Transforms RandomForest estimators non-consecutive labels to consecut…

…ive labels where appropriate (rapidsai#4780)

This PR closes rapidsai#4478 by transforming non-consecutive labels outside of [0,n) to consecutive labels inside [0,n) similar to what Scikit-learn does under the hood.

Closes rapidsai#691

Authors:
  - https://github.com/VamsiTallam95

Approvers:
  - Micka (https://github.com/lowener)
  - Dante Gama Dessavre (https://github.com/dantegd)
  - Corey J. Nolet (https://github.com/cjnolet)

URL: rapidsai#4780

python/cuml/ensemble/randomforest_common.pyx

@@ -36,6 +36,7 @@ from cuml.ensemble.randomforest_shared import treelite_serialize, \
 from cuml.ensemble.randomforest_shared cimport *
 from cuml.common import input_to_cuml_array
 from cuml.common.array_descriptor import CumlArrayDescriptor
+from cuml.prims.label.classlabels import make_monotonic, check_labels
 
 
 class BaseRandomForestModel(Base):
@@ -283,11 +284,10 @@ class BaseRandomForestModel(Base):
                                 " `int32`")
             self.classes_ = cp.unique(y_m)
             self.num_classes = len(self.classes_)
-            for i in range(self.num_classes):
-                if i not in self.classes_:
-                    raise ValueError("The labels need "
-                                     "to be consecutive values from "
-                                     "0 to the number of unique label values")
+            self.use_monotonic = not check_labels(
+                y_m, cp.arange(self.num_classes, dtype=np.int32))
+            if self.use_monotonic:
+                y_m, _ = make_monotonic(y_m)
 
         else:
             y_m, _, _, y_dtype = \

python/cuml/ensemble/randomforestclassifier.pyx

@@ -46,6 +46,7 @@ from libc.stdint cimport uintptr_t, uint64_t
 from libc.stdlib cimport calloc, malloc, free
 
 from numba import cuda
+from cuml.prims.label.classlabels import check_labels, invert_labels
 
 from pylibraft.common.handle cimport handle_t
 cimport cuml.common.cuda
@@ -431,6 +432,8 @@ class RandomForestClassifier(BaseRandomForestModel,
 
         X_m, y_m, max_feature_val = self._dataset_setup_for_fit(X, y,
                                                                 convert_dtype)
+        # Track the labels to see if update is necessary
+        self.update_labels = not check_labels(y_m, self.classes_)
         cdef uintptr_t X_ptr, y_ptr
 
         X_ptr = X_m.ptr
@@ -611,6 +614,9 @@ class RandomForestClassifier(BaseRandomForestModel,
                                            fil_sparse_format=fil_sparse_format,
                                            predict_proba=False)
 
+        if self.update_labels:
+            preds = preds.to_output().astype(self.classes_.dtype)
+            preds = invert_labels(preds, self.classes_)
         return preds
 
     @insert_into_docstring(parameters=[('dense', '(n_samples, n_features)')],

python/cuml/tests/test_random_forest.py

@@ -307,6 +307,64 @@ def test_rf_classification(small_clf, datatype, max_samples, max_features):
     assert fil_acc >= (cuml_acc - 0.07)  # to be changed to 0.02. see issue #3910: https://github.com/rapidsai/cuml/issues/3910 # noqa
 
 
+@pytest.mark.parametrize(
+    "max_samples", [unit_param(1.0), quality_param(0.90), stress_param(0.95)]
+)
+@pytest.mark.parametrize("datatype", [np.float32, np.float64])
+def test_rf_classification_unorder(
+                small_clf, datatype, max_samples, max_features=1, a=2, b=5):
+    use_handle = True
+
+    X, y = small_clf
+    X = X.astype(datatype)
+    y = y.astype(np.int32)
+    # affine transformation
+    y = a*y+b
+    X_train, X_test, y_train, y_test = train_test_split(
+        X, y, train_size=0.8, random_state=0
+    )
+    # Create a handle for the cuml model
+    handle, stream = get_handle(use_handle, n_streams=1)
+
+    # Initialize, fit and predict using cuML's
+    # random forest classification model
+    cuml_model = curfc(
+        max_features=max_features,
+        max_samples=max_samples,
+        n_bins=16,
+        split_criterion=0,
+        min_samples_leaf=2,
+        random_state=123,
+        n_streams=1,
+        n_estimators=40,
+        handle=handle,
+        max_leaves=-1,
+        max_depth=16,
+    )
+    cuml_model.fit(X_train, y_train)
+
+    fil_preds = cuml_model.predict(
+        X_test, predict_model="GPU", threshold=0.5, algo="auto"
+    )
+    cu_preds = cuml_model.predict(X_test, predict_model="CPU")
+    fil_preds = np.reshape(fil_preds, np.shape(cu_preds))
+    cuml_acc = accuracy_score(y_test, cu_preds)
+    fil_acc = accuracy_score(y_test, fil_preds)
+    if X.shape[0] < 500000:
+        sk_model = skrfc(
+            n_estimators=40,
+            max_depth=16,
+            min_samples_split=2,
+            max_features=max_features,
+            random_state=10,
+        )
+        sk_model.fit(X_train, y_train)
+        sk_preds = sk_model.predict(X_test)
+        sk_acc = accuracy_score(y_test, sk_preds)
+        assert fil_acc >= (sk_acc - 0.07)
+    assert fil_acc >= (cuml_acc - 0.07)  # to be changed to 0.02. see issue #3910: https://github.com/rapidsai/cuml/issues/3910 # noqa
+
+
 @pytest.mark.parametrize(
     "max_samples", [unit_param(1.0), quality_param(0.90), stress_param(0.95)]
 )