Copy attributes from the original class to the proxy (#6306)

This makes the proxy look and feel more like the original class.

This copies several attributes, not just `__name__` and makes sure the signature of the constructor looks like that of the scikit-learn class. This is useful for people who inspect it during coding (like some editors) or `ipython`'s `LogisticRegression?`.

~~For some reason the test in `test_basic_estimator.py` fails because the proxy doesn't have a `__name__` attribute. Trying to work out why, if someone can spot the mistake let me know~~

This is an alternative/extended version of #6305

Authors:
  - Tim Head (https://github.com/betatim)

Approvers:
  - Simon Adorf (https://github.com/csadorf)

URL: #6306

python/cuml/cuml/experimental/accel/estimator_proxy.py

@@ -1,5 +1,5 @@
 #
-# Copyright (c) 2024, NVIDIA CORPORATION.
+# Copyright (c) 2024-2025, NVIDIA CORPORATION.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -277,6 +277,26 @@ def __reduce__(self):
                 ),
             )
 
+    # Help make the proxy class look more like the original class
+    for attr in (
+        "__module__",
+        "__name__",
+        "__qualname__",
+        "__doc__",
+        "__annotate__",
+        "__type_params__",
+    ):
+        try:
+            value = getattr(original_class_a, attr)
+        except AttributeError:
+            pass
+        else:
+            setattr(ProxyEstimator, attr, value)
+
+    ProxyEstimator.__init__.__signature__ = inspect.signature(
+        original_class_a.__init__
+    )
+
     logger.debug(
         f"Created proxy estimator: ({module_b}, {original_class_name}, {ProxyEstimator})"
     )

python/cuml/cuml/tests/experimental/accel/test_basic_estimators.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2024, NVIDIA CORPORATION.
+# Copyright (c) 2024-2025, NVIDIA CORPORATION.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -140,3 +140,32 @@ def test_k_neighbors_regressor():
         for metric in ["euclidean", "manhattan"]:
             knr = KNeighborsRegressor().fit(X, y)
             knr.predict(X)
+
+
+def test_proxy_facade():
+    # Check that the proxy estimator pretends to look like the
+    # class it is proxying
+
+    # A random estimator, shouldn't matter which one as all are proxied
+    # the same way.
+    # We need an instance to get access to the `_cpu_model_class`
+    # but we want to compare to the PCA class
+    pca = PCA()
+    for attr in (
+        "__module__",
+        "__name__",
+        "__qualname__",
+        "__doc__",
+        "__annotate__",
+        "__type_params__",
+    ):
+        # if the original class has this attribute, the proxy should
+        # have it as well and the values should match
+        try:
+            original_value = getattr(pca._cpu_model_class, attr)
+        except AttributeError:
+            pass
+        else:
+            proxy_value = getattr(PCA, attr)
+
+            assert original_value == proxy_value

python/cuml/cuml/tests/experimental/accel/test_pipeline.py

@@ -1,5 +1,5 @@
 #
-# Copyright (c) 2024, NVIDIA CORPORATION.
+# Copyright (c) 2024-2025, NVIDIA CORPORATION.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -31,7 +31,7 @@
     KNeighborsClassifier,
     KNeighborsRegressor,
 )
-from sklearn.pipeline import Pipeline
+from sklearn.pipeline import Pipeline, make_pipeline
 from sklearn.datasets import make_classification, make_regression
 from sklearn.model_selection import train_test_split
 from sklearn.metrics import accuracy_score, mean_squared_error
@@ -145,3 +145,12 @@ def test_umap_with_logistic_regression(classification_data):
     # Fit and predict
     pipeline.fit(X_train, y_train)
     pipeline.predict(X_test)
+
+
+def test_automatic_step_naming():
+    # The automatically generated names of estimators should be the
+    # same with and without accelerator.
+    pipeline = make_pipeline(PCA(), LogisticRegression())
+
+    assert "pca" in pipeline.named_steps
+    assert "logisticregression" in pipeline.named_steps