MAINT: Modify to use new random API in benchmarks/sparse_csgraph_matc…

…hing.py

benchmarks/benchmarks/sparse_csgraph_matching.py

@@ -18,8 +18,8 @@ def setup(self, n, density):
         # Create random sparse matrices. Note that we could use
         # scipy.sparse.rand for this purpose, but simply using np.random and
         # disregarding duplicates is quite a bit faster.
-        np.random.seed(42)
-        d = np.random.randint(0, n, size=(int(n*n*density), 2))
+        rng = np.random.default_rng(42)
+        d = rng.integers(0, n, size=(int(n*n*density), 2))
         graph = scipy.sparse.csr_matrix((np.ones(len(d)), (d[:, 0], d[:, 1])),
                                         shape=(n, n))
         self.graph = graph
@@ -31,29 +31,29 @@ def time_maximum_bipartite_matching(self, n, density):
 # For benchmarking min_weight_full_bipartite_matching, we rely on some of
 # the classes defined in Burkard, Dell'Amico, Martello -- Assignment Problems,
 # 2009, Section 4.10.1.
-def random_uniform(shape):
-    return scipy.sparse.csr_matrix(np.random.uniform(1, 100, shape))
+def random_uniform(shape, rng):
+    return scipy.sparse.csr_matrix(rng.uniform(1, 100, shape))
 
 
-def random_uniform_sparse(shape):
-    return scipy.sparse.random(shape[0], shape[1], density=0.1, format='csr')
+def random_uniform_sparse(shape, rng):
+    return scipy.sparse.random(shape[0], shape[1], density=0.1, format='csr', random_state=rng)
 
 
-def random_uniform_integer(shape):
-    return scipy.sparse.csr_matrix(np.random.randint(1, 1000, shape))
+def random_uniform_integer(shape, rng):
+    return scipy.sparse.csr_matrix(rng.integers(1, 1000, shape))
 
 
-def random_geometric(shape):
-    P = np.random.randint(1, 1000, size=(shape[0], 2))
-    Q = np.random.randint(1, 1000, size=(shape[1], 2))
+def random_geometric(shape, rng):
+    P = rng.integers(1, 1000, size=(shape[0], 2))
+    Q = rng.integers(1, 1000, size=(shape[1], 2))
     return scipy.sparse.csr_matrix(cdist(P, Q, 'sqeuclidean'))
 
 
-def random_two_cost(shape):
-    return scipy.sparse.csr_matrix(np.random.choice((1, 1000000), shape))
+def random_two_cost(shape, rng):
+    return scipy.sparse.csr_matrix(rng.choice((1, 1000000), shape))
 
 
-def machol_wien(shape):
+def machol_wien(shape, rng):
     # Machol--Wien instances being harder than the other examples, we cut
     # down the size of the instance by 5.
     return scipy.sparse.csr_matrix(
@@ -71,15 +71,15 @@ class MinWeightFullBipartiteMatching(Benchmark):
     ]
 
     def setup(self, shape, input_type):
-        np.random.seed(42)
+        rng = np.random.default_rng(42)
         input_func = {'random_uniform': random_uniform,
                       'random_uniform_sparse': random_uniform_sparse,
                       'random_uniform_integer': random_uniform_integer,
                       'random_geometric': random_geometric,
                       'random_two_cost': random_two_cost,
                       'machol_wien': machol_wien}[input_type]
 
-        self.biadjacency_matrix = input_func(shape)
+        self.biadjacency_matrix = input_func(shape, rng)
 
     def time_evaluation(self, *args):
         min_weight_full_bipartite_matching(self.biadjacency_matrix)