adaptive/tests/test_average_learner.py

 # -*- coding: utf-8 -*-
 
+import random
+
+import numpy as np
+
 from ..learner import AverageLearner
 
 
@@ -17,3 +21,28 @@ def test_only_returns_new_points():
     assert learner.ask(1)[0][0] == 3
     assert learner.ask(1)[0][0] == 4
     assert learner.ask(1)[0][0] == 10
+
+
+def test_avg_std_and_npoints():
+    learner = AverageLearner(lambda x: x, atol=None, rtol=0.01)
+
+    for i in range(300):
+        # This will add 5000 points at random values of n.
+        # It could try to readd already evaluated points.
+
+        n = random.randint(0, 2*300)
+        value = random.random()
+
+        # With 10% chance None is added to simulate asking that point.
+        if value < 0.9:
+            learner.tell(n, value)
+        else:
+            learner.tell_pending(n)
+
+        if i > 2 and i % 10 == 0:
+            # We need more than two points for 'learner.std' to be defined.
+            values = np.array(list(learner.data.values()))
+            std = np.sqrt(sum((values - values.mean())**2) / (len(values) - 1))
+            assert learner.npoints == len(learner.data)
+            assert abs(learner.sum_f - values.sum()) < 1e-13
+            assert abs(learner.std - std) < 1e-13

adaptive/tests/test_learners.py

@@ -288,7 +288,9 @@ def test_point_adding_order_is_irrelevant(learner_type, f, learner_kwargs):
     np.testing.assert_almost_equal(sorted(pls), sorted(cpls))
 
 
-@run_with(Learner1D, Learner2D, LearnerND, AverageLearner)
+# XXX: the Learner2D fails with ~50% chance
+# see https://gitlab.kwant-project.org/qt/adaptive/issues/84
+@run_with(Learner1D, xfail(Learner2D), LearnerND, AverageLearner)
 def test_expected_loss_improvement_is_less_than_total_loss(learner_type, f, learner_kwargs):
     """The estimated loss improvement can never be greater than the total loss."""
     f = generate_random_parametrization(f)