Deprecate convert_to_* methods in favor of as_* methods in cuml.ensemble

docs/source/pickling_cuml_models.ipynb

@@ -199,7 +199,7 @@
     "\n",
     "Here are the steps to export the model:\n",
     "\n",
-    "1. Call `convert_to_treelite_model().serialize()` to obtain the checkpoint file from the cuML Random Forest model."
+    "1. Call `as_treelite().serialize()` to obtain the checkpoint file from the cuML Random Forest model."
    ]
   },
   {
@@ -219,7 +219,7 @@
     "\n",
     "checkpoint_path = './checkpoint.tl'\n",
     "# Export cuML RF model as Treelite checkpoint\n",
-    "clf.convert_to_treelite_model().serialize(checkpoint_path)"
+    "clf.as_treelite().serialize(checkpoint_path)"
    ]
   },
   {

python/cuml/cuml/ensemble/randomforest_common.pyx

@@ -343,20 +343,39 @@ class BaseRandomForestModel(Base, InteropMixin):
         return self.n_estimators
 
     def convert_to_treelite_model(self):
+        """Deprecated, use `as_treelite`."""
+        warnings.warn(
+            "`convert_to_treelite_model` was deprecated in 25.10 and will be "
+            "removed in 25.12. Please use `as_treelite` instead.",
+            FutureWarning,
+        )
+        return self.as_treelite()
+
+    def convert_to_fil_model(
+        self, layout="depth_first", default_chunk_size=None, align_bytes=None
+    ):
+        """Deprecated, use `as_fil`."""
+        warnings.warn(
+            "`convert_to_fil_model` was deprecated in 25.10 and will be "
+            "removed in 25.12. Please use `as_fil` instead.",
+            FutureWarning,
+        )
+        return self.as_fil(
+            layout=layout, default_chunk_size=default_chunk_size, align_bytes=align_bytes
+        )
+
+    def as_treelite(self):
         """
-        Converts the cuML RF model to a Treelite model
+        Converts this estimator to a Treelite model.
 
         Returns
         -------
-        tl_to_fil_model : treelite.Model
+        treelite.Model
         """
         return treelite.Model.deserialize_bytes(self._treelite_model_bytes)
 
-    def convert_to_fil_model(
-        self,
-        layout="depth_first",
-        default_chunk_size=None,
-        align_bytes=None,
+    def as_fil(
+        self, layout="depth_first", default_chunk_size=None, align_bytes=None,
     ):
         """
         Create a Forest Inference (FIL) model from the trained cuML
@@ -501,13 +520,13 @@ class BaseRandomForestModel(Base, InteropMixin):
                         verbose
                     )
 
-        # XXX: Theoretically we could wrap `tl_handle` with `treelite.Model` to manage
-        # ownership, and keep the loaded model around. However, this only works if the
-        # `libtreelite` used by `treelite` matches the one that `cuml` is linked against.
-        # This is currently true for conda environments, but not for wheels where
-        # `treelite` contains its own separate version. So for now we need to do this
-        # serialize-and-reload dance. If/when this is fixed we instead store the loaded
-        # model and use that everywhere.
+        # XXX: Theoretically we could wrap `tl_handle` with `treelite.Model` to
+        # manage ownership, and keep the loaded model around. However, this
+        # only works if the `libtreelite` is ABI compatible with the one used
+        # by `cuml`. This is currently true for conda environments, but not for
+        # wheels where `cuml` and `treelite` use different manylinux ABIs. So
+        # for now we need to do this serialize-and-reload dance. If/when this
+        # is fixed we could instead store the loaded model and use that instead.
         cdef const char* tl_bytes = NULL
         cdef size_t tl_bytes_len
         safe_treelite_call(
@@ -540,9 +559,9 @@ class BaseRandomForestModel(Base, InteropMixin):
         ):
             # default parameters, get (or create) the cached fil model
             if (fil_model := getattr(self, "_fil_model", None)) is None:
-                fil_model = self._fil_model = self.convert_to_fil_model()
+                fil_model = self._fil_model = self.as_fil()
         else:
-            fil_model = self.convert_to_fil_model(
+            fil_model = self.as_fil(
                 layout=layout,
                 default_chunk_size=default_chunk_size,
                 align_bytes=align_bytes,
@@ -554,7 +573,7 @@ class BaseRandomForestModel(Base, InteropMixin):
             warnings.warn(
                 (
                     "`predict_model` is deprecated (and ignored) and will be removed "
-                    "in 25.12. To infer on CPU use `model.convert_to_fil_model` to get "
+                    "in 25.12. To infer on CPU use `model.as_fil` to get "
                     "a `FIL` instance which may then be used to perform inference on "
                     "both CPU and GPU."
                 ),

python/cuml/cuml/ensemble/randomforestclassifier.py

@@ -279,9 +279,8 @@ def predict(
 
             .. deprecated:: 25.10
                 `predict_model` is deprecated (and ignored) and will be removed
-                in 25.12. To infer on CPU use `model.convert_to_fil_model` to get
-                a `FIL` instance which may then be used to perform inference on
-                both CPU and GPU.
+                in 25.12. To infer on CPU use `model.as_fil` to get a `FIL` instance
+                which may then be used to perform inference on both CPU and GPU.
 
         Returns
         -------
@@ -400,9 +399,8 @@ def score(
 
             .. deprecated:: 25.10
                 `predict_model` is deprecated (and ignored) and will be removed
-                in 25.12. To infer on CPU use `model.convert_to_fil_model` to get
-                a `FIL` instance which may then be used to perform inference on
-                both CPU and GPU.
+                in 25.12. To infer on CPU use `model.as_fil` to get a `FIL` instance
+                which may then be used to perform inference on both CPU and GPU.
 
         Returns
         -------

python/cuml/cuml/ensemble/randomforestregressor.py

@@ -265,9 +265,8 @@ def predict(
 
             .. deprecated:: 25.10
                 `predict_model` is deprecated (and ignored) and will be removed
-                in 25.12. To infer on CPU use `model.convert_to_fil_model` to get
-                a `FIL` instance which may then be used to perform inference on
-                both CPU and GPU.
+                in 25.12. To infer on CPU use `model.as_fil` to get a `FIL` instance
+                which may then be used to perform inference on both CPU and GPU.
 
         Returns
         -------
@@ -337,9 +336,8 @@ def score(
 
             .. deprecated:: 25.10
                 `predict_model` is deprecated (and ignored) and will be removed
-                in 25.12. To infer on CPU use `model.convert_to_fil_model` to get
-                a `FIL` instance which may then be used to perform inference on
-                both CPU and GPU.
+                in 25.12. To infer on CPU use `model.as_fil` to get a `FIL` instance
+                which may then be used to perform inference on both CPU and GPU.
 
         Returns
         -------

python/cuml/cuml/explainer/tree_shap.pyx

@@ -175,7 +175,7 @@ cdef class TreeExplainer:
             tl_model = treelite.frontend.from_lightgbm(model)
         # cuML RF model object
         elif isinstance(model, (cuml.RandomForestClassifier, cuml.RandomForestRegressor)):
-            tl_model = model.convert_to_treelite_model()
+            tl_model = model.as_treelite()
         # scikit-learn RF model object
         elif isinstance(model, treelite.Model):
             tl_model = model

python/cuml/tests/dask/test_dask_random_forest.py

@@ -362,7 +362,7 @@ def test_rf_data_count(client, max_depth, n_estimators):
     X_dask, y_dask = _prep_training_data(client, X, y, partitions_per_worker=2)
     dask_model.fit(X_dask, y_dask)
     model = dask_model.get_combined_model()
-    json_obj = json.loads(model.convert_to_treelite_model().dump_as_json())
+    json_obj = json.loads(model.as_treelite().dump_as_json())
 
     def check_count(node, nodes):
         if "left_child" in node:

python/cuml/tests/test_fil.py

@@ -945,7 +945,7 @@ def test_device_selection(device_id, model_kind, tmp_path):
                 n_streams=1,
             )
             cuml_model.fit(cp.array(X), cp.array(y))
-            fm = cuml_model.convert_to_fil_model()
+            fm = cuml_model.as_fil()
     else:
         raise NotImplementedError()
 

python/cuml/tests/test_random_forest.py

@@ -707,14 +707,14 @@ def test_rf_classification_sparse(small_clf, datatype, fil_layout):
     acc = accuracy_score(y_test, preds)
     np.testing.assert_almost_equal(acc, cuml_model.score(X_test, y_test))
 
-    fil_model = cuml_model.convert_to_fil_model()
+    fil_model = cuml_model.as_fil()
 
     with cuml.using_output_type("numpy"):
         fil_model_preds = fil_model.predict(X_test)
         fil_model_acc = accuracy_score(y_test, fil_model_preds)
         assert acc == fil_model_acc
 
-    tl_model = cuml_model.convert_to_treelite_model()
+    tl_model = cuml_model.as_treelite()
     assert num_trees == tl_model.num_tree
     assert X.shape[1] == tl_model.num_feature
 
@@ -771,15 +771,15 @@ def test_rf_regression_sparse(special_reg, datatype, fil_layout):
     preds = cuml_model.predict(X_test, layout=fil_layout)
     r2 = r2_score(y_test, preds)
 
-    fil_model = cuml_model.convert_to_fil_model()
+    fil_model = cuml_model.as_fil()
 
     with cuml.using_output_type("numpy"):
         fil_model_preds = fil_model.predict(X_test)
         fil_model_preds = np.reshape(fil_model_preds, np.shape(y_test))
         fil_model_r2 = r2_score(y_test, fil_model_preds)
         assert r2 == fil_model_r2
 
-    tl_model = cuml_model.convert_to_treelite_model()
+    tl_model = cuml_model.as_treelite()
     assert num_trees == tl_model.num_tree
     assert X.shape[1] == tl_model.num_feature
 
@@ -1092,9 +1092,7 @@ def test_rf_regression_with_identical_labels():
         max_depth=1,
     )
     model.fit(X, y)
-    trees = json.loads(model.convert_to_treelite_model().dump_as_json())[
-        "trees"
-    ]
+    trees = json.loads(model.as_treelite().dump_as_json())["trees"]
     assert len(trees) == 1
     assert len(trees[0]["nodes"]) == 1
     assert trees[0]["nodes"][0] == {
@@ -1112,7 +1110,7 @@ def test_rf_regressor_gtil_integration(tmpdir):
     expected_pred = clf.predict(X).reshape((-1, 1, 1))
 
     checkpoint_path = os.path.join(tmpdir, "checkpoint.tl")
-    clf.convert_to_treelite_model().serialize(checkpoint_path)
+    clf.as_treelite().serialize(checkpoint_path)
 
     tl_model = treelite.Model.deserialize(checkpoint_path)
     out_pred = treelite.gtil.predict(tl_model, X)
@@ -1127,7 +1125,7 @@ def test_rf_binary_classifier_gtil_integration(tmpdir):
     expected_pred = clf.predict_proba(X).reshape((-1, 1, 2))
 
     checkpoint_path = os.path.join(tmpdir, "checkpoint.tl")
-    clf.convert_to_treelite_model().serialize(checkpoint_path)
+    clf.as_treelite().serialize(checkpoint_path)
 
     tl_model = treelite.Model.deserialize(checkpoint_path)
     out_pred = treelite.gtil.predict(tl_model, X)
@@ -1142,7 +1140,7 @@ def test_rf_multiclass_classifier_gtil_integration(tmpdir):
     expected_prob = clf.predict_proba(X).reshape((X.shape[0], 1, -1))
 
     checkpoint_path = os.path.join(tmpdir, "checkpoint.tl")
-    clf.convert_to_treelite_model().serialize(checkpoint_path)
+    clf.as_treelite().serialize(checkpoint_path)
 
     tl_model = treelite.Model.deserialize(checkpoint_path)
     out_prob = treelite.gtil.predict(tl_model, X, pred_margin=True)
@@ -1256,3 +1254,18 @@ def test_accuracy_metric_deprecated():
         model = cuml.RandomForestRegressor(accuracy_metric="mse")
     score = model.fit(X, y).score(X, y)
     np.testing.assert_allclose(score, mean_squared_error(y, model.predict(X)))
+
+
+def test_convert_methods_deprecated():
+    X, y = make_regression(n_samples=500)
+    model = cuml.RandomForestRegressor().fit(X, y)
+
+    with pytest.warns(FutureWarning, match="convert_to_treelite_model"):
+        tl = model.convert_to_treelite_model()
+
+    assert isinstance(tl, treelite.Model)
+
+    with pytest.warns(FutureWarning, match="convert_to_fil_model"):
+        fil = model.convert_to_fil_model()
+
+    assert isinstance(fil, cuml.fil.ForestInference)