TST Add non-regression test for scikit-learn#7981

jjerphan · lesteve · jjerphan · commit 15f39d1a0554 · 2023-01-19T11:17:41.000+01:00
This reproducer is adapted from the one of this message: scikit-learn#7981 (comment) Co-authored-by: Loïc Estève <loic.esteve@ymail.com>
diff --git a/sklearn/metrics/tests/test_pairwise.py b/sklearn/metrics/tests/test_pairwise.py
@@ -2,6 +2,7 @@
 from types import GeneratorType
 
 import numpy as np
+from joblib import Parallel
 from numpy import linalg
 
 from scipy.sparse import dok_matrix, csr_matrix, issparse
@@ -15,7 +16,7 @@
     # should be used instead.
     from scipy.spatial.distance import minkowski as wminkowski
 
-from sklearn.utils.fixes import sp_version, parse_version
+from sklearn.utils.fixes import sp_version, parse_version, delayed
 
 import pytest
 
@@ -1549,3 +1550,14 @@ def test_numeric_pairwise_distances_datatypes(metric, global_dtype, y_is_x):
     dist = pairwise_distances(X, Y, metric=metric, **params)
 
     assert_allclose(dist, expected_dist)
+
+
+def test_sparse_manhattan_readonly_dataset():
+    # Non-regression test for: https://github.com/scikit-learn/scikit-learn/issues/7981
+    matrices1 = [csr_matrix(np.ones((5, 5)))]
+    matrices2 = [csr_matrix(np.ones((5, 5)))]
+    # Joblib memory maps datasets which makes them read only.
+    # The following call was reporting as failing in 7981, but this must pass.
+    Parallel(n_jobs=-1, max_nbytes=0)(
+        delayed(manhattan_distances)(m1, m2) for m1, m2 in zip(matrices1, matrices2)
+    )
