Update raft, use cuml normalization, add hellinger to neighors

rapidsai · Apr 1, 2021 · 4d1c1c9 · 4d1c1c9
1 parent 5cbf694
commit 4d1c1c9
Show file tree

Hide file tree

Showing 3 changed files with 7 additions and 6 deletions.
diff --git a/cpp/cmake/Dependencies.cmake b/cpp/cmake/Dependencies.cmake
@@ -39,7 +39,7 @@ else(DEFINED ENV{RAFT_PATH})
 
   ExternalProject_Add(raft
     GIT_REPOSITORY    https://github.com/rapidsai/raft.git
-    GIT_TAG           a57cf7df757b24230454e442c83f8491f97a4843
+    GIT_TAG           d1fd927bc4ec67bfd765620b5fa93f17c54cfa70
     PREFIX            ${RAFT_DIR}
     CONFIGURE_COMMAND ""
     BUILD_COMMAND     ""

diff --git a/python/cuml/neighbors/__init__.py b/python/cuml/neighbors/__init__.py
@@ -51,5 +51,5 @@
     "brute": set(["euclidean", "l2", "inner_product",
                   "l1", "cityblock", "manhattan", "taxicab",
                   "canberra", "linf", "chebyshev", "jaccard",
-                  "minkowski", "lp", "cosine"])
+                  "minkowski", "lp", "cosine", "hellinger"])
 }
diff --git a/python/cuml/test/test_metrics.py b/python/cuml/test/test_metrics.py
@@ -63,6 +63,7 @@
 from sklearn.metrics import mean_squared_log_error as sklearn_msle
 
 from cuml.common import has_scipy
+from cuml.common.sparsefuncs import csr_row_normalize_l1
 
 from cuml.metrics import roc_auc_score
 from cuml.metrics import precision_recall_curve
@@ -1119,18 +1120,18 @@ def prepare_sparse_data(size0, size1, dtype, density, metric):
     # create sparse array, then normalize every row to one
     data = cupyx.scipy.sparse.random(size0, size1,
                                      dtype=dtype,
-                                     random_state=123, density=density)
+                                     random_state=123, density=density).tocsr()
     if metric == 'hellinger':
-        data = (cupyx.scipy.sparse.diags(cp.array(1 / data.sum(1).T)[0],
-                                         0).tocoo()) * data
+        data = csr_row_normalize_l1(data)
     return data
 
 
 @pytest.mark.parametrize("metric", PAIRWISE_DISTANCE_SPARSE_METRICS.keys())
 @pytest.mark.parametrize("matrix_size, density", [
     ((3, 3), 0.7),
     ((5, 40), 0.2)])
-def test_sparse_pairwise_distances(metric: str, matrix_size, density: float):
+def test_sparse_pairwise_distances_corner_cases(metric: str, matrix_size,
+                                                density: float):
     # Test the sparse_pairwise_distance helper function.
     # Use sparse input for sklearn calls when possible
     sk_sparse = metric in ['cityblock', 'cosine', 'euclidean', 'l1', 'l2',