-Based Spatial Clustering of Applications with Noise
+"""
+
+# Author: Robert Layton <robertlayton@gmail.com>
+#         Joel Nothman <joel.nothman@gmail.com>
+#         Lars Buitinck
+#
+# License: BSD 3 clause
+
+import warnings
+from numbers import Integral, Real
+
+import numpy as np
+from scipy import sparse
+
+from ..base import BaseEstimator, ClusterMixin, _fit_context
+from ..metrics.pairwise import _VALID_METRICS
+from ..neighbors import NearestNeighbors
+from ..utils._param_validation import Interval, StrOptions, validate_params
+from ..utils.validation import _check_sample_weight
+from ._dbscan_inner import dbscan_inner
+
+
+@validate_params(
+    {
+        "X": ["array-like", "sparse matrix"],
+        "sample_weight": ["array-like", None],
+    },
+    prefer_skip_nested_validation=False,
+)
+def dbscan(
+    X,
+    eps=0.5,
+    *,
+    min_samples=5,
+    metric="minkowski",
+    metric_params=None,
+    algorithm="auto",
+    leaf_size=30,
+    p=2,
+    sample_weight=None,
+    n_jobs=None,
+):
+    """Perform DBSCAN clustering from vector array or distance matrix.
+
+    Read more in the :ref:`User Guide <dbscan>`.
+
+    Parameters
+    ----------
+    X : {array-like, sparse (CSR) matrix} of shape (n_samples, n_features) or \
+            (n_samples, n_samples)
+        A feature array, or array of distances between samples if
+        ``metric='precomputed'``.
+
+    eps : float, default=0.5
+        The maximum distance between two samples for one to be considered
+        as in the neighborhood of the other. This is not a maximum bound
+        on the distances of points within a cluster. This is the most
+        important DBSCAN parameter to choose appropriately for your data set
+        and distance function.
+
+    min_samples : int, default=5
+        The number of samples (or total weight) in a neighborhood for a point
+        to be considered as a core point. This includes the point itself.
+
+    metric : str or callable, default='minkowski'
+        The metric to use when calculating distance between instances in a
+        feature array. If metric is a string or callable, it must be one of
+        the options allowed by :func:`sklearn.metrics.pairwise_distances` for
+        its metric parameter.
+        If metric is "precomputed", X is assumed to be a distance matrix and
+        must be square during fit.
+        X may be a :term:`sparse graph <sparse graph>`,
+        in which case only "nonzero" elements may be considered neighbors.
+
+    metric_params : dict, default=None
+        Additional keyword arguments for the metric function.
+
+        .. versionadded:: 0.19
+
+    algorithm : {'auto', 'ball_tree', 'kd_tree', 'brute'}, default='auto'
+        The algorithm to be used by the NearestNeighbors module
+        to compute pointwise distances and find nearest neighbors.
+        See NearestNeighbors module documentation for details.
+
+    leaf_size : int, default=30
+        Leaf size passed to BallTree or cKDTree. This can affect the speed
+        of the construction and query, as well as the memory required
+        to store the tree. The optimal value depends
+        on the nature of the problem.
+
+    p : float, default=2
+        The power of the Minkowski metric to be used to calculate distance
+        between points.
+
+    sample_weight : array-like of shape (n_samples,), default=None
+        Weight of each sample, such that a sample with a weight of at least
+        ``min_samples`` is by itself a core sample; a sample with negative
+        weight may inhibit its eps-neighbor from being core.
+        Note that weights are absolute, and default to 1.
+
+    n_jobs : int, default=None
+        The number of parallel jobs to run for neighbors search. ``None`` means
+        1 unless in a :obj:`joblib.parallel_backend` context. ``-1`` means
+        using all processors. See :term:`Glossary <n_jobs>` for more details.
+        If precomputed distance are used, parallel execution is not available
+        and thus n_jobs will have no effect.
+
+    Returns
+    -------
+    core_samples : ndarray of shape (n_core_samples,)
+        Indices of core samples.
+
+    labels : ndarray of shape (n_samples,)
+        Cluster labels for each point.  Noisy samples are given the label -1.
+
+    See Also
+    --------
+    DBSCAN : An estimator interface for this clustering algorithm.
+    OPTICS : A similar estimator interface clustering at multiple values of
+        eps. Our implementation is optimized for memory usage.
+
+    Notes
+    -----
+    For an example, see :ref:`examples/cluster/plot_dbscan.py
+    <sphx_glr_auto_examples_cluster_plot_dbscan.py>`.
+
+    This implementation bulk-computes all neighborhood queries, which increases
+    the memory complexity to O(n.d) where d is the average number of neighbors,
+    while original DBSCAN had memory complexity O(n). It may attract a higher
+    memory complexity when querying these nearest neighborhoods, depending
+    on the ``algorithm``.
+
+    One way to avoid the query complexity is to pre-compute sparse
+    neighborhoods in chunks using
+    :func:`NearestNeighbors.radius_neighbors_graph
+    <sklearn.neighbors.NearestNeighbors.radius_neighbors_graph>` with
+    ``mode='distance'``, then using ``metric='precomputed'`` here.
+
+    Another way to reduce memory and computation time is to remove
+    (near-)duplicate points and use ``sample_weight`` instead.
+
+    :class:`~sklearn.cluster.OPTICS` provides a similar clustering with lower
+    memory usage.
+
+    References
+    ----------
+    Ester, M., H. P. Kriegel, J. Sander, and X. Xu, `"A Density-Based
+    Algorithm for Discovering Clusters in Large Spatial Databases with Noise"
+    <https://www.dbs.ifi.lmu.de/Publikationen/Papers/KDD-96.final.frame.pdf>`_.
+    In: Proceedings of the 2nd International Conference on Knowledge Discovery
+    and Data Mining, Portland, OR, AAAI Press, pp. 226-231. 1996
+
+    Schubert, E., Sander, J., Ester, M., Kriegel, H. P., & Xu, X. (2017).
+    :doi:`"DBSCAN revisited, revisited: why and how you should (still) use DBSCAN."
+    <10.1145/3068335>`
+    ACM Transactions on Database Systems (TODS), 42(3), 19.
+
+    Examples
+    --------
+    >>> from sklearn.cluster import dbscan
+    >>> X = [[1, 2], [2, 2], [2, 3], [8, 7], [8, 8], [25, 80]]
+    >>> core_samples, labels = dbscan(X, eps=3, min_samples=2)
+    >>> core_samples
+    array([0, 1, 2, 3, 4])
+    >>> labels
+    array([ 0,  0,  0,  1,  1, -1])
+    """
+
+    est = DBSCAN(
+        eps=eps,
+        min_samples=min_samples,
+        metric=metric,
+        metric_params=metric_params,
+        algorithm=algorithm,
+        leaf_size=leaf_size,
+        p=p,
+        n_jobs=n_jobs,
+    )
+    est.fit(X, sample_weight=sample_weight)
+    return est.core_sample_indices_, est.labels_
+
+
+class DBSCAN(ClusterMixin, BaseEstimator):
+    """Perform DBSCAN clustering from vector array or distance matrix.
+
+    DBSCAN - Density-Based Spatial Clustering of Applications with Noise.
+    Finds core samples of high density and expands clusters from them.
+    Good for data which contains clusters of similar density.
+
+    This implementation has a worst case memory complexity of :math:`O({n}^2)`,
+    which can occur when the `eps` param is large and `min_samples` is low,
+    while the original DBSCAN only uses linear memory.
+    For further details, see the Notes below.
+
+    Read more in the :ref:`User Guide <dbscan>`.
+
+    Parameters
+    ----------
+    eps : float, default=0.5
+        The maximum distance between two samples for one to be considered
+        as in the neighborhood of the other. This is not a maximum bound
+        on the distances of points within a cluster. This is the most
+        important DBSCAN parameter to choose appropriately for your data set
+        and distance function.
+
+    min_samples : int, default=5
+        The number of samples (or total weight) in a neighborhood for a point to
+        be considered as a core point. This includes the point itself. If
+        `min_samples` is set to a higher value, DBSCAN will find denser clusters,
+        whereas if it is set to a lower value, the found clusters will be more
+        sparse.
+
+    metric : str, or callable, default='euclidean'
+        The metric to use when calculating distance between instances in a
+        feature array. If metric is a string or callable, it must be one of
+        the options allowed by :func:`sklearn.metrics.pairwise_distances` for
+        its metric parameter.
+        If metric is "precomputed", X is assumed to be a distance matrix and
+        must be square. X may be a :term:`sparse graph`, in which
+        case only "nonzero" elements may be considered neighbors for DBSCAN.
+
+        .. versionadded:: 0.17
+           metric *precomputed* to accept precomputed sparse matrix.
+
+    metric_params : dict, default=None
+        Additional keyword arguments for the metric function.
+
+        .. versionadded:: 0.19
+
+    algorithm : {'auto', 'ball_tree', 'kd_tree', 'brute'}, default='auto'
+        The algorithm to be used by the NearestNeighbors module
+        to compute pointwise distances and find nearest neighbors.
+        See NearestNeighbors module documentation for details.
+
+    leaf_size : int, default=30
+        Leaf size passed to BallTree or cKDTree. This can affect the speed
+        of the construction and query, as well as the memory required
+        to store the tree. The optimal value depends
+        on the nature of the problem.
+
+    p : float, default=None
+        The power of the Minkowski metric to be used to calculate distance
+        between points. If None, then ``p=2`` (equivalent to the Euclidean
+        distance).
+
+    n_jobs : int, default=None
+        The number of parallel jobs to run.
+        ``None`` means 1 unless in a :obj:`joblib.parallel_backend` context.
+        ``-1`` means using all processors. See :term:`Glossary <n_jobs>`
+        for more details.
+
+    Attributes
+    ----------
+    core_sample_indices_ : ndarray of shape (n_core_samples,)
+        Indices of core samples.
+
+    components_ : ndarray of shape (n_core_samples, n_features)
+        Copy of each core sample found by training.
+
+    labels_ : ndarray of shape (n_samples)
+        Cluster labels for each point in the dataset given to fit().
+        Noisy samples are given the label -1.
+
+    n_features_in_ : int
+        Number of features seen during :term:`fit`.
+
+        .. versionadded:: 0.24
+
+    feature_names_in_ : ndarray of shape (`n_features_in_`,)
+        Names of features seen during :term:`fit`. Defined only when `X`
+        has feature names that are all strings.
+
+        .. versionadded:: 1.0
+
+    See Also
+    --------
+    OPTICS : A similar clustering at multiple values of eps. Our implementation
+        is optimized for memory usage.
+
+    Notes
+    -----
+    For an example, see :ref:`examples/cluster/plot_dbscan.py
+    <sphx_glr_auto_examples_cluster_plot_dbscan.py>`.
+
+    This implementation bulk-computes all neighborhood queries, which increases
+    the memory complexity to O(n.d) where d is the average number of neighbors,
+    while original DBSCAN had memory complexity O(n). It may attract a higher
+    memory complexity when querying these nearest neighborhoods, depending
+    on the ``algorithm``.
+
+    One way to avoid the query complexity is to pre-compute sparse
+    neighborhoods in chunks using
+    :func:`NearestNeighbors.radius_neighbors_graph
+    <sklearn.neighbors.NearestNeighbors.radius_neighbors_graph>` with
+    ``mode='distance'``, then using ``metric='precomputed'`` here.
+
+    Another way to reduce memory and computation time is to remove
+    (near-)duplicate points and use ``sample_weight`` instead.
+
+    :class:`~sklearn.cluster.OPTICS` provides a similar clustering with lower memory
+    usage.
+
+    References
+    ----------
+    Ester, M., H. P. Kriegel, J. Sander, and X. Xu, `"A Density-Based
+    Algorithm for Discovering Clusters in Large Spatial Databases with Noise"
+    <https://www.dbs.ifi.lmu.de/Publikationen/Papers/KDD-96.final.frame.pdf>`_.
+    In: Proceedings of the 2nd International Conference on Knowledge Discovery
+    and Data Mining, Portland, OR, AAAI Press, pp. 226-231. 1996
+
+    Schubert, E., Sander, J., Ester, M., Kriegel, H. P., & Xu, X. (2017).
+    :doi:`"DBSCAN revisited, revisited: why and how you should (still) use DBSCAN."
+    <10.1145/3068335>`
+    ACM Transactions on Database Systems (TODS), 42(3), 19.
+
+    Examples
+    --------
+    >>> from sklearn.cluster import DBSCAN
+    >>> import numpy as np
+    >>> X = np.array([[1, 2], [2, 2], [2, 3],
+    ...               [8, 7], [8, 8], [25, 80]])
+    >>> clustering = DBSCAN(eps=3, min_samples=2).fit(X)
+    >>> clustering.labels_
+    array([ 0,  0,  0,  1,  1, -1])
+    >>> clustering
+    DBSCAN(eps=3, min_samples=2)
+    """
+
+    _parameter_constraints: dict = {
+        "eps": [Interval(Real, 0.0, None, closed="neither")],
+        "min_samples": [Interval(Integral, 1, None, closed="left")],
+        "metric": [
+            StrOptions(set(_VALID_METRICS) | {"precomputed"}),
+            callable,
+        ],
+        "metric_params": [dict, None],
+        "algorithm": [StrOptions({"auto", "ball_tree", "kd_tree", "brute"})],
+        "leaf_size": [Interval(Integral, 1, None, closed="left")],
+        "p": [Interval(Real, 0.0, None, closed="left"), None],
+        "n_jobs": [Integral, None],
+    }
+
+    def __init__(
+        self,
+        eps=0.5,
+        *,
+        min_samples=5,
+        metric="euclidean",
+        metric_params=None,
+        algorithm="auto",
+        leaf_size=30,
+        p=None,
+        n_jobs=None,
+    ):
+        self.eps = eps
+        self.min_samples = min_samples
+        self.metric = metric
+        self.metric_params = metric_params
+        self.algorithm = algorithm
+        self.leaf_size = leaf_size
+        self.p = p
+        self.n_jobs = n_jobs
+
+    @_fit_context(
+        # DBSCAN.metric is not validated yet
+        prefer_skip_nested_validation=False
+    )
+    def fit(self, X, y=None, sample_weight=None):
+        """Perform DBSCAN clustering from features, or distance matrix.
+
+        Parameters
+        ----------
+        X : {array-like, sparse matrix} of shape (n_samples, n_features), or \
+            (n_samples, n_samples)
+            Training instances to cluster, or distances between instances if
+            ``metric='precomputed'``. If a sparse matrix is provided, it will
+            be converted into a sparse ``csr_matrix``.
+
+        y : Ignored
+            Not used, present here for API consistency by convention.
+
+        sample_weight : array-like of shape (n_samples,), default=None
+            Weight of each sample, such that a sample with a weight of at least
+            ``min_samples`` is by itself a core sample; a sample with a
+            negative weight may inhibit its eps-neighbor from being core.
+            Note that weights are absolute, and default to 1.
+
+        Returns
+        -------
+        self : object
+            Returns a fitted instance of self.
+        """
+        X = self._validate_data(X, accept_sparse="csr")
+
+        if sample_weight is not None:
+            sample_weight = _check_sample_weight(sample_weight, X)
+
+        # Calculate neighborhood for all samples. This leaves the original
+        # point in, which needs to be considered later (i.e. point i is in the
+        # neighborhood of point i. While True, its useless information)
+        if self.metric == "precomputed" and sparse.issparse(X):
+            # set the diagonal to explicit values, as a point is its own
+            # neighbor
+            X = X.copy()  # copy to avoid in-place modification
+            with warnings.catch_warnings():
+                warnings.simplefilter("ignore", sparse.SparseEfficiencyWarning)
+                X.setdiag(X.diagonal())
+
+        neighbors_model = NearestNeighbors(
+            radius=self.eps,
+            algorithm=self.algorithm,
+            leaf_size=self.leaf_size,
+            metric=self.metric,
+            metric_params=self.metric_params,
+            p=self.p,
+            n_jobs=self.n_jobs,
+        )
+        neighbors_model.fit(X)
+        # This has worst case O(n^2) memory complexity
+        neighborhoods = neighbors_model.radius_neighbors(X, return_distance=False)
+
+        if sample_weight is None:
+            n_neighbors = np.array([len(neighbors) for neighbors in neighborhoods])
+        else:
+            n_neighbors = np.array(
+                [np.sum(sample_weight[neighbors]) for neighbors in neighborhoods]
+            )
+
+        # Initially, all samples are noise.
+        labels = np.full(X.shape[0], -1, dtype=np.intp)
+
+        # A list of all core samples found.
+        core_samples = np.asarray(n_neighbors >= self.min_samples, dtype=np.uint8)
+        dbscan_inner(core_samples, neighborhoods, labels)
+
+        self.core_sample_indices_ = np.where(core_samples)[0]
+        self.labels_ = labels
+
+        if len(self.core_sample_indices_):
+            # fix for scipy sparse indexing issue
+            self.components_ = X[self.core_sample_indices_].copy()
+        else:
+            # no core samples
+            self.components_ = np.empty((0, X.shape[1]))
+        return self
+
+    def fit_predict(self, X, y=None, sample_weight=None):
+        """Compute clusters from a data or distance matrix and predict labels.
+
+        Parameters
+        ----------
+        X : {array-like, sparse matrix} of shape (n_samples, n_features), or \
+            (n_samples, n_samples)
+            Training instances to cluster, or distances between instances if
+            ``metric='precomputed'``. If a sparse matrix is provided, it will
+            be converted into a sparse ``csr_matrix``.
+
+        y : Ignored
+            Not used, present here for API consistency by convention.
+
+        sample_weight : array-like of shape (n_samples,), default=None
+            Weight of each sample, such that a sample with a weight of at least
+            ``min_samples`` is by itself a core sample; a sample with a
+            negative weight may inhibit its eps-neighbor from being core.
+            Note that weights are absolute, and default to 1.
+
+        Returns
+        -------
+        labels : ndarray of shape (n_samples,)
+            Cluster labels. Noisy samples are given the label -1.
+        """
+        self.fit(X, sample_weight=sample_weight)
+        return self.labels_
+
+    def _more_tags(self):
+        return {"pairwise": self.metric == "precomputed"}

.venv/lib/python3.12/site-packages/sklearn/cluster/_dbscan_inner.pyx

@@ -0,0 +1,40 @@
+# Fast inner loop for DBSCAN.
+# Author: Lars Buitinck
+# License: 3-clause BSD
+
+from libcpp.vector cimport vector
+
+from ..utils._typedefs cimport uint8_t, intp_t
+
+
+def dbscan_inner(const uint8_t[::1] is_core,
+                 object[:] neighborhoods,
+                 intp_t[::1] labels):
+    cdef intp_t i, label_num = 0, v
+    cdef intp_t[:] neighb
+    cdef vector[intp_t] stack
+
+    for i in range(labels.shape[0]):
+        if labels[i] != -1 or not is_core[i]:
+            continue
+
+        # Depth-first search starting from i, ending at the non-core points.
+        # This is very similar to the classic algorithm for computing connected
+        # components, the difference being that we label non-core points as
+        # part of a cluster (component), but don't expand their neighborhoods.
+        while True:
+            if labels[i] == -1:
+                labels[i] = label_num
+                if is_core[i]:
+                    neighb = neighborhoods[i]
+                    for i in range(neighb.shape[0]):
+                        v = neighb[i]
+                        if labels[v] == -1:
+                            stack.push_back(v)
+
+            if stack.size() == 0:
+                break
+            i = stack.back()
+            stack.pop_back()
+
+        label_num += 1

.venv/lib/python3.12/site-packages/sklearn/cluster/_feature_agglomeration.py

@@ -0,0 +1,92 @@
+"""
+Feature agglomeration. Base classes and functions for performing feature
+agglomeration.
+"""
+
+# Author: V. Michel, A. Gramfort
+# License: BSD 3 clause
+
+
+import numpy as np
+from scipy.sparse import issparse
+
+from ..base import TransformerMixin
+from ..utils import metadata_routing
+from ..utils.deprecation import _deprecate_Xt_in_inverse_transform
+from ..utils.validation import check_is_fitted
+
+###############################################################################
+# Mixin class for feature agglomeration.
+
+
+class AgglomerationTransform(TransformerMixin):
+    """
+    A class for feature agglomeration via the transform interface.
+    """
+
+    # This prevents ``set_split_inverse_transform`` to be generated for the
+    # non-standard ``Xt`` arg on ``inverse_transform``.
+    # TODO(1.7): remove when Xt is removed for inverse_transform.
+    __metadata_request__inverse_transform = {"Xt": metadata_routing.UNUSED}
+
+    def transform(self, X):
+        """
+        Transform a new matrix using the built clustering.
+
+        Parameters
+        ----------
+        X : array-like of shape (n_samples, n_features) or \
+                (n_samples, n_samples)
+            A M by N array of M observations in N dimensions or a length
+            M array of M one-dimensional observations.
+
+        Returns
+        -------
+        Y : ndarray of shape (n_samples, n_clusters) or (n_clusters,)
+            The pooled values for each feature cluster.
+        """
+        check_is_fitted(self)
+
+        X = self._validate_data(X, reset=False)
+        if self.pooling_func == np.mean and not issparse(X):
+            size = np.bincount(self.labels_)
+            n_samples = X.shape[0]
+            # a fast way to compute the mean of grouped features
+            nX = np.array(
+                [np.bincount(self.labels_, X[i, :]) / size for i in range(n_samples)]
+            )
+        else:
+            nX = [
+                self.pooling_func(X[:, self.labels_ == l], axis=1)
+                for l in np.unique(self.labels_)
+            ]
+            nX = np.array(nX).T
+        return nX
+
+    def inverse_transform(self, X=None, *, Xt=None):
+        """
+        Inverse the transformation and return a vector of size `n_features`.
+
+        Parameters
+        ----------
+        X : array-like of shape (n_samples, n_clusters) or (n_clusters,)
+            The values to be assigned to each cluster of samples.
+
+        Xt : array-like of shape (n_samples, n_clusters) or (n_clusters,)
+            The values to be assigned to each cluster of samples.
+
+            .. deprecated:: 1.5
+                `Xt` was deprecated in 1.5 and will be removed in 1.7. Use `X` instead.
+
+        Returns
+        -------
+        X : ndarray of shape (n_samples, n_features) or (n_features,)
+            A vector of size `n_samples` with the values of `Xred` assigned to
+            each of the cluster of samples.
+        """
+        X = _deprecate_Xt_in_inverse_transform(X, Xt)
+
+        check_is_fitted(self)
+
+        unil, inverse = np.unique(self.labels_, return_inverse=True)
+        return X[..., inverse]

.venv/lib/python3.12/site-packages/sklearn/cluster/_hdbscan/_linkage.pyx

@@ -0,0 +1,272 @@
+# Minimum spanning tree single linkage implementation for hdbscan
+# Authors: Leland McInnes <leland.mcinnes@gmail.com>
+#          Steve Astels <sastels@gmail.com>
+#          Meekail Zain <zainmeekail@gmail.com>
+# Copyright (c) 2015, Leland McInnes
+# All rights reserved.
+
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are met:
+
+# 1. Redistributions of source code must retain the above copyright notice,
+# this list of conditions and the following disclaimer.
+
+# 2. Redistributions in binary form must reproduce the above copyright notice,
+# this list of conditions and the following disclaimer in the documentation
+# and/or other materials provided with the distribution.
+
+# 3. Neither the name of the copyright holder nor the names of its contributors
+# may be used to endorse or promote products derived from this software without
+# specific prior written permission.
+
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+# AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+# ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+# LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+# CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+# SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+# INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+# CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+# ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+# POSSIBILITY OF SUCH DAMAGE.
+
+cimport numpy as cnp
+from libc.float cimport DBL_MAX
+
+import numpy as np
+from ...metrics._dist_metrics cimport DistanceMetric64
+from ...cluster._hierarchical_fast cimport UnionFind
+from ...cluster._hdbscan._tree cimport HIERARCHY_t
+from ...cluster._hdbscan._tree import HIERARCHY_dtype
+from ...utils._typedefs cimport intp_t, float64_t, int64_t, uint8_t
+
+cnp.import_array()
+
+cdef extern from "numpy/arrayobject.h":
+    intp_t * PyArray_SHAPE(cnp.PyArrayObject *)
+
+# Numpy structured dtype representing a single ordered edge in Prim's algorithm
+MST_edge_dtype = np.dtype([
+    ("current_node", np.int64),
+    ("next_node", np.int64),
+    ("distance", np.float64),
+])
+
+# Packed shouldn't make a difference since they're all 8-byte quantities,
+# but it's included just to be safe.
+ctypedef packed struct MST_edge_t:
+    int64_t current_node
+    int64_t next_node
+    float64_t distance
+
+cpdef cnp.ndarray[MST_edge_t, ndim=1, mode='c'] mst_from_mutual_reachability(
+    cnp.ndarray[float64_t, ndim=2] mutual_reachability
+):
+    """Compute the Minimum Spanning Tree (MST) representation of the mutual-
+    reachability graph using Prim's algorithm.
+
+    Parameters
+    ----------
+    mutual_reachability : ndarray of shape (n_samples, n_samples)
+        Array of mutual-reachabilities between samples.
+
+    Returns
+    -------
+    mst : ndarray of shape (n_samples - 1,), dtype=MST_edge_dtype
+        The MST representation of the mutual-reahability graph. The MST is
+        represented as a collecteion of edges.
+    """
+    cdef:
+        # Note: we utilize ndarray's over memory-views to make use of numpy
+        # binary indexing and sub-selection below.
+        cnp.ndarray[int64_t, ndim=1, mode='c'] current_labels
+        cnp.ndarray[float64_t, ndim=1, mode='c'] min_reachability, left, right
+        cnp.ndarray[MST_edge_t, ndim=1, mode='c'] mst
+
+        cnp.ndarray[uint8_t, mode='c'] label_filter
+
+        int64_t n_samples = PyArray_SHAPE(<cnp.PyArrayObject*> mutual_reachability)[0]
+        int64_t current_node, new_node_index, new_node, i
+
+    mst = np.empty(n_samples - 1, dtype=MST_edge_dtype)
+    current_labels = np.arange(n_samples, dtype=np.int64)
+    current_node = 0
+    min_reachability = np.full(n_samples, fill_value=np.inf, dtype=np.float64)
+    for i in range(0, n_samples - 1):
+        label_filter = current_labels != current_node
+        current_labels = current_labels[label_filter]
+        left = min_reachability[label_filter]
+        right = mutual_reachability[current_node][current_labels]
+        min_reachability = np.minimum(left, right)
+
+        new_node_index = np.argmin(min_reachability)
+        new_node = current_labels[new_node_index]
+        mst[i].current_node = current_node
+        mst[i].next_node = new_node
+        mst[i].distance = min_reachability[new_node_index]
+        current_node = new_node
+
+    return mst
+
+
+cpdef cnp.ndarray[MST_edge_t, ndim=1, mode='c'] mst_from_data_matrix(
+    const float64_t[:, ::1] raw_data,
+    const float64_t[::1] core_distances,
+    DistanceMetric64 dist_metric,
+    float64_t alpha=1.0
+):
+    """Compute the Minimum Spanning Tree (MST) representation of the mutual-
+    reachability graph generated from the provided `raw_data` and
+    `core_distances` using Prim's algorithm.
+
+    Parameters
+    ----------
+    raw_data : ndarray of shape (n_samples, n_features)
+        Input array of data samples.
+
+    core_distances : ndarray of shape (n_samples,)
+        An array containing the core-distance calculated for each corresponding
+        sample.
+
+    dist_metric : DistanceMetric
+        The distance metric to use when calculating pairwise distances for
+        determining mutual-reachability.
+
+    Returns
+    -------
+    mst : ndarray of shape (n_samples - 1,), dtype=MST_edge_dtype
+        The MST representation of the mutual-reahability graph. The MST is
+        represented as a collecteion of edges.
+    """
+
+    cdef:
+        uint8_t[::1] in_tree
+        float64_t[::1] min_reachability
+        int64_t[::1] current_sources
+        cnp.ndarray[MST_edge_t, ndim=1, mode='c'] mst
+
+        int64_t current_node, source_node, new_node, next_node_source
+        int64_t i, j, n_samples, num_features
+
+        float64_t current_node_core_dist, new_reachability, mutual_reachability_distance
+        float64_t next_node_min_reach, pair_distance, next_node_core_dist
+
+    n_samples = raw_data.shape[0]
+    num_features = raw_data.shape[1]
+
+    mst = np.empty(n_samples - 1, dtype=MST_edge_dtype)
+
+    in_tree = np.zeros(n_samples, dtype=np.uint8)
+    min_reachability = np.full(n_samples, fill_value=np.inf, dtype=np.float64)
+    current_sources = np.ones(n_samples, dtype=np.int64)
+
+    current_node = 0
+
+    for i in range(0, n_samples - 1):
+
+        in_tree[current_node] = 1
+
+        current_node_core_dist = core_distances[current_node]
+
+        new_reachability = DBL_MAX
+        source_node = 0
+        new_node = 0
+
+        for j in range(n_samples):
+            if in_tree[j]:
+                continue
+
+            next_node_min_reach = min_reachability[j]
+            next_node_source = current_sources[j]
+
+            pair_distance = dist_metric.dist(
+                &raw_data[current_node, 0],
+                &raw_data[j, 0],
+                num_features
+            )
+
+            pair_distance /= alpha
+
+            next_node_core_dist = core_distances[j]
+            mutual_reachability_distance = max(
+                current_node_core_dist,
+                next_node_core_dist,
+                pair_distance
+            )
+            if mutual_reachability_distance > next_node_min_reach:
+                if next_node_min_reach < new_reachability:
+                    new_reachability = next_node_min_reach
+                    source_node = next_node_source
+                    new_node = j
+                continue
+
+            if mutual_reachability_distance < next_node_min_reach:
+                min_reachability[j] = mutual_reachability_distance
+                current_sources[j] = current_node
+                if mutual_reachability_distance < new_reachability:
+                    new_reachability = mutual_reachability_distance
+                    source_node = current_node
+                    new_node = j
+            else:
+                if next_node_min_reach < new_reachability:
+                    new_reachability = next_node_min_reach
+                    source_node = next_node_source
+                    new_node = j
+
+        mst[i].current_node = source_node
+        mst[i].next_node = new_node
+        mst[i].distance = new_reachability
+        current_node = new_node
+
+    return mst
+
+cpdef cnp.ndarray[HIERARCHY_t, ndim=1, mode="c"] make_single_linkage(const MST_edge_t[::1] mst):
+    """Construct a single-linkage tree from an MST.
+
+    Parameters
+    ----------
+    mst : ndarray of shape (n_samples - 1,), dtype=MST_edge_dtype
+        The MST representation of the mutual-reahability graph. The MST is
+        represented as a collecteion of edges.
+
+    Returns
+    -------
+    single_linkage : ndarray of shape (n_samples - 1,), dtype=HIERARCHY_dtype
+        The single-linkage tree tree (dendrogram) built from the MST. Each
+        of the array represents the following:
+
+        - left node/cluster
+        - right node/cluster
+        - distance
+        - new cluster size
+    """
+    cdef:
+        cnp.ndarray[HIERARCHY_t, ndim=1, mode="c"] single_linkage
+
+        # Note mst.shape[0] is one fewer than the number of samples
+        int64_t n_samples = mst.shape[0] + 1
+        intp_t current_node_cluster, next_node_cluster
+        int64_t current_node, next_node, i
+        float64_t distance
+        UnionFind U = UnionFind(n_samples)
+
+    single_linkage = np.zeros(n_samples - 1, dtype=HIERARCHY_dtype)
+
+    for i in range(n_samples - 1):
+
+        current_node = mst[i].current_node
+        next_node = mst[i].next_node
+        distance = mst[i].distance
+
+        current_node_cluster = U.fast_find(current_node)
+        next_node_cluster = U.fast_find(next_node)
+
+        single_linkage[i].left_node = current_node_cluster
+        single_linkage[i].right_node = next_node_cluster
+        single_linkage[i].value = distance
+        single_linkage[i].cluster_size = U.size[current_node_cluster] + U.size[next_node_cluster]
+
+        U.union(current_node_cluster, next_node_cluster)
+
+    return single_linkage

.venv/lib/python3.12/site-packages/sklearn/cluster/_hdbscan/_reachability.pyx

@@ -0,0 +1,212 @@
+# mutual reachability distance computations
+# Authors: Leland McInnes <leland.mcinnes@gmail.com>
+#          Meekail Zain <zainmeekail@gmail.com>
+#          Guillaume Lemaitre <g.lemaitre58@gmail.com>
+# Copyright (c) 2015, Leland McInnes
+# All rights reserved.
+
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are met:
+
+# 1. Redistributions of source code must retain the above copyright notice,
+# this list of conditions and the following disclaimer.
+
+# 2. Redistributions in binary form must reproduce the above copyright notice,
+# this list of conditions and the following disclaimer in the documentation
+# and/or other materials provided with the distribution.
+
+# 3. Neither the name of the copyright holder nor the names of its contributors
+# may be used to endorse or promote products derived from this software without
+# specific prior written permission.
+
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+# AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+# ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+# LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+# CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+# SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+# INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+# CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+# ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+# POSSIBILITY OF SUCH DAMAGE.
+
+cimport numpy as cnp
+
+import numpy as np
+from scipy.sparse import issparse
+from cython cimport floating, integral
+from libc.math cimport isfinite, INFINITY
+from ...utils._typedefs cimport intp_t
+cnp.import_array()
+
+
+def mutual_reachability_graph(
+    distance_matrix, min_samples=5, max_distance=0.0
+):
+    """Compute the weighted adjacency matrix of the mutual reachability graph.
+
+    The mutual reachability distance used to build the graph is defined as::
+
+        max(d_core(x_p), d_core(x_q), d(x_p, x_q))
+
+    and the core distance `d_core` is defined as the distance between a point
+    `x_p` and its k-th nearest neighbor.
+
+    Note that all computations are done in-place.
+
+    Parameters
+    ----------
+    distance_matrix : {ndarray, sparse matrix} of shape (n_samples, n_samples)
+        Array of distances between samples. If sparse, the array must be in
+        `CSR` format.
+
+    min_samples : int, default=5
+        The parameter `k` used to calculate the distance between a point
+        `x_p` and its k-th nearest neighbor.
+
+    max_distance : float, default=0.0
+        The distance which `np.inf` is replaced with. When the true mutual-
+        reachability distance is measured to be infinite, it is instead
+        truncated to `max_dist`. Only used when `distance_matrix` is a sparse
+        matrix.
+
+    Returns
+    -------
+    mututal_reachability_graph: {ndarray, sparse matrix} of shape \
+            (n_samples, n_samples)
+        Weighted adjacency matrix of the mutual reachability graph.
+
+    References
+    ----------
+    .. [1] Campello, R. J., Moulavi, D., & Sander, J. (2013, April).
+       Density-based clustering based on hierarchical density estimates.
+       In Pacific-Asia Conference on Knowledge Discovery and Data Mining
+       (pp. 160-172). Springer Berlin Heidelberg.
+    """
+    further_neighbor_idx = min_samples - 1
+    if issparse(distance_matrix):
+        if distance_matrix.format != "csr":
+            raise ValueError(
+                "Only sparse CSR matrices are supported for `distance_matrix`."
+            )
+        _sparse_mutual_reachability_graph(
+            distance_matrix.data,
+            distance_matrix.indices,
+            distance_matrix.indptr,
+            distance_matrix.shape[0],
+            further_neighbor_idx=further_neighbor_idx,
+            max_distance=max_distance,
+        )
+    else:
+        _dense_mutual_reachability_graph(
+            distance_matrix, further_neighbor_idx=further_neighbor_idx
+        )
+    return distance_matrix
+
+
+def _dense_mutual_reachability_graph(
+    floating[:, :] distance_matrix,
+    intp_t further_neighbor_idx,
+):
+    """Dense implementation of mutual reachability graph.
+
+    The computation is done in-place, i.e. the distance matrix is modified
+    directly.
+
+    Parameters
+    ----------
+    distance_matrix : ndarray of shape (n_samples, n_samples)
+        Array of distances between samples.
+
+    further_neighbor_idx : int
+        The index of the furthest neighbor to use to define the core distances.
+    """
+    cdef:
+        intp_t i, j, n_samples = distance_matrix.shape[0]
+        floating mutual_reachibility_distance
+        floating[::1] core_distances
+
+    # We assume that the distance matrix is symmetric. We choose to sort every
+    # row to have the same implementation than the sparse case that requires
+    # CSR matrix.
+    core_distances = np.ascontiguousarray(
+        np.partition(
+            distance_matrix, further_neighbor_idx, axis=1
+        )[:, further_neighbor_idx]
+    )
+
+    with nogil:
+        # TODO: Update w/ prange with thread count based on
+        # _openmp_effective_n_threads
+        for i in range(n_samples):
+            for j in range(n_samples):
+                mutual_reachibility_distance = max(
+                    core_distances[i],
+                    core_distances[j],
+                    distance_matrix[i, j],
+                )
+                distance_matrix[i, j] = mutual_reachibility_distance
+
+
+def _sparse_mutual_reachability_graph(
+    cnp.ndarray[floating, ndim=1, mode="c"] data,
+    cnp.ndarray[integral, ndim=1, mode="c"] indices,
+    cnp.ndarray[integral, ndim=1, mode="c"] indptr,
+    intp_t n_samples,
+    intp_t further_neighbor_idx,
+    floating max_distance,
+):
+    """Sparse implementation of mutual reachability graph.
+
+    The computation is done in-place, i.e. the distance matrix is modified
+    directly. This implementation only accepts `CSR` format sparse matrices.
+
+    Parameters
+    ----------
+    distance_matrix : sparse matrix of shape (n_samples, n_samples)
+        Sparse matrix of distances between samples. The sparse format should
+        be `CSR`.
+
+    further_neighbor_idx : int
+        The index of the furthest neighbor to use to define the core distances.
+
+    max_distance : float
+        The distance which `np.inf` is replaced with. When the true mutual-
+        reachability distance is measured to be infinite, it is instead
+        truncated to `max_dist`. Only used when `distance_matrix` is a sparse
+        matrix.
+    """
+    cdef:
+        integral i, col_ind, row_ind
+        floating mutual_reachibility_distance
+        floating[:] core_distances
+        floating[:] row_data
+
+    if floating is float:
+        dtype = np.float32
+    else:
+        dtype = np.float64
+
+    core_distances = np.empty(n_samples, dtype=dtype)
+
+    for i in range(n_samples):
+        row_data = data[indptr[i]:indptr[i + 1]]
+        if further_neighbor_idx < row_data.size:
+            core_distances[i] = np.partition(
+                row_data, further_neighbor_idx
+            )[further_neighbor_idx]
+        else:
+            core_distances[i] = INFINITY
+
+    with nogil:
+        for row_ind in range(n_samples):
+            for i in range(indptr[row_ind], indptr[row_ind + 1]):
+                col_ind = indices[i]
+                mutual_reachibility_distance = max(
+                    core_distances[row_ind], core_distances[col_ind], data[i]
+                )
+                if isfinite(mutual_reachibility_distance):
+                    data[i] = mutual_reachibility_distance
+                elif max_distance > 0:
+                    data[i] = max_distance

.venv/lib/python3.12/site-packages/sklearn/cluster/_hdbscan/_tree.pxd

@@ -0,0 +1,49 @@
+# Copyright (c) 2015, Leland McInnes
+# All rights reserved.
+
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are met:
+
+# 1. Redistributions of source code must retain the above copyright notice,
+# this list of conditions and the following disclaimer.
+
+# 2. Redistributions in binary form must reproduce the above copyright notice,
+# this list of conditions and the following disclaimer in the documentation
+# and/or other materials provided with the distribution.
+
+# 3. Neither the name of the copyright holder nor the names of its contributors
+# may be used to endorse or promote products derived from this software without
+# specific prior written permission.
+
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+# AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+# ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+# LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+# CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+# SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+# INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+# CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+# ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+# POSSIBILITY OF SUCH DAMAGE.
+
+from ...utils._typedefs cimport intp_t, float64_t, uint8_t
+cimport numpy as cnp
+
+# This corresponds to the scipy.cluster.hierarchy format
+ctypedef packed struct HIERARCHY_t:
+    intp_t left_node
+    intp_t right_node
+    float64_t value
+    intp_t cluster_size
+
+# Effectively an edgelist encoding a parent/child pair, along with a value and
+# the corresponding cluster_size in each row providing a tree structure.
+ctypedef packed struct CONDENSED_t:
+    intp_t parent
+    intp_t child
+    float64_t value
+    intp_t cluster_size
+
+cdef extern from "numpy/arrayobject.h":
+    intp_t * PyArray_SHAPE(cnp.PyArrayObject *)

.venv/lib/python3.12/site-packages/sklearn/cluster/_hdbscan/_tree.pyx

@@ -0,0 +1,799 @@
+# Tree handling (condensing, finding stable clusters) for hdbscan
+# Authors: Leland McInnes
+# Copyright (c) 2015, Leland McInnes
+# All rights reserved.
+
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are met:
+
+# 1. Redistributions of source code must retain the above copyright notice,
+# this list of conditions and the following disclaimer.
+
+# 2. Redistributions in binary form must reproduce the above copyright notice,
+# this list of conditions and the following disclaimer in the documentation
+# and/or other materials provided with the distribution.
+
+# 3. Neither the name of the copyright holder nor the names of its contributors
+# may be used to endorse or promote products derived from this software without
+# specific prior written permission.
+
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+# AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+# ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+# LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+# CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+# SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+# INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+# CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+# ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+# POSSIBILITY OF SUCH DAMAGE.
+
+
+cimport numpy as cnp
+from libc.math cimport isinf
+import cython
+
+import numpy as np
+
+cnp.import_array()
+
+cdef extern from "numpy/arrayobject.h":
+    intp_t * PyArray_SHAPE(cnp.PyArrayObject *)
+
+cdef cnp.float64_t INFTY = np.inf
+cdef cnp.intp_t NOISE = -1
+
+HIERARCHY_dtype = np.dtype([
+    ("left_node", np.intp),
+    ("right_node", np.intp),
+    ("value", np.float64),
+    ("cluster_size", np.intp),
+])
+
+CONDENSED_dtype = np.dtype([
+    ("parent", np.intp),
+    ("child", np.intp),
+    ("value", np.float64),
+    ("cluster_size", np.intp),
+])
+
+cpdef tuple tree_to_labels(
+    const HIERARCHY_t[::1] single_linkage_tree,
+    cnp.intp_t min_cluster_size=10,
+    cluster_selection_method="eom",
+    bint allow_single_cluster=False,
+    cnp.float64_t cluster_selection_epsilon=0.0,
+    max_cluster_size=None,
+):
+    cdef:
+        cnp.ndarray[CONDENSED_t, ndim=1, mode='c'] condensed_tree
+        cnp.ndarray[cnp.intp_t, ndim=1, mode='c'] labels
+        cnp.ndarray[cnp.float64_t, ndim=1, mode='c'] probabilities
+
+    condensed_tree = _condense_tree(single_linkage_tree, min_cluster_size)
+    labels, probabilities = _get_clusters(
+        condensed_tree,
+        _compute_stability(condensed_tree),
+        cluster_selection_method,
+        allow_single_cluster,
+        cluster_selection_epsilon,
+        max_cluster_size,
+    )
+
+    return (labels, probabilities)
+
+cdef list bfs_from_hierarchy(
+    const HIERARCHY_t[::1] hierarchy,
+    cnp.intp_t bfs_root
+):
+    """
+    Perform a breadth first search on a tree in scipy hclust format.
+    """
+
+    cdef list process_queue, next_queue, result
+    cdef cnp.intp_t n_samples = hierarchy.shape[0] + 1
+    cdef cnp.intp_t node
+    process_queue = [bfs_root]
+    result = []
+
+    while process_queue:
+        result.extend(process_queue)
+        # By construction, node i is formed by the union of nodes
+        # hierarchy[i - n_samples, 0] and hierarchy[i - n_samples, 1]
+        process_queue = [
+            x - n_samples
+            for x in process_queue
+            if x >= n_samples
+        ]
+        if process_queue:
+            next_queue = []
+            for node in process_queue:
+                next_queue.extend(
+                    [
+                        hierarchy[node].left_node,
+                        hierarchy[node].right_node,
+                    ]
+                )
+            process_queue = next_queue
+    return result
+
+
+cpdef cnp.ndarray[CONDENSED_t, ndim=1, mode='c'] _condense_tree(
+    const HIERARCHY_t[::1] hierarchy,
+    cnp.intp_t min_cluster_size=10
+):
+    """Condense a tree according to a minimum cluster size. This is akin
+    to the runt pruning procedure of Stuetzle. The result is a much simpler
+    tree that is easier to visualize. We include extra information on the
+    lambda value at which individual points depart clusters for later
+    analysis and computation.
+
+    Parameters
+    ----------
+    hierarchy : ndarray of shape (n_samples,), dtype=HIERARCHY_dtype
+        A single linkage hierarchy in scipy.cluster.hierarchy format.
+
+    min_cluster_size : int, optional (default 10)
+        The minimum size of clusters to consider. Clusters smaller than this
+        are pruned from the tree.
+
+    Returns
+    -------
+    condensed_tree : ndarray of shape (n_samples,), dtype=CONDENSED_dtype
+        Effectively an edgelist encoding a parent/child pair, along with a
+        value and the corresponding cluster_size in each row providing a tree
+        structure.
+    """
+
+    cdef:
+        cnp.intp_t root = 2 * hierarchy.shape[0]
+        cnp.intp_t n_samples = hierarchy.shape[0] + 1
+        cnp.intp_t next_label = n_samples + 1
+        list result_list, node_list = bfs_from_hierarchy(hierarchy, root)
+
+        cnp.intp_t[::1] relabel
+        cnp.uint8_t[::1] ignore
+
+        cnp.intp_t node, sub_node, left, right
+        cnp.float64_t lambda_value, distance
+        cnp.intp_t left_count, right_count
+        HIERARCHY_t children
+
+    relabel = np.empty(root + 1, dtype=np.intp)
+    relabel[root] = n_samples
+    result_list = []
+    ignore = np.zeros(len(node_list), dtype=bool)
+
+    for node in node_list:
+        if ignore[node] or node < n_samples:
+            continue
+
+        children = hierarchy[node - n_samples]
+        left = children.left_node
+        right = children.right_node
+        distance = children.value
+        if distance > 0.0:
+            lambda_value = 1.0 / distance
+        else:
+            lambda_value = INFTY
+
+        if left >= n_samples:
+            left_count = hierarchy[left - n_samples].cluster_size
+        else:
+            left_count = 1
+
+        if right >= n_samples:
+            right_count = <cnp.intp_t> hierarchy[right - n_samples].cluster_size
+        else:
+            right_count = 1
+
+        if left_count >= min_cluster_size and right_count >= min_cluster_size:
+            relabel[left] = next_label
+            next_label += 1
+            result_list.append(
+                (relabel[node], relabel[left], lambda_value, left_count)
+            )
+
+            relabel[right] = next_label
+            next_label += 1
+            result_list.append(
+                (relabel[node], relabel[right], lambda_value, right_count)
+            )
+
+        elif left_count < min_cluster_size and right_count < min_cluster_size:
+            for sub_node in bfs_from_hierarchy(hierarchy, left):
+                if sub_node < n_samples:
+                    result_list.append(
+                        (relabel[node], sub_node, lambda_value, 1)
+                    )
+                ignore[sub_node] = True
+
+            for sub_node in bfs_from_hierarchy(hierarchy, right):
+                if sub_node < n_samples:
+                    result_list.append(
+                        (relabel[node], sub_node, lambda_value, 1)
+                    )
+                ignore[sub_node] = True
+
+        elif left_count < min_cluster_size:
+            relabel[right] = relabel[node]
+            for sub_node in bfs_from_hierarchy(hierarchy, left):
+                if sub_node < n_samples:
+                    result_list.append(
+                        (relabel[node], sub_node, lambda_value, 1)
+                    )
+                ignore[sub_node] = True
+
+        else:
+            relabel[left] = relabel[node]
+            for sub_node in bfs_from_hierarchy(hierarchy, right):
+                if sub_node < n_samples:
+                    result_list.append(
+                        (relabel[node], sub_node, lambda_value, 1)
+                    )
+                ignore[sub_node] = True
+
+    return np.array(result_list, dtype=CONDENSED_dtype)
+
+
+cdef dict _compute_stability(
+    cnp.ndarray[CONDENSED_t, ndim=1, mode='c'] condensed_tree
+):
+
+    cdef:
+        cnp.float64_t[::1] result, births
+        cnp.intp_t[:] parents = condensed_tree['parent']
+
+        cnp.intp_t parent, cluster_size, result_index, idx
+        cnp.float64_t lambda_val
+        CONDENSED_t condensed_node
+        cnp.intp_t largest_child = condensed_tree['child'].max()
+        cnp.intp_t smallest_cluster = np.min(parents)
+        cnp.intp_t num_clusters = np.max(parents) - smallest_cluster + 1
+        dict stability_dict = {}
+
+    largest_child = max(largest_child, smallest_cluster)
+    births = np.full(largest_child + 1, np.nan, dtype=np.float64)
+
+    for idx in range(PyArray_SHAPE(<cnp.PyArrayObject*> condensed_tree)[0]):
+        condensed_node = condensed_tree[idx]
+        births[condensed_node.child] = condensed_node.value
+
+    births[smallest_cluster] = 0.0
+
+    result = np.zeros(num_clusters, dtype=np.float64)
+    for idx in range(PyArray_SHAPE(<cnp.PyArrayObject*> condensed_tree)[0]):
+        condensed_node = condensed_tree[idx]
+        parent = condensed_node.parent
+        lambda_val = condensed_node.value
+        cluster_size = condensed_node.cluster_size
+
+        result_index = parent - smallest_cluster
+        result[result_index] += (lambda_val - births[parent]) * cluster_size
+
+    for idx in range(num_clusters):
+        stability_dict[idx + smallest_cluster] = result[idx]
+
+    return stability_dict
+
+
+cdef list bfs_from_cluster_tree(
+    cnp.ndarray[CONDENSED_t, ndim=1, mode='c'] condensed_tree,
+    cnp.intp_t bfs_root
+):
+
+    cdef:
+        list result = []
+        cnp.ndarray[cnp.intp_t, ndim=1] process_queue = (
+            np.array([bfs_root], dtype=np.intp)
+        )
+        cnp.ndarray[cnp.intp_t, ndim=1] children = condensed_tree['child']
+        cnp.intp_t[:] parents = condensed_tree['parent']
+
+    while len(process_queue) > 0:
+        result.extend(process_queue.tolist())
+        process_queue = children[np.isin(parents, process_queue)]
+
+    return result
+
+
+cdef cnp.float64_t[::1] max_lambdas(cnp.ndarray[CONDENSED_t, ndim=1, mode='c'] condensed_tree):
+
+    cdef:
+        cnp.intp_t parent, current_parent, idx
+        cnp.float64_t lambda_val, max_lambda
+        cnp.float64_t[::1] deaths
+        cnp.intp_t largest_parent = condensed_tree['parent'].max()
+
+    deaths = np.zeros(largest_parent + 1, dtype=np.float64)
+    current_parent = condensed_tree[0].parent
+    max_lambda = condensed_tree[0].value
+
+    for idx in range(1, PyArray_SHAPE(<cnp.PyArrayObject*> condensed_tree)[0]):
+        parent = condensed_tree[idx].parent
+        lambda_val = condensed_tree[idx].value
+
+        if parent == current_parent:
+            max_lambda = max(max_lambda, lambda_val)
+        else:
+            deaths[current_parent] = max_lambda
+            current_parent = parent
+            max_lambda = lambda_val
+
+    deaths[current_parent] = max_lambda  # value for last parent
+    return deaths
+
+
+@cython.final
+cdef class TreeUnionFind:
+
+    cdef cnp.intp_t[:, ::1] data
+    cdef cnp.uint8_t[::1] is_component
+
+    def __init__(self, size):
+        cdef cnp.intp_t idx
+        self.data = np.zeros((size, 2), dtype=np.intp)
+        for idx in range(size):
+            self.data[idx, 0] = idx
+        self.is_component = np.ones(size, dtype=np.uint8)
+
+    cdef void union(self, cnp.intp_t x, cnp.intp_t y):
+        cdef cnp.intp_t x_root = self.find(x)
+        cdef cnp.intp_t y_root = self.find(y)
+
+        if self.data[x_root, 1] < self.data[y_root, 1]:
+            self.data[x_root, 0] = y_root
+        elif self.data[x_root, 1] > self.data[y_root, 1]:
+            self.data[y_root, 0] = x_root
+        else:
+            self.data[y_root, 0] = x_root
+            self.data[x_root, 1] += 1
+        return
+
+    cdef cnp.intp_t find(self, cnp.intp_t x):
+        if self.data[x, 0] != x:
+            self.data[x, 0] = self.find(self.data[x, 0])
+            self.is_component[x] = False
+        return self.data[x, 0]
+
+
+cpdef cnp.ndarray[cnp.intp_t, ndim=1, mode='c'] labelling_at_cut(
+        const HIERARCHY_t[::1] linkage,
+        cnp.float64_t cut,
+        cnp.intp_t min_cluster_size
+):
+    """Given a single linkage tree and a cut value, return the
+    vector of cluster labels at that cut value. This is useful
+    for Robust Single Linkage, and extracting DBSCAN results
+    from a single HDBSCAN run.
+
+    Parameters
+    ----------
+    linkage : ndarray of shape (n_samples,), dtype=HIERARCHY_dtype
+        The single linkage tree in scipy.cluster.hierarchy format.
+
+    cut : double
+        The cut value at which to find clusters.
+
+    min_cluster_size : int
+        The minimum cluster size; clusters below this size at
+        the cut will be considered noise.
+
+    Returns
+    -------
+    labels : ndarray of shape (n_samples,)
+        The cluster labels for each point in the data set;
+        a label of -1 denotes a noise assignment.
+    """
+
+    cdef:
+        cnp.intp_t n, cluster, root, n_samples, cluster_label
+        cnp.intp_t[::1] unique_labels, cluster_size
+        cnp.ndarray[cnp.intp_t, ndim=1, mode='c'] result
+        TreeUnionFind union_find
+        dict cluster_label_map
+        HIERARCHY_t node
+
+    root = 2 * linkage.shape[0]
+    n_samples = root // 2 + 1
+    result = np.empty(n_samples, dtype=np.intp)
+    union_find = TreeUnionFind(root + 1)
+
+    cluster = n_samples
+    for node in linkage:
+        if node.value < cut:
+            union_find.union(node.left_node, cluster)
+            union_find.union(node.right_node, cluster)
+        cluster += 1
+
+    cluster_size = np.zeros(cluster, dtype=np.intp)
+    for n in range(n_samples):
+        cluster = union_find.find(n)
+        cluster_size[cluster] += 1
+        result[n] = cluster
+
+    cluster_label_map = {-1: NOISE}
+    cluster_label = 0
+    unique_labels = np.unique(result)
+
+    for cluster in unique_labels:
+        if cluster_size[cluster] < min_cluster_size:
+            cluster_label_map[cluster] = NOISE
+        else:
+            cluster_label_map[cluster] = cluster_label
+            cluster_label += 1
+
+    for n in range(n_samples):
+        result[n] = cluster_label_map[result[n]]
+
+    return result
+
+
+cpdef cnp.ndarray[cnp.intp_t, ndim=1, mode='c'] _do_labelling(
+        cnp.ndarray[CONDENSED_t, ndim=1, mode='c'] condensed_tree,
+        set clusters,
+        dict cluster_label_map,
+        cnp.intp_t allow_single_cluster,
+        cnp.float64_t cluster_selection_epsilon
+):
+    """Given a condensed tree, clusters and a labeling map for the clusters,
+    return an array containing the labels of each point based on cluster
+    membership. Note that this is where points may be marked as noisy
+    outliers. The determination of some points as noise is in large, single-
+    cluster datasets is controlled by the `allow_single_cluster` and
+    `cluster_selection_epsilon` parameters.
+
+    Parameters
+    ----------
+    condensed_tree : ndarray of shape (n_samples,), dtype=CONDENSED_dtype
+        Effectively an edgelist encoding a parent/child pair, along with a
+        value and the corresponding cluster_size in each row providing a tree
+        structure.
+
+    clusters : set
+        The set of nodes corresponding to identified clusters. These node
+        values should be the same as those present in `condensed_tree`.
+
+    cluster_label_map : dict
+        A mapping from the node values present in `clusters` to the labels
+        which will be returned.
+
+    Returns
+    -------
+    labels : ndarray of shape (n_samples,)
+        The cluster labels for each point in the data set;
+        a label of -1 denotes a noise assignment.
+    """
+
+    cdef:
+        cnp.intp_t root_cluster
+        cnp.ndarray[cnp.intp_t, ndim=1, mode='c'] result
+        cnp.ndarray[cnp.intp_t, ndim=1] parent_array, child_array
+        cnp.ndarray[cnp.float64_t, ndim=1] lambda_array
+        TreeUnionFind union_find
+        cnp.intp_t n, parent, child, cluster
+        cnp.float64_t threshold
+
+    child_array = condensed_tree['child']
+    parent_array = condensed_tree['parent']
+    lambda_array = condensed_tree['value']
+
+    root_cluster = np.min(parent_array)
+    result = np.empty(root_cluster, dtype=np.intp)
+    union_find = TreeUnionFind(np.max(parent_array) + 1)
+
+    for n in range(PyArray_SHAPE(<cnp.PyArrayObject*> condensed_tree)[0]):
+        child = child_array[n]
+        parent = parent_array[n]
+        if child not in clusters:
+            union_find.union(parent, child)
+
+    for n in range(root_cluster):
+        cluster = union_find.find(n)
+        label = NOISE
+        if cluster != root_cluster:
+            label = cluster_label_map[cluster]
+        elif len(clusters) == 1 and allow_single_cluster:
+            # There can only be one edge with this particular child hence this
+            # expression extracts a unique, scalar lambda value.
+            parent_lambda = lambda_array[child_array == n]
+            if cluster_selection_epsilon != 0.0:
+                threshold = 1 / cluster_selection_epsilon
+            else:
+                # The threshold should be calculated per-sample based on the
+                # largest lambda of any simbling node.
+                threshold = lambda_array[parent_array == cluster].max()
+            if parent_lambda >= threshold:
+                label = cluster_label_map[cluster]
+
+        result[n] = label
+
+    return result
+
+
+cdef cnp.ndarray[cnp.float64_t, ndim=1, mode='c'] get_probabilities(
+    cnp.ndarray[CONDENSED_t, ndim=1, mode='c'] condensed_tree,
+    dict cluster_map,
+    cnp.intp_t[::1] labels
+):
+
+    cdef:
+        cnp.ndarray[cnp.float64_t, ndim=1, mode='c'] result
+        cnp.float64_t[:] lambda_array
+        cnp.float64_t[::1] deaths
+        cnp.intp_t[:] child_array, parent_array
+        cnp.intp_t root_cluster, n, point, cluster_num, cluster
+        cnp.float64_t max_lambda, lambda_val
+
+    child_array = condensed_tree['child']
+    parent_array = condensed_tree['parent']
+    lambda_array = condensed_tree['value']
+
+    result = np.zeros(labels.shape[0])
+    deaths = max_lambdas(condensed_tree)
+    root_cluster = np.min(parent_array)
+
+    for n in range(PyArray_SHAPE(<cnp.PyArrayObject*> condensed_tree)[0]):
+        point = child_array[n]
+        if point >= root_cluster:
+            continue
+
+        cluster_num = labels[point]
+        if cluster_num == -1:
+            continue
+
+        cluster = cluster_map[cluster_num]
+        max_lambda = deaths[cluster]
+        if max_lambda == 0.0 or isinf(lambda_array[n]):
+            result[point] = 1.0
+        else:
+            lambda_val = min(lambda_array[n], max_lambda)
+            result[point] = lambda_val / max_lambda
+
+    return result
+
+
+cpdef list recurse_leaf_dfs(
+    cnp.ndarray[CONDENSED_t, ndim=1, mode='c'] cluster_tree,
+    cnp.intp_t current_node
+):
+    cdef cnp.intp_t[:] children
+    cdef cnp.intp_t child
+
+    children = cluster_tree[cluster_tree['parent'] == current_node]['child']
+    if children.shape[0] == 0:
+        return [current_node,]
+    else:
+        return sum([recurse_leaf_dfs(cluster_tree, child) for child in children], [])
+
+
+cpdef list get_cluster_tree_leaves(cnp.ndarray[CONDENSED_t, ndim=1, mode='c'] cluster_tree):
+    cdef cnp.intp_t root
+    if PyArray_SHAPE(<cnp.PyArrayObject*> cluster_tree)[0] == 0:
+        return []
+    root = cluster_tree['parent'].min()
+    return recurse_leaf_dfs(cluster_tree, root)
+
+cdef cnp.intp_t traverse_upwards(
+    cnp.ndarray[CONDENSED_t, ndim=1, mode='c'] cluster_tree,
+    cnp.float64_t cluster_selection_epsilon,
+    cnp.intp_t leaf,
+    cnp.intp_t allow_single_cluster
+):
+    cdef cnp.intp_t root, parent
+    cdef cnp.float64_t parent_eps
+
+    root = cluster_tree['parent'].min()
+    parent = cluster_tree[cluster_tree['child'] == leaf]['parent']
+    if parent == root:
+        if allow_single_cluster:
+            return parent
+        else:
+            return leaf  # return node closest to root
+
+    parent_eps = 1 / cluster_tree[cluster_tree['child'] == parent]['value']
+    if parent_eps > cluster_selection_epsilon:
+        return parent
+    else:
+        return traverse_upwards(
+            cluster_tree,
+            cluster_selection_epsilon,
+            parent,
+            allow_single_cluster
+        )
+
+cdef set epsilon_search(
+    set leaves,
+    cnp.ndarray[CONDENSED_t, ndim=1, mode='c'] cluster_tree,
+    cnp.float64_t cluster_selection_epsilon,
+    cnp.intp_t allow_single_cluster
+):
+    cdef:
+        list selected_clusters = list()
+        list processed = list()
+        cnp.intp_t leaf, epsilon_child, sub_node
+        cnp.float64_t eps
+        cnp.uint8_t[:] leaf_nodes
+        cnp.ndarray[cnp.intp_t, ndim=1] children = cluster_tree['child']
+        cnp.ndarray[cnp.float64_t, ndim=1] distances = cluster_tree['value']
+
+    for leaf in leaves:
+        leaf_nodes = children == leaf
+        eps = 1 / distances[leaf_nodes][0]
+        if eps < cluster_selection_epsilon:
+            if leaf not in processed:
+                epsilon_child = traverse_upwards(
+                    cluster_tree,
+                    cluster_selection_epsilon,
+                    leaf,
+                    allow_single_cluster
+                )
+                selected_clusters.append(epsilon_child)
+
+                for sub_node in bfs_from_cluster_tree(cluster_tree, epsilon_child):
+                    if sub_node != epsilon_child:
+                        processed.append(sub_node)
+        else:
+            selected_clusters.append(leaf)
+
+    return set(selected_clusters)
+
+
+@cython.wraparound(True)
+cdef tuple _get_clusters(
+    cnp.ndarray[CONDENSED_t, ndim=1, mode='c'] condensed_tree,
+    dict stability,
+    cluster_selection_method='eom',
+    cnp.uint8_t allow_single_cluster=False,
+    cnp.float64_t cluster_selection_epsilon=0.0,
+    max_cluster_size=None
+):
+    """Given a tree and stability dict, produce the cluster labels
+    (and probabilities) for a flat clustering based on the chosen
+    cluster selection method.
+
+    Parameters
+    ----------
+    condensed_tree : ndarray of shape (n_samples,), dtype=CONDENSED_dtype
+        Effectively an edgelist encoding a parent/child pair, along with a
+        value and the corresponding cluster_size in each row providing a tree
+        structure.
+
+    stability : dict
+        A dictionary mapping cluster_ids to stability values
+
+    cluster_selection_method : string, optional (default 'eom')
+        The method of selecting clusters. The default is the
+        Excess of Mass algorithm specified by 'eom'. The alternate
+        option is 'leaf'.
+
+    allow_single_cluster : boolean, optional (default False)
+        Whether to allow a single cluster to be selected by the
+        Excess of Mass algorithm.
+
+    cluster_selection_epsilon: double, optional (default 0.0)
+        A distance threshold for cluster splits.
+
+    max_cluster_size: int, default=None
+        The maximum size for clusters located by the EOM clusterer. Can
+        be overridden by the cluster_selection_epsilon parameter in
+        rare cases.
+
+    Returns
+    -------
+    labels : ndarray of shape (n_samples,)
+        An integer array of cluster labels, with -1 denoting noise.
+
+    probabilities : ndarray (n_samples,)
+        The cluster membership strength of each sample.
+
+    stabilities : ndarray (n_clusters,)
+        The cluster coherence strengths of each cluster.
+    """
+    cdef:
+        list node_list
+        cnp.ndarray[CONDENSED_t, ndim=1, mode='c'] cluster_tree
+        cnp.uint8_t[::1] child_selection
+        cnp.ndarray[cnp.intp_t, ndim=1, mode='c'] labels
+        dict is_cluster, cluster_sizes
+        cnp.float64_t subtree_stability
+        cnp.intp_t node, sub_node, cluster, n_samples
+        cnp.ndarray[cnp.float64_t, ndim=1, mode='c'] probs
+
+    # Assume clusters are ordered by numeric id equivalent to
+    # a topological sort of the tree; This is valid given the
+    # current implementation above, so don't change that ... or
+    # if you do, change this accordingly!
+    if allow_single_cluster:
+        node_list = sorted(stability.keys(), reverse=True)
+    else:
+        node_list = sorted(stability.keys(), reverse=True)[:-1]
+        # (exclude root)
+
+    cluster_tree = condensed_tree[condensed_tree['cluster_size'] > 1]
+    is_cluster = {cluster: True for cluster in node_list}
+    n_samples = np.max(condensed_tree[condensed_tree['cluster_size'] == 1]['child']) + 1
+
+    if max_cluster_size is None:
+        max_cluster_size = n_samples + 1  # Set to a value that will never be triggered
+    cluster_sizes = {
+        child: cluster_size for child, cluster_size
+        in zip(cluster_tree['child'], cluster_tree['cluster_size'])
+    }
+    if allow_single_cluster:
+        # Compute cluster size for the root node
+        cluster_sizes[node_list[-1]] = np.sum(
+            cluster_tree[cluster_tree['parent'] == node_list[-1]]['cluster_size'])
+
+    if cluster_selection_method == 'eom':
+        for node in node_list:
+            child_selection = (cluster_tree['parent'] == node)
+            subtree_stability = np.sum([
+                stability[child] for
+                child in cluster_tree['child'][child_selection]])
+            if subtree_stability > stability[node] or cluster_sizes[node] > max_cluster_size:
+                is_cluster[node] = False
+                stability[node] = subtree_stability
+            else:
+                for sub_node in bfs_from_cluster_tree(cluster_tree, node):
+                    if sub_node != node:
+                        is_cluster[sub_node] = False
+
+        if cluster_selection_epsilon != 0.0 and PyArray_SHAPE(<cnp.PyArrayObject*> cluster_tree)[0] > 0:
+            eom_clusters = [c for c in is_cluster if is_cluster[c]]
+            selected_clusters = []
+            # first check if eom_clusters only has root node, which skips epsilon check.
+            if (len(eom_clusters) == 1 and eom_clusters[0] == cluster_tree['parent'].min()):
+                if allow_single_cluster:
+                    selected_clusters = eom_clusters
+            else:
+                selected_clusters = epsilon_search(
+                    set(eom_clusters),
+                    cluster_tree,
+                    cluster_selection_epsilon,
+                    allow_single_cluster
+                )
+            for c in is_cluster:
+                if c in selected_clusters:
+                    is_cluster[c] = True
+                else:
+                    is_cluster[c] = False
+
+    elif cluster_selection_method == 'leaf':
+        leaves = set(get_cluster_tree_leaves(cluster_tree))
+        if len(leaves) == 0:
+            for c in is_cluster:
+                is_cluster[c] = False
+            is_cluster[condensed_tree['parent'].min()] = True
+
+        if cluster_selection_epsilon != 0.0:
+            selected_clusters = epsilon_search(
+                leaves,
+                cluster_tree,
+                cluster_selection_epsilon,
+                allow_single_cluster
+            )
+        else:
+            selected_clusters = leaves
+
+        for c in is_cluster:
+            if c in selected_clusters:
+                is_cluster[c] = True
+            else:
+                is_cluster[c] = False
+
+    clusters = set([c for c in is_cluster if is_cluster[c]])
+    cluster_map = {c: n for n, c in enumerate(sorted(list(clusters)))}
+    reverse_cluster_map = {n: c for c, n in cluster_map.items()}
+
+    labels = _do_labelling(
+        condensed_tree,
+        clusters,
+        cluster_map,
+        allow_single_cluster,
+        cluster_selection_epsilon
+    )
+    probs = get_probabilities(condensed_tree, reverse_cluster_map, labels)
+
+    return (labels, probs)

.venv/lib/python3.12/site-packages/sklearn/cluster/_hdbscan/meson.build

@@ -0,0 +1,16 @@
+cluster_hdbscan_extension_metadata = {
+  '_linkage': {'sources': ['_linkage.pyx', metrics_cython_tree]},
+  '_reachability': {'sources': ['_reachability.pyx']},
+  '_tree': {'sources': ['_tree.pyx']}
+}
+
+foreach ext_name, ext_dict : cluster_hdbscan_extension_metadata
+  py.extension_module(
+    ext_name,
+    ext_dict.get('sources'),
+    dependencies: [np_dep],
+    cython_args: cython_args,
+    subdir: 'sklearn/cluster/_hdbscan',
+    install: true
+  )
+endforeach

.venv/lib/python3.12/site-packages/sklearn/cluster/_hdbscan/tests/test_reachibility.py

@@ -0,0 +1,63 @@
+import numpy as np
+import pytest
+
+from sklearn.cluster._hdbscan._reachability import mutual_reachability_graph
+from sklearn.utils._testing import (
+    _convert_container,
+    assert_allclose,
+)
+
+
+def test_mutual_reachability_graph_error_sparse_format():
+    """Check that we raise an error if the sparse format is not CSR."""
+    rng = np.random.RandomState(0)
+    X = rng.randn(10, 10)
+    X = X.T @ X
+    np.fill_diagonal(X, 0.0)
+    X = _convert_container(X, "sparse_csc")
+
+    err_msg = "Only sparse CSR matrices are supported"
+    with pytest.raises(ValueError, match=err_msg):
+        mutual_reachability_graph(X)
+
+
+@pytest.mark.parametrize("array_type", ["array", "sparse_csr"])
+def test_mutual_reachability_graph_inplace(array_type):
+    """Check that the operation is happening inplace."""
+    rng = np.random.RandomState(0)
+    X = rng.randn(10, 10)
+    X = X.T @ X
+    np.fill_diagonal(X, 0.0)
+    X = _convert_container(X, array_type)
+
+    mr_graph = mutual_reachability_graph(X)
+
+    assert id(mr_graph) == id(X)
+
+
+def test_mutual_reachability_graph_equivalence_dense_sparse():
+    """Check that we get the same results for dense and sparse implementation."""
+    rng = np.random.RandomState(0)
+    X = rng.randn(5, 5)
+    X_dense = X.T @ X
+    X_sparse = _convert_container(X_dense, "sparse_csr")
+
+    mr_graph_dense = mutual_reachability_graph(X_dense, min_samples=3)
+    mr_graph_sparse = mutual_reachability_graph(X_sparse, min_samples=3)
+
+    assert_allclose(mr_graph_dense, mr_graph_sparse.toarray())
+
+
+@pytest.mark.parametrize("array_type", ["array", "sparse_csr"])
+@pytest.mark.parametrize("dtype", [np.float32, np.float64])
+def test_mutual_reachability_graph_preserve_dtype(array_type, dtype):
+    """Check that the computation preserve dtype thanks to fused types."""
+    rng = np.random.RandomState(0)
+    X = rng.randn(10, 10)
+    X = (X.T @ X).astype(dtype)
+    np.fill_diagonal(X, 0.0)
+    X = _convert_container(X, array_type)
+
+    assert X.dtype == dtype
+    mr_graph = mutual_reachability_graph(X)
+    assert mr_graph.dtype == dtype

.venv/lib/python3.12/site-packages/sklearn/cluster/_hierarchical_fast.pxd

@@ -0,0 +1,9 @@
+from ..utils._typedefs cimport intp_t
+
+cdef class UnionFind:
+    cdef intp_t next_label
+    cdef intp_t[:] parent
+    cdef intp_t[:] size
+
+    cdef void union(self, intp_t m, intp_t n) noexcept
+    cdef intp_t fast_find(self, intp_t n) noexcept

.venv/lib/python3.12/site-packages/sklearn/cluster/_hierarchical_fast.pyx

@@ -0,0 +1,506 @@
+# Author: Gael Varoquaux <gael.varoquaux@normalesup.org>
+
+import numpy as np
+cimport cython
+
+from ..metrics._dist_metrics cimport DistanceMetric64
+from ..utils._fast_dict cimport IntFloatDict
+from ..utils._typedefs cimport float64_t, intp_t, uint8_t
+
+# C++
+from cython.operator cimport dereference as deref, preincrement as inc
+from libcpp.map cimport map as cpp_map
+from libc.math cimport fmax, INFINITY
+
+
+###############################################################################
+# Utilities for computing the ward momentum
+
+def compute_ward_dist(
+    const float64_t[::1] m_1,
+    const float64_t[:, ::1] m_2,
+    const intp_t[::1] coord_row,
+    const intp_t[::1] coord_col,
+    float64_t[::1] res
+):
+    cdef intp_t size_max = coord_row.shape[0]
+    cdef intp_t n_features = m_2.shape[1]
+    cdef intp_t i, j, row, col
+    cdef float64_t pa, n
+
+    for i in range(size_max):
+        row = coord_row[i]
+        col = coord_col[i]
+        n = (m_1[row] * m_1[col]) / (m_1[row] + m_1[col])
+        pa = 0.
+        for j in range(n_features):
+            pa += (m_2[row, j] / m_1[row] - m_2[col, j] / m_1[col]) ** 2
+        res[i] = pa * n
+
+
+###############################################################################
+# Utilities for cutting and exploring a hierarchical tree
+
+def _hc_get_descendent(intp_t node, children, intp_t n_leaves):
+    """
+    Function returning all the descendent leaves of a set of nodes in the tree.
+
+    Parameters
+    ----------
+    node : integer
+        The node for which we want the descendents.
+
+    children : list of pairs, length n_nodes
+        The children of each non-leaf node. Values less than `n_samples` refer
+        to leaves of the tree. A greater value `i` indicates a node with
+        children `children[i - n_samples]`.
+
+    n_leaves : integer
+        Number of leaves.
+
+    Returns
+    -------
+    descendent : list of int
+    """
+    ind = [node]
+    if node < n_leaves:
+        return ind
+    descendent = []
+
+    # It is actually faster to do the accounting of the number of
+    # elements is the list ourselves: len is a lengthy operation on a
+    # chained list
+    cdef intp_t i, n_indices = 1
+
+    while n_indices:
+        i = ind.pop()
+        if i < n_leaves:
+            descendent.append(i)
+            n_indices -= 1
+        else:
+            ind.extend(children[i - n_leaves])
+            n_indices += 1
+    return descendent
+
+
+def hc_get_heads(intp_t[:] parents, copy=True):
+    """Returns the heads of the forest, as defined by parents.
+
+    Parameters
+    ----------
+    parents : array of integers
+        The parent structure defining the forest (ensemble of trees)
+    copy : boolean
+        If copy is False, the input 'parents' array is modified inplace
+
+    Returns
+    -------
+    heads : array of integers of same shape as parents
+        The indices in the 'parents' of the tree heads
+
+    """
+    cdef intp_t parent, node0, node, size
+    if copy:
+        parents = np.copy(parents)
+    size = parents.size
+
+    # Start from the top of the tree and go down
+    for node0 in range(size - 1, -1, -1):
+        node = node0
+        parent = parents[node]
+        while parent != node:
+            parents[node0] = parent
+            node = parent
+            parent = parents[node]
+    return parents
+
+
+def _get_parents(
+    nodes,
+    heads,
+    const intp_t[:] parents,
+    uint8_t[::1] not_visited
+):
+    """Returns the heads of the given nodes, as defined by parents.
+
+    Modifies 'heads' and 'not_visited' in-place.
+
+    Parameters
+    ----------
+    nodes : list of integers
+        The nodes to start from
+    heads : list of integers
+        A list to hold the results (modified inplace)
+    parents : array of integers
+        The parent structure defining the tree
+    not_visited
+        The tree nodes to consider (modified inplace)
+
+    """
+    cdef intp_t parent, node
+
+    for node in nodes:
+        parent = parents[node]
+        while parent != node:
+            node = parent
+            parent = parents[node]
+        if not_visited[node]:
+            not_visited[node] = 0
+            heads.append(node)
+
+
+###############################################################################
+# merge strategies implemented on IntFloatDicts
+
+# These are used in the hierarchical clustering code, to implement
+# merging between two clusters, defined as a dict containing node number
+# as keys and edge weights as values.
+
+
+def max_merge(
+    IntFloatDict a,
+    IntFloatDict b,
+    const intp_t[:] mask,
+    intp_t n_a,
+    intp_t n_b
+):
+    """Merge two IntFloatDicts with the max strategy: when the same key is
+    present in the two dicts, the max of the two values is used.
+
+    Parameters
+    ==========
+    a, b : IntFloatDict object
+        The IntFloatDicts to merge
+    mask : ndarray array of dtype integer and of dimension 1
+        a mask for keys to ignore: if not mask[key] the corresponding key
+        is skipped in the output dictionary
+    n_a, n_b : float
+        n_a and n_b are weights for a and b for the merge strategy.
+        They are not used in the case of a max merge.
+
+    Returns
+    =======
+    out : IntFloatDict object
+        The IntFloatDict resulting from the merge
+    """
+    cdef IntFloatDict out_obj = IntFloatDict.__new__(IntFloatDict)
+    cdef cpp_map[intp_t, float64_t].iterator a_it = a.my_map.begin()
+    cdef cpp_map[intp_t, float64_t].iterator a_end = a.my_map.end()
+    cdef intp_t key
+    cdef float64_t value
+    # First copy a into out
+    while a_it != a_end:
+        key = deref(a_it).first
+        if mask[key]:
+            out_obj.my_map[key] = deref(a_it).second
+        inc(a_it)
+
+    # Then merge b into out
+    cdef cpp_map[intp_t, float64_t].iterator out_it = out_obj.my_map.begin()
+    cdef cpp_map[intp_t, float64_t].iterator out_end = out_obj.my_map.end()
+    cdef cpp_map[intp_t, float64_t].iterator b_it = b.my_map.begin()
+    cdef cpp_map[intp_t, float64_t].iterator b_end = b.my_map.end()
+    while b_it != b_end:
+        key = deref(b_it).first
+        value = deref(b_it).second
+        if mask[key]:
+            out_it = out_obj.my_map.find(key)
+            if out_it == out_end:
+                # Key not found
+                out_obj.my_map[key] = value
+            else:
+                deref(out_it).second = fmax(deref(out_it).second, value)
+        inc(b_it)
+    return out_obj
+
+
+def average_merge(
+    IntFloatDict a,
+    IntFloatDict b,
+    const intp_t[:] mask,
+    intp_t n_a,
+    intp_t n_b
+):
+    """Merge two IntFloatDicts with the average strategy: when the
+    same key is present in the two dicts, the weighted average of the two
+    values is used.
+
+    Parameters
+    ==========
+    a, b : IntFloatDict object
+        The IntFloatDicts to merge
+    mask : ndarray array of dtype integer and of dimension 1
+        a mask for keys to ignore: if not mask[key] the corresponding key
+        is skipped in the output dictionary
+    n_a, n_b : float
+        n_a and n_b are weights for a and b for the merge strategy.
+        They are used for a weighted mean.
+
+    Returns
+    =======
+    out : IntFloatDict object
+        The IntFloatDict resulting from the merge
+    """
+    cdef IntFloatDict out_obj = IntFloatDict.__new__(IntFloatDict)
+    cdef cpp_map[intp_t, float64_t].iterator a_it = a.my_map.begin()
+    cdef cpp_map[intp_t, float64_t].iterator a_end = a.my_map.end()
+    cdef intp_t key
+    cdef float64_t value
+    cdef float64_t n_out = <float64_t> (n_a + n_b)
+    # First copy a into out
+    while a_it != a_end:
+        key = deref(a_it).first
+        if mask[key]:
+            out_obj.my_map[key] = deref(a_it).second
+        inc(a_it)
+
+    # Then merge b into out
+    cdef cpp_map[intp_t, float64_t].iterator out_it = out_obj.my_map.begin()
+    cdef cpp_map[intp_t, float64_t].iterator out_end = out_obj.my_map.end()
+    cdef cpp_map[intp_t, float64_t].iterator b_it = b.my_map.begin()
+    cdef cpp_map[intp_t, float64_t].iterator b_end = b.my_map.end()
+    while b_it != b_end:
+        key = deref(b_it).first
+        value = deref(b_it).second
+        if mask[key]:
+            out_it = out_obj.my_map.find(key)
+            if out_it == out_end:
+                # Key not found
+                out_obj.my_map[key] = value
+            else:
+                deref(out_it).second = (n_a * deref(out_it).second
+                                        + n_b * value) / n_out
+        inc(b_it)
+    return out_obj
+
+
+###############################################################################
+# An edge object for fast comparisons
+
+cdef class WeightedEdge:
+    cdef public intp_t a
+    cdef public intp_t b
+    cdef public float64_t weight
+
+    def __init__(self, float64_t weight, intp_t a, intp_t b):
+        self.weight = weight
+        self.a = a
+        self.b = b
+
+    def __richcmp__(self, WeightedEdge other, int op):
+        """Cython-specific comparison method.
+
+        op is the comparison code::
+            <   0
+            ==  2
+            >   4
+            <=  1
+            !=  3
+            >=  5
+        """
+        if op == 0:
+            return self.weight < other.weight
+        elif op == 1:
+            return self.weight <= other.weight
+        elif op == 2:
+            return self.weight == other.weight
+        elif op == 3:
+            return self.weight != other.weight
+        elif op == 4:
+            return self.weight > other.weight
+        elif op == 5:
+            return self.weight >= other.weight
+
+    def __repr__(self):
+        return "%s(weight=%f, a=%i, b=%i)" % (self.__class__.__name__,
+                                              self.weight,
+                                              self.a, self.b)
+
+
+################################################################################
+# Efficient labelling/conversion of MSTs to single linkage hierarchies
+
+cdef class UnionFind(object):
+
+    def __init__(self, N):
+        self.parent = np.full(2 * N - 1, -1., dtype=np.intp, order='C')
+        self.next_label = N
+        self.size = np.hstack((np.ones(N, dtype=np.intp),
+                               np.zeros(N - 1, dtype=np.intp)))
+
+    cdef void union(self, intp_t m, intp_t n) noexcept:
+        self.parent[m] = self.next_label
+        self.parent[n] = self.next_label
+        self.size[self.next_label] = self.size[m] + self.size[n]
+        self.next_label += 1
+        return
+
+    @cython.wraparound(True)
+    cdef intp_t fast_find(self, intp_t n) noexcept:
+        cdef intp_t p
+        p = n
+        # find the highest node in the linkage graph so far
+        while self.parent[n] != -1:
+            n = self.parent[n]
+        # provide a shortcut up to the highest node
+        while self.parent[p] != n:
+            p, self.parent[p] = self.parent[p], n
+        return n
+
+
+def _single_linkage_label(const float64_t[:, :] L):
+    """
+    Convert an linkage array or MST to a tree by labelling clusters at merges.
+    This is done by using a Union find structure to keep track of merges
+    efficiently. This is the private version of the function that assumes that
+    ``L`` has been properly validated. See ``single_linkage_label`` for the
+    user facing version of this function.
+
+    Parameters
+    ----------
+    L: array of shape (n_samples - 1, 3)
+        The linkage array or MST where each row specifies two samples
+        to be merged and a distance or weight at which the merge occurs. This
+         array is assumed to be sorted by the distance/weight.
+
+    Returns
+    -------
+    A tree in the format used by scipy.cluster.hierarchy.
+    """
+
+    cdef float64_t[:, ::1] result_arr
+
+    cdef intp_t left, left_cluster, right, right_cluster, index
+    cdef float64_t delta
+
+    result_arr = np.zeros((L.shape[0], 4), dtype=np.float64)
+    U = UnionFind(L.shape[0] + 1)
+
+    for index in range(L.shape[0]):
+
+        left = <intp_t> L[index, 0]
+        right = <intp_t> L[index, 1]
+        delta = L[index, 2]
+
+        left_cluster = U.fast_find(left)
+        right_cluster = U.fast_find(right)
+
+        result_arr[index][0] = left_cluster
+        result_arr[index][1] = right_cluster
+        result_arr[index][2] = delta
+        result_arr[index][3] = U.size[left_cluster] + U.size[right_cluster]
+
+        U.union(left_cluster, right_cluster)
+
+    return np.asarray(result_arr)
+
+
+@cython.wraparound(True)
+def single_linkage_label(L):
+    """
+    Convert an linkage array or MST to a tree by labelling clusters at merges.
+    This is done by using a Union find structure to keep track of merges
+    efficiently.
+
+    Parameters
+    ----------
+    L: array of shape (n_samples - 1, 3)
+        The linkage array or MST where each row specifies two samples
+        to be merged and a distance or weight at which the merge occurs. This
+         array is assumed to be sorted by the distance/weight.
+
+    Returns
+    -------
+    A tree in the format used by scipy.cluster.hierarchy.
+    """
+    # Validate L
+    if L[:, :2].min() < 0 or L[:, :2].max() >= 2 * L.shape[0] + 1:
+        raise ValueError("Input MST array is not a validly formatted MST array")
+
+    is_sorted = lambda x: np.all(x[:-1] <= x[1:])
+    if not is_sorted(L[:, 2]):
+        raise ValueError("Input MST array must be sorted by weight")
+
+    return _single_linkage_label(L)
+
+
+# Implements MST-LINKAGE-CORE from https://arxiv.org/abs/1109.2378
+def mst_linkage_core(
+        const float64_t [:, ::1] raw_data,
+        DistanceMetric64 dist_metric):
+    """
+    Compute the necessary elements of a minimum spanning
+    tree for computation of single linkage clustering. This
+    represents the MST-LINKAGE-CORE algorithm (Figure 6) from
+    :arxiv:`Daniel Mullner, "Modern hierarchical, agglomerative clustering
+    algorithms" <1109.2378>`.
+
+    In contrast to the scipy implementation is never computes
+    a full distance matrix, generating distances only as they
+    are needed and releasing them when no longer needed.
+
+    Parameters
+    ----------
+    raw_data: array of shape (n_samples, n_features)
+        The array of feature data to be clustered. Must be C-aligned
+
+    dist_metric: DistanceMetric64
+        A DistanceMetric64 object conforming to the API from
+        ``sklearn.metrics._dist_metrics.pxd`` that will be
+        used to compute distances.
+
+    Returns
+    -------
+    mst_core_data: array of shape (n_samples, 3)
+        An array providing information from which one
+        can either compute an MST, or the linkage hierarchy
+        very efficiently. See :arxiv:`Daniel Mullner, "Modern hierarchical,
+        agglomerative clustering algorithms" <1109.2378>` algorithm
+        MST-LINKAGE-CORE for more details.
+    """
+    cdef:
+        intp_t n_samples = raw_data.shape[0]
+        uint8_t[:] in_tree = np.zeros(n_samples, dtype=bool)
+        float64_t[:, ::1] result = np.zeros((n_samples - 1, 3))
+
+        intp_t current_node = 0
+        intp_t new_node
+        intp_t i
+        intp_t j
+        intp_t num_features = raw_data.shape[1]
+
+        float64_t right_value
+        float64_t left_value
+        float64_t new_distance
+
+        float64_t[:] current_distances = np.full(n_samples, INFINITY)
+
+    for i in range(n_samples - 1):
+
+        in_tree[current_node] = 1
+
+        new_distance = INFINITY
+        new_node = 0
+
+        for j in range(n_samples):
+            if in_tree[j]:
+                continue
+
+            right_value = current_distances[j]
+            left_value = dist_metric.dist(&raw_data[current_node, 0],
+                                          &raw_data[j, 0],
+                                          num_features)
+
+            if left_value < right_value:
+                current_distances[j] = left_value
+
+            if current_distances[j] < new_distance:
+                new_distance = current_distances[j]
+                new_node = j
+
+        result[i, 0] = current_node
+        result[i, 1] = new_node
+        result[i, 2] = new_distance
+        current_node = new_node
+
+    return np.array(result)