Merge branch 'bugfix/first-iteration'

correct point choosing in the first few steps of the 1DLearner

Closes #55

See merge request qt/adaptive!59

Author: jbweston

adaptive/learner/learner1D.py

@@ -278,13 +278,14 @@ def choose_points(self, n, add_data=True):
             if bound not in self.data and bound not in self.data_interp:
                 points.append(bound)
 
-        # Ensure we return exactly 'n' points.
-        if points:
-            loss_improvements = [float('inf')] * n
-            if n <= 2:
-                points = points[:n]
-            else:
-                points = np.linspace(*self.bounds, n)
+        if len(points) == 2:
+            # First time
+            loss_improvements = [np.inf] * n
+            points = np.linspace(*self.bounds, n)
+        elif len(points) == 1:
+            # Second time, if we previously returned just self.bounds[0]
+            loss_improvements = [np.inf] * n
+            points = np.linspace(*self.bounds, n + 1)[1:]
         else:
             def xs(x, n):
                 if n == 1:

adaptive/tests/test_learner.py

@@ -321,6 +321,30 @@ def test_learner_performance_is_invariant_under_scaling(learner_type, f, learner
     assert abs(learner.loss() - control.loss()) / learner.loss() < 1e-11
 
 
+def test_learner1d_first_iteration():
+    """Edge cases where we ask for a few points at the start."""
+    learner = Learner1D(lambda x: None, (-1, 1))
+    points, loss_improvements = learner.choose_points(2)
+    assert set(points) == set([-1, 1])
+
+    learner = Learner1D(lambda x: None, (-1, 1))
+    points, loss_improvements = learner.choose_points(3)
+    assert set(points) == set([-1, 0, 1])
+
+    learner = Learner1D(lambda x: None, (-1, 1))
+    points, loss_improvements = learner.choose_points(1)
+    assert len(points) == 1 and points[0] in [-1, 1]
+    rest = set([-1, 0, 1]) - set(points)
+    points, loss_improvements = learner.choose_points(2)
+    assert set(points) == set(rest)
+
+    learner = Learner1D(lambda x: None, (-1, 1))
+    points, loss_improvements = learner.choose_points(1)
+    to_see = set([-1, 1]) - set(points)
+    points, loss_improvements = learner.choose_points(1)
+    assert set(points) == set(to_see)
+
+
 @pytest.mark.xfail
 @run_with(Learner1D, Learner2D)
 def test_convergence_for_arbitrary_ordering(learner_type, f, learner_kwargs):
@@ -339,4 +363,4 @@ def test_learner_subdomain(learner_type, f, learner_kwargs):
     """Learners that never receive data outside of a subdomain should
        perform 'similarly' to learners defined on that subdomain only."""
     # XXX: not sure how to implement this. How do we measure "performance"?
-    raise NotImplementedError()
+    raise NotImplementedError()