Silhouette Score make_monotonic for non-monotonic label set (#3619)

closes #3584

Authors:
  - Divye Gala (@divyegala)

Approvers:
  - Dante Gama Dessavre (@dantegd)

URL: #3619

python/cuml/metrics/cluster/silhouette_score.pyx

@@ -24,6 +24,7 @@ from cuml.metrics.pairwise_distances import _determine_metric
 from cuml.raft.common.handle cimport handle_t
 from cuml.raft.common.handle import Handle
 from cuml.metrics.distance_type cimport DistanceType
+from cuml.prims.label.classlabels import make_monotonic, check_labels
 
 cdef extern from "cuml/metrics/metrics.hpp" namespace "ML::Metrics::Batched":
     float silhouette_score(
@@ -105,6 +106,16 @@ def _silhouette_coeff(
         labels.to_output(output_type='cupy', output_dtype='int')
     ).shape[0]
 
+    if not check_labels(labels, cp.arange(n_labels, dtype=np.int32)):
+        mono_labels, _ = make_monotonic(labels, copy=True)
+        mono_labels, _, _, _ = input_to_cuml_array(
+            mono_labels,
+            order='C',
+            convert_to_dtype=np.int32
+        )
+    else:
+        mono_labels = labels
+
     cdef uintptr_t scores_ptr
     if sil_scores is None:
         scores_ptr = <uintptr_t> NULL
@@ -122,7 +133,7 @@ def _silhouette_coeff(
                                 <float*> <uintptr_t> data.ptr,
                                 <int> n_rows,
                                 <int> n_cols,
-                                <int*> <uintptr_t> labels.ptr,
+                                <int*> <uintptr_t> mono_labels.ptr,
                                 <int> n_labels,
                                 <float*> scores_ptr,
                                 <int> chunksize,
@@ -132,7 +143,7 @@ def _silhouette_coeff(
                                 <double*> <uintptr_t> data.ptr,
                                 <int> n_rows,
                                 <int> n_cols,
-                                <int*> <uintptr_t> labels.ptr,
+                                <int*> <uintptr_t> mono_labels.ptr,
                                 <int> n_labels,
                                 <double*> scores_ptr,
                                 <int> chunksize,

python/cuml/test/test_metrics.py

@@ -270,6 +270,23 @@ def test_silhouette_samples_batched(metric, chunk_divider, labeled_clusters):
         assert False
 
 
+def test_silhouette_score_batched_non_monotonic():
+    vecs = np.array([[0.0, 0.0, 0.0], [1.0, 1.0, 1.0],
+                    [2.0, 2.0, 2.0], [10.0, 10.0, 10.0]])
+    labels = np.array([0, 0, 1, 3])
+
+    cuml_score = cu_silhouette_score(X=vecs, labels=labels)
+    sk_score = sk_silhouette_score(X=vecs, labels=labels)
+    assert_almost_equal(cuml_score, sk_score, decimal=2)
+
+    vecs = np.array([[0.0, 0.0, 0.0], [1.0, 1.0, 1.0], [10.0, 10.0, 10.0]])
+    labels = np.array([1, 1, 3])
+
+    cuml_score = cu_silhouette_score(X=vecs, labels=labels)
+    sk_score = sk_silhouette_score(X=vecs, labels=labels)
+    assert_almost_equal(cuml_score, sk_score, decimal=2)
+
+
 def score_homogeneity(ground_truth, predictions, use_handle):
     return score_labeling_with_handle(cuml.metrics.homogeneity_score,
                                       ground_truth,