Merge pull request #384 from jcapriot/random_warnings

Warn for non-repeatable random tests in a testing environment

Author: jcapriot

discretize/tests.py

@@ -25,6 +25,8 @@
   assert_isadjoint
 """  # NOQA D205
 
+import warnings
+
 import numpy as np
 import scipy.sparse as sp
 
@@ -81,6 +83,23 @@
 _happiness_rng = np.random.default_rng()
 
 
+def _warn_random_test():
+    stack = inspect.stack()
+    in_pytest = any(x[0].f_globals["__name__"].startswith("_pytest.") for x in stack)
+    in_nosetest = any(x[0].f_globals["__name__"].startswith("nose.") for x in stack)
+
+    if in_pytest or in_nosetest:
+        test = "pytest" if in_pytest else "nosetest"
+        warnings.warn(
+            f"You are running a {test} without setting a random seed, the results might not be"
+            "repeatable. For repeatable tests please pass an argument to `random seed` that is"
+            "not `None`.",
+            UserWarning,
+            stacklevel=3,
+        )
+    return in_pytest or in_nosetest
+
+
 def setup_mesh(mesh_type, nC, nDim, random_seed=None):
     """Generate arbitrary mesh for testing.
 
@@ -110,6 +129,8 @@ def setup_mesh(mesh_type, nC, nDim, random_seed=None):
         A discretize mesh of class specified by the input argument *mesh_type*
     """
     if "random" in mesh_type:
+        if random_seed is None:
+            _warn_random_test()
         rng = np.random.default_rng(random_seed)
     if "TensorMesh" in mesh_type:
         if "uniform" in mesh_type:
@@ -649,6 +670,8 @@ def check_derivative(
     x0 = mkvc(x0)
 
     if dx is None:
+        if random_seed is None:
+            _warn_random_test()
         rng = np.random.default_rng(random_seed)
         dx = rng.standard_normal(len(x0))
 
@@ -867,6 +890,8 @@ def assert_isadjoint(
     """
     __tracebackhide__ = True
 
+    if random_seed is None:
+        _warn_random_test()
     rng = np.random.default_rng(random_seed)
 
     def random(size, iscomplex):

tests/base/test_tests.py

@@ -4,7 +4,13 @@
 import subprocess
 import numpy as np
 import scipy.sparse as sp
-from discretize.tests import assert_isadjoint, check_derivative, assert_expected_order
+from discretize.tests import (
+    assert_isadjoint,
+    check_derivative,
+    assert_expected_order,
+    _warn_random_test,
+    setup_mesh,
+)
 
 
 class TestAssertIsAdjoint:
@@ -166,3 +172,22 @@ def test_import_time():
 
     # Currently we check t < 1.0s.
     assert float(out.stderr.decode("utf-8")[:-1]) < 1.0
+
+
+def test_random_test_warning():
+
+    match = r"You are running a pytest without setting a random seed.*"
+    with pytest.warns(UserWarning, match=match):
+        _warn_random_test()
+
+    def simple_deriv(x):
+        return np.sin(x), lambda y: np.cos(x) * y
+
+    with pytest.warns(UserWarning, match=match):
+        check_derivative(simple_deriv, np.zeros(10), plotIt=False)
+
+    with pytest.warns(UserWarning, match=match):
+        setup_mesh("randomTensorMesh", 10, 1)
+
+    with pytest.warns(UserWarning, match=match):
+        assert_isadjoint(lambda x: x, lambda x: x, 5, 5)