TST: Set seed when using basin hopping

Basin hopping uses the global RandomState by default.  Add a fixture
to set the global state to a fixed and then restore it to its original
value after the test has completed.

xref statsmodels#4230

statsmodels/conftest.py

@@ -1,3 +1,4 @@
+import numpy as np
 import pytest
 
 
@@ -68,3 +69,24 @@ def close():
 
  yield close
  close()
+
+
+@pytest.fixture()
+def reset_randomstate():
+ """
+ Fixture that set the global RandomState to the fixed seed 1
+
+ Notes
+ -----
+ Used by passing as an argument to the function that uses the global
+ RandomState
+
+ def test_some_plot(reset_randomstate):
+ <test code>
+
+ Returns the state after the test function exits
+ """
+ state = np.random.get_state()
+ np.random.seed(1)
+ yield
+ np.random.set_state(state)

statsmodels/discrete/tests/test_count_model.py

@@ -269,7 +269,7 @@ def test_t(self):
  t_test = self.res1.t_test(unit_matrix)
  assert_allclose(self.res1.tvalues, t_test.tvalue)
 
- def test_minimize(self):
+ def test_minimize(self, reset_randomstate):
  # check additional optimizers using the `minimize` option
  model = self.res1.model
  # use the same start_params, but avoid recomputing
@@ -380,7 +380,7 @@ def test_fit_regularized(self):
  atol=1e-1, rtol=1e-1)
 
  # possibly slow, adds 25 seconds
- def test_minimize(self):
+ def test_minimize(self, reset_randomstate):
  # check additional optimizers using the `minimize` option
  model = self.res1.model
  # use the same start_params, but avoid recomputing

statsmodels/discrete/tests/test_discrete.py

@@ -424,6 +424,7 @@ def setup_class(cls):
  res2.probit()
  cls.res2 = res2
  fit = Probit(data.endog, data.exog).fit
+ np.random.seed(1)
  cls.res1 = fit(method="basinhopping", disp=0, niter=5,
  minimizer={'method' : 'L-BFGS-B', 'tol' : 1e-8})
 

statsmodels/tsa/tests/test_holtwinters.py

@@ -327,7 +327,7 @@ def test_invalid_start_param_length():
  mod.fit(start_params=np.array([0.5]))
 
 
-def test_basin_hopping():
+def test_basin_hopping(reset_randomstate):
  mod = ExponentialSmoothing(housing_data, trend='add')
  res = mod.fit()
  res2 = mod.fit(use_basinhopping=True)