Fix RandomForestRegressor default max_features (#6862)

This was a bug introduced with the removal of `max_features="auto"`. The default value for `RandomForestRegressor` should be `1.0`, while the default for `RandomForestClassifier` should be `"sqrt"`. They were both erroneously set to `"sqrt"`.

Authors:
  - Jim Crist-Harif (https://github.com/jcrist)

Approvers:
  - Philip Hyunsu Cho (https://github.com/hcho3)
  - Simon Adorf (https://github.com/csadorf)

URL: #6862

Author: jcrist

python/cuml/cuml/accel/tests/scikit-learn/xfail-list.yaml

@@ -475,6 +475,7 @@
   - "sklearn.ensemble.tests.test_forest::test_min_weight_fraction_leaf[RandomForestRegressor]"
   - "sklearn.ensemble.tests.test_forest::test_missing_value_is_predictive[RandomForestClassifier]"
   - "sklearn.ensemble.tests.test_forest::test_missing_value_is_predictive[RandomForestRegressor]"
+  - "sklearn.ensemble.tests.test_forest::test_missing_values_is_resilient[make_regression-RandomForestRegressor]"
   - "sklearn.ensemble.tests.test_forest::test_mse_criterion_object_segfault_smoke_test[RandomForestRegressor]"
   - "sklearn.ensemble.tests.test_forest::test_multioutput[RandomForestClassifier]"
   - "sklearn.ensemble.tests.test_forest::test_multioutput[RandomForestRegressor]"
@@ -484,7 +485,6 @@
   - "sklearn.ensemble.tests.test_forest::test_poisson_y_positive_check"
   - "sklearn.ensemble.tests.test_forest::test_probability[RandomForestClassifier]"
   - "sklearn.ensemble.tests.test_forest::test_regression_criterion[friedman_mse-RandomForestRegressor]"
-  - "sklearn.ensemble.tests.test_forest::test_regression_criterion[squared_error-RandomForestRegressor]"
   - "sklearn.ensemble.tests.test_forest::test_sparse_input[coo_array-RandomForestClassifier]"
   - "sklearn.ensemble.tests.test_forest::test_sparse_input[coo_array-RandomForestRegressor]"
   - "sklearn.ensemble.tests.test_forest::test_sparse_input[coo_matrix-RandomForestClassifier]"

python/cuml/cuml/ensemble/randomforestregressor.pyx

@@ -266,6 +266,7 @@ class RandomForestRegressor(BaseRandomForestModel,
     @device_interop_preparation
     def __init__(self, *,
                  split_criterion=2,
+                 max_features=1.0,
                  accuracy_metric='r2',
                  handle=None,
                  verbose=False,
@@ -274,6 +275,7 @@ class RandomForestRegressor(BaseRandomForestModel,
         self.RF_type = REGRESSION
         super().__init__(
             split_criterion=split_criterion,
+            max_features=max_features,
             accuracy_metric=accuracy_metric,
             handle=handle,
             verbose=verbose,

python/cuml/cuml/tests/explainer/test_explainer_kernel_shap.py

@@ -219,7 +219,9 @@ def test_kernel_housing_dataset(housing_dataset):
     y_train = y_train.astype(np.float32)
     y_test = y_test.astype(np.float32)
 
-    cumodel = cuml.RandomForestRegressor().fit(X_train, y_train)
+    cumodel = cuml.RandomForestRegressor(max_features="sqrt").fit(
+        X_train, y_train
+    )
 
     explainer = KernelExplainer(
         model=cumodel.predict, data=X_train[:100], output_type="numpy"

python/cuml/cuml/tests/test_random_forest.py

@@ -199,6 +199,36 @@ def special_reg(request):
     return X, y
 
 
+def test_default_parameters():
+    reg_params = curfr().get_params()
+    clf_params = curfc().get_params()
+
+    # Different default max_features
+    assert reg_params["max_features"] == 1.0
+    assert clf_params["max_features"] == "sqrt"
+
+    # Different default split_criterion
+    assert reg_params["split_criterion"] == 2
+    assert clf_params["split_criterion"] == 0
+
+    # Different accuracy_metric
+    assert reg_params["accuracy_metric"] == "r2"
+    assert clf_params["accuracy_metric"] is None
+
+    # Drop differing params
+    for name in [
+        "max_features",
+        "split_criterion",
+        "accuracy_metric",
+        "handle",
+    ]:
+        reg_params.pop(name)
+        clf_params.pop(name)
+
+    # The rest are the same
+    assert reg_params == clf_params
+
+
 @pytest.mark.parametrize("max_depth", [2, 4])
 @pytest.mark.parametrize(
     "split_criterion", ["poisson", "gamma", "inverse_gaussian"]