@@ -129,7 +129,7 @@ def curvature_loss_function(area_factor=1, euclid_factor=0.02, horizontal_factor

     @uses_nth_neighbors(1)

     def curvature_loss(xs, ys):

         xs_middle = xs[1:3]

-        ys_middle = xs[1:3]

+        ys_middle = ys[1:3]

         triangle_loss_ = triangle_loss(xs, ys)

         default_loss_ = default_loss(xs_middle, ys_middle)

@@ -654,7 +654,8 @@ class Learner1D(BaseLearner):

         return self.data

     def _set_data(self, data):

-        self.tell_many(*zip(*data.items()))

+        if data:

+            self.tell_many(*zip(*data.items()))

 def loss_manager(x_scale):

@@ -670,7 +671,7 @@ def finite_loss(ival, loss, x_scale):

     sort intervals that have infinite loss."""

     # If the loss is infinite we return the

     # distance between the two points.

-    if math.isinf(loss):

+    if math.isinf(loss) or math.isnan(loss):

         loss = (ival[1] - ival[0]) / x_scale

         if len(ival) == 3:

             # Used when constructing quals. Last item is

@@ -757,7 +757,8 @@ class LearnerND(BaseLearner):

         return self.data

     def _set_data(self, data):

-        self.tell_many(*zip(*data.items()))

+        if data:

+            self.tell_many(*zip(*data.items()))

     def _get_iso(self, level=0.0, which='surface'):

         if which == 'surface':

@@ -363,3 +363,15 @@ def test_curvature_loss_vectors():

     learner = Learner1D(f, (-1, 1), loss_per_interval=loss)

     simple(learner, goal=lambda l: l.npoints > 100)

     assert learner.npoints > 100

+

+

+def test_NaN_loss():

+    # see https://github.com/python-adaptive/adaptive/issues/145

+    def f(x):

+        a = 0.01

+        if random.random() < 0.2:

+            return np.NaN

+        return x + a**2 / (a**2 + x**2)

+

+    learner = Learner1D(f, bounds=(-1, 1))

+    simple(learner, lambda l: l.npoints > 100)