Merge pull request scipy#23739 from mdhaber/remove_seed

MAINT: stats: remove use of `np.random.seed`

Author: andyfaff

scipy/stats/tests/common_tests.py

@@ -197,12 +197,8 @@ def check_named_args(distfn, x, shape_args, defaults, meths):
 def check_random_state_property(distfn, args):
     # check the random_state attribute of a distribution *instance*
 
-    # This test fiddles with distfn.random_state. This breaks other tests,
-    # hence need to save it and then restore.
-    rndm = distfn.random_state
-
     # baseline: this relies on the global state
-    np.random.seed(1234)
+    np.random.seed(1234)  # valid use of np.random.seed
     distfn.random_state = None
     r0 = distfn.rvs(*args, size=8)
 
@@ -231,9 +227,6 @@ def check_random_state_property(distfn, args):
     # ... and that does not alter the instance-level random_state!
     npt.assert_equal(distfn.random_state.get_state(), orig_state)
 
-    # finally, restore the random_state
-    distfn.random_state = rndm
-
 
 def check_meth_dtype(distfn, arg, meths):
     q0 = [0.25, 0.5, 0.75]

scipy/stats/tests/test_axis_nan_policy.py

@@ -989,9 +989,9 @@ def test_non_broadcastable(hypotest, args, kwds, n_samples, n_outputs, paired,
 
 def test_masked_array_2_sentinel_array():
     # prepare arrays
-    np.random.seed(0)
-    A = np.random.rand(10, 11, 12)
-    B = np.random.rand(12)
+    rng = np.random.default_rng(805715284)
+    A = rng.random((10, 11, 12))
+    B = rng.random(12)
     mask = A < 0.5
     A = np.ma.masked_array(A, mask)
 
@@ -1122,10 +1122,10 @@ def test_masked_stat_1d():
 @pytest.mark.parametrize(("axis"), range(-3, 3))
 def test_masked_stat_3d(axis):
     # basic test of _axis_nan_policy_factory with 3D masked sample
-    np.random.seed(0)
-    a = np.random.rand(3, 4, 5)
-    b = np.random.rand(4, 5)
-    c = np.random.rand(4, 1)
+    rng = np.random.default_rng(3679428403)
+    a = rng.random((3, 4, 5))
+    b = rng.random((4, 5))
+    c = rng.random((4, 1))
 
     mask_a = a < 0.1
     mask_c = [False, False, False, True]

scipy/stats/tests/test_binned_statistic.py

@@ -159,9 +159,9 @@ def test_1d_bincode(self):
 
     def test_1d_range_keyword(self):
         # Regression test for gh-3063, range can be (min, max) or [(min, max)]
-        np.random.seed(9865)
+        rng = np.random.default_rng(6823616729)
         x = np.arange(30)
-        data = np.random.random(30)
+        data = rng.random(30)
 
         mean, bins, _ = binned_statistic(x[:15], data[:15])
         mean_range, bins_range, _ = binned_statistic(x, data, range=[(0, 14)])
@@ -479,8 +479,9 @@ def test_dd_binnumbers_unraveled(self):
 
     def test_dd_binned_statistic_result(self):
         # NOTE: tests the reuse of bin_edges from previous call
-        x = np.random.random((10000, 3))
-        v = np.random.random(10000)
+        rng = np.random.default_rng(8111360615)
+        x = rng.random((10000, 3))
+        v = rng.random(10000)
         bins = np.linspace(0, 1, 10)
         bins = (bins, bins, bins)
 
@@ -494,8 +495,9 @@ def test_dd_binned_statistic_result(self):
         assert_allclose(stat, stat2)
 
     def test_dd_zero_dedges(self):
-        x = np.random.random((10000, 3))
-        v = np.random.random(10000)
+        rng = np.random.default_rng(1132724173)
+        x = rng.random((10000, 3))
+        v = rng.random(10000)
         bins = np.linspace(0, 1, 10)
         bins = np.append(bins, 1)
         bins = (bins, bins, bins)

scipy/stats/tests/test_continuous_basic.py

@@ -587,7 +587,6 @@ def test_gh1320_regression():
 
 def test_method_of_moments():
     # example from https://en.wikipedia.org/wiki/Method_of_moments_(statistics)
-    np.random.seed(1234)
     x = [0, 0, 0, 0, 1]
     a = 1/5 - 2*np.sqrt(3)/5
     b = 1/5 + 2*np.sqrt(3)/5

scipy/stats/tests/test_discrete_basic.py

@@ -168,8 +168,9 @@ def test_ppf_with_loc(dist, args):
     except TypeError:
         distfn = dist
     #check with a negative, no and positive relocation.
-    np.random.seed(1942349)
-    re_locs = [np.random.randint(-10, -1), 0, np.random.randint(1, 10)]
+    rng = np.random.default_rng(5108587887)
+
+    re_locs = [rng.integers(-10, -1), 0, rng.integers(1, 10)]
     _a, _b = distfn.support(*args)
     for loc in re_locs:
         npt.assert_array_equal(
@@ -185,17 +186,17 @@ def test_isf_with_loc(dist, args):
     except TypeError:
         distfn = dist
     # check with a negative, no and positive relocation.
-    np.random.seed(1942349)
-    re_locs = [np.random.randint(-10, -1), 0, np.random.randint(1, 10)]
+    rng = np.random.default_rng(4030503535)
+    re_locs = [rng.integers(-10, -1), 0, rng.integers(1, 10)]
     _a, _b = distfn.support(*args)
     for loc in re_locs:
         expected = _b + loc, _a - 1 + loc
         res = distfn.isf(0., *args, loc=loc), distfn.isf(1., *args, loc=loc)
         npt.assert_array_equal(expected, res)
     # test broadcasting behaviour
-    re_locs = [np.random.randint(-10, -1, size=(5, 3)),
+    re_locs = [rng.integers(-10, -1, size=(5, 3)),
                np.zeros((5, 3)),
-               np.random.randint(1, 10, size=(5, 3))]
+               rng.integers(1, 10, size=(5, 3))]
     _a, _b = distfn.support(*args)
     for loc in re_locs:
         expected = _b + loc, _a - 1 + loc

scipy/stats/tests/test_discrete_distns.py

@@ -177,17 +177,18 @@ def test_issue_11134():
 
 
 def test_issue_7406():
-    np.random.seed(0)
-    assert_equal(binom.ppf(np.random.rand(10), 0, 0.5), 0)
+    rng = np.random.default_rng(4763112764)
+    assert_equal(binom.ppf(rng.random(10), 0, 0.5), 0)
 
     # Also check that endpoints (q=0, q=1) are correct
     assert_equal(binom.ppf(0, 0, 0.5), -1)
     assert_equal(binom.ppf(1, 0, 0.5), 0)
 
 
 def test_issue_5122():
+    rng = np.random.default_rng(8312492117)
     p = 0
-    n = np.random.randint(100, size=10)
+    n = rng.integers(100, size=10)
 
     x = 0
     ppf = binom.ppf(x, n, p)
@@ -397,17 +398,19 @@ def test_mean_against_mpmath(self, a, n, ref):
 
 
 class TestNCH:
-    np.random.seed(2)  # seeds 0 and 1 had some xl = xu; randint failed
-    shape = (2, 4, 3)
-    max_m = 100
-    m1 = np.random.randint(1, max_m, size=shape)    # red balls
-    m2 = np.random.randint(1, max_m, size=shape)    # white balls
-    N = m1 + m2                                     # total balls
-    n = randint.rvs(0, N, size=N.shape)             # number of draws
-    xl = np.maximum(0, n-m2)                        # lower bound of support
-    xu = np.minimum(n, m1)                          # upper bound of support
-    x = randint.rvs(xl, xu, size=xl.shape)
-    odds = np.random.rand(*x.shape)*2
+    def setup_method(self):
+        rng = np.random.default_rng(7162434334)
+        shape = (2, 4, 3)
+        max_m = 100
+        m1 = rng.integers(1, max_m, size=shape)  # red balls
+        m2 = rng.integers(1, max_m, size=shape)  # white balls
+        N = m1 + m2  # total balls
+        n = randint.rvs(0, N, size=N.shape, random_state=rng)  # number of draws
+        xl = np.maximum(0, n-m2)  # lower bound of support
+        xu = np.minimum(n, m1)  # upper bound of support
+        x = randint.rvs(xl, xu, size=xl.shape, random_state=rng)
+        odds = rng.random(x.shape)*2
+        self.x, self.N, self.m1, self.n, self.odds = x, N, m1, n, odds
 
     # test output is more readable when function names (strings) are passed
     @pytest.mark.parametrize('dist_name',