Use docstring-parser instead of numpydoc.

conda/environments/all_cuda-129_arch-aarch64.yaml

@@ -22,6 +22,7 @@ dependencies:
 - dask-cuda==26.4.*,>=0.0.0a0
 - dask-cudf==26.4.*,>=0.0.0a0
 - dask-ml>=2024
+- docstring_parser
 - doxygen=1.9.1
 - gcc_linux-aarch64=14.*
 - graphviz
@@ -46,8 +47,6 @@ dependencies:
 - numba-cuda>=0.22.1
 - numba>=0.60.0,<0.62.0
 - numpy>=1.23,<3.0
-- numpydoc
-- numpydoc<1.9
 - nvidia-ml-py>=12
 - packaging
 - pre-commit

conda/environments/all_cuda-129_arch-x86_64.yaml

@@ -22,6 +22,7 @@ dependencies:
 - dask-cuda==26.4.*,>=0.0.0a0
 - dask-cudf==26.4.*,>=0.0.0a0
 - dask-ml>=2024
+- docstring_parser
 - doxygen=1.9.1
 - gcc_linux-64=14.*
 - graphviz
@@ -46,8 +47,6 @@ dependencies:
 - numba-cuda>=0.22.1
 - numba>=0.60.0,<0.62.0
 - numpy>=1.23,<3.0
-- numpydoc
-- numpydoc<1.9
 - nvidia-ml-py>=12
 - packaging
 - pre-commit

conda/environments/all_cuda-131_arch-aarch64.yaml

@@ -22,6 +22,7 @@ dependencies:
 - dask-cuda==26.4.*,>=0.0.0a0
 - dask-cudf==26.4.*,>=0.0.0a0
 - dask-ml>=2024
+- docstring_parser
 - doxygen=1.9.1
 - gcc_linux-aarch64=14.*
 - graphviz
@@ -46,8 +47,6 @@ dependencies:
 - numba-cuda>=0.22.1
 - numba>=0.60.0,<0.62.0
 - numpy>=1.23,<3.0
-- numpydoc
-- numpydoc<1.9
 - nvidia-ml-py>=12
 - packaging
 - pre-commit

conda/environments/all_cuda-131_arch-x86_64.yaml

@@ -22,6 +22,7 @@ dependencies:
 - dask-cuda==26.4.*,>=0.0.0a0
 - dask-cudf==26.4.*,>=0.0.0a0
 - dask-ml>=2024
+- docstring_parser
 - doxygen=1.9.1
 - gcc_linux-64=14.*
 - graphviz
@@ -46,8 +47,6 @@ dependencies:
 - numba-cuda>=0.22.1
 - numba>=0.60.0,<0.62.0
 - numpy>=1.23,<3.0
-- numpydoc
-- numpydoc<1.9
 - nvidia-ml-py>=12
 - packaging
 - pre-commit

dependencies.yaml

@@ -400,7 +400,6 @@ dependencies:
           - ipython
           - ipykernel
           - nbsphinx
-          - numpydoc
           # https://github.com/pydata/pydata-sphinx-theme/issues/1539
           - pydata-sphinx-theme!=0.14.2
           - recommonmark
@@ -496,8 +495,7 @@ dependencies:
           - hdbscan>=0.8.39,<0.8.40
           - hypothesis>=6.0,<7
           - nltk
-          # upstream sklearn docstring tests require numpydoc<1.9
-          - numpydoc<1.9
+          - docstring_parser
           # 'nvidia-ml-py' provides the 'pynvml' module
           - nvidia-ml-py>=12
           - pyyaml

python/cuml/pyproject.toml

@@ -108,11 +108,11 @@ classifiers = [
 test = [
     "certifi",
     "cython>=3.0.0,<3.2.0",
+    "docstring_parser",
     "hdbscan>=0.8.39,<0.8.40",
     "hypothesis>=6.0,<7",
     "ipython>=7.32.0",
     "nltk",
-    "numpydoc<1.9",
     "nvidia-ml-py>=12",
     "pynndescent",
     "pytest-benchmark",

python/cuml/tests/test_base.py

@@ -4,10 +4,11 @@
 import inspect
 
 import numpy as np
-import numpydoc.docscrape
 import pandas as pd
 import pylibraft.common.handle
 import pytest
+from docstring_parser import DocstringStyle
+from docstring_parser import parse as parse_docstring
 from sklearn.datasets import (
     make_classification,
     make_multilabel_classification,
@@ -74,8 +75,8 @@ class that derives from `Base`, We ensure that 1) the base arguments exist
     # verbose: r"^[ ]{4}verbose :.*\n(^(?![ ]{0,4}(?![ ]{4,})).*(\n))+"
     # handle: r"^[ ]{4}handle :.*\n(^(?![ ]{0,4}(?![ ]{4,})).*(\n))+"
 
-    def get_param_doc(param_doc_obj, name: str):
-        found_doc = next((x for x in param_doc_obj if x.name == name), None)
+    def get_param_doc(param_list, name: str):
+        found_doc = next((x for x in param_list if x.arg_name == name), None)
 
         assert found_doc is not None, "Could not find {} in docstring".format(
             name
@@ -85,13 +86,17 @@ def get_param_doc(param_doc_obj, name: str):
 
     # Load the base class signature, parse the docstring and pull out params
     base_sig = inspect.signature(cuml.Base, follow_wrapped=True)
-    base_doc = numpydoc.docscrape.NumpyDocString(cuml.Base.__doc__)
-    base_doc_params = base_doc["Parameters"]
+    base_parsed = parse_docstring(
+        cuml.Base.__doc__ or "", style=DocstringStyle.NUMPYDOC
+    )
+    base_doc_params = base_parsed.params or []
 
     # Load the current class signature, parse the docstring and pull out params
     klass_sig = inspect.signature(klass, follow_wrapped=True)
-    klass_doc = numpydoc.docscrape.NumpyDocString(klass.__doc__ or "")
-    klass_doc_params = klass_doc["Parameters"]
+    klass_parsed = parse_docstring(
+        klass.__doc__ or "", style=DocstringStyle.NUMPYDOC
+    )
+    klass_doc_params = klass_parsed.params or []
 
     for name, param in base_sig.parameters.items():
         # Ensure the base param exists in the derived
@@ -113,12 +118,13 @@ def get_param_doc(param_doc_obj, name: str):
 
             base_item_doc = get_param_doc(base_doc_params, name)
 
-            assert found_doc.type == base_item_doc.type, (
+            assert found_doc.type_name == base_item_doc.type_name, (
                 f"Docstring mismatch for {name}"
             )
 
-            found = " ".join(found_doc.desc)
-            expected = " ".join(base_item_doc.desc)
+            # Normalize whitespace for comparison (content only, not formatting)
+            found = " ".join((found_doc.description or "").split())
+            expected = " ".join((base_item_doc.description or "").split())
 
             assert found == expected, f"Docstring mismatch for {name}"