LearnerND scale output values before computing loss

Closes #78 (closed)

In a way similar to the Learner1D

Recompute all losses when the scale increases with a factor 1.1.

My simple measurements indicate that with the ring_of_fire as test function the learner slows down approximately 10%. But we get in return that we scale everything to equivalent sizes before computing the loss.

TODO:

• Add some test(s).

adaptive/learner/learnerND.py

@@ -9,13 +9,13 @@ import numpy as np
 from scipy import interpolate
 import scipy.spatial
 
-from .base_learner import BaseLearner
+from adaptive.learner.base_learner import BaseLearner
 
-from ..notebook_integration import ensure_holoviews, ensure_plotly
-from .triangulation import (Triangulation, point_in_simplex,
-                            circumsphere, simplex_volume_in_embedding,
-                            fast_det)
-from ..utils import restore, cache_latest
+from adaptive.notebook_integration import ensure_holoviews, ensure_plotly
+from adaptive.learner.triangulation import (
+    Triangulation, point_in_simplex, circumsphere,
+    simplex_volume_in_embedding, fast_det)
+from adaptive.utils import restore, cache_latest
 
 
 def volume(simplex, ys=None):
@@ -177,8 +177,14 @@ class LearnerND(BaseLearner):
         # triangulation of the pending points inside a specific simplex
         self._subtriangulations = dict()  # simplex → triangulation
 
-        # scale to unit
+        # scale to unit hypercube
+        # for the input
         self._transform = np.linalg.inv(np.diag(np.diff(self._bbox).flat))
+        # for the output
+        self._min_value = None
+        self._max_value = None
+        self._output_multiplier = 1 # If we do not know anything, do not scale the values
+        self._recompute_losses_factor = 1.1
 
         # create a private random number generator with fixed seed
         self._random = random.Random(1)
@@ -277,6 +283,7 @@ class LearnerND(BaseLearner):
             to_delete, to_add = tri.add_point(
                 point, simplex, transform=self._transform)
             self.update_losses(to_delete, to_add)
+            self._update_range(value)
 
     def _simplex_exists(self, simplex):
         simplex = tuple(sorted(simplex))
@@ -447,10 +454,8 @@ class LearnerND(BaseLearner):
                                      if p not in self.data)
 
         for simplex in to_add:
-            vertices = self.tri.get_vertices(simplex)
-            values = [self.data[tuple(v)] for v in vertices]
-            loss = float(self.loss_per_simplex(vertices, values))
-            self._losses[simplex] = float(loss)
+            loss = self.compute_loss(simplex)
+            self._losses[simplex] = loss
 
             for p in pending_points_unbound:
                 self._try_adding_pending_point_to_simplex(p, simplex)
@@ -462,6 +467,83 @@ class LearnerND(BaseLearner):
             self._update_subsimplex_losses(
                 simplex, self._subtriangulations[simplex].simplices)
 
+    def compute_loss(self, simplex):
+        # get the loss
+        vertices = self.tri.get_vertices(simplex)
+        values = [self.data[tuple(v)] for v in vertices]
+
+        # scale them to a cube with sides 1
+        vertices = vertices @ self._transform
+        values = self._output_multiplier * values
+
+        # compute the loss on the scaled simplex
+        return float(self.loss_per_simplex(vertices, values))
+
+    def recompute_all_losses(self):
+        """Recompute all losses and pending losses."""
+        # amortized O(N) complexity
+        if self.tri is None:
+            return
+
+        # reset the _simplex_queue
+        self._simplex_queue = []
+
+        # recompute all losses
+        for simplex in self.tri.simplices:
+            loss = self.compute_loss(simplex)
+            self._losses[simplex] = loss
+
+            # now distribute it around the the children if they are present
+            if simplex not in self._subtriangulations:
+                heapq.heappush(self._simplex_queue, (-loss, simplex, None))
+                continue
+
+            self._update_subsimplex_losses(
+                simplex, self._subtriangulations[simplex].simplices)
+
+    @property
+    def _scale(self):
+        # get the output scale
+        return self._max_value - self._min_value
+
+    def _update_range(self, new_output):
+        if self._min_value is None or self._max_value is None:
+            # this is the first point, nothing to do, just set the range
+            self._min_value = np.array(new_output)
+            self._max_value = np.array(new_output)
+            self._old_scale = self._scale
+            return False
+
+        # if range in one or more directions is doubled, then update all losses
+        self._min_value = np.minimum(self._min_value, new_output)
+        self._max_value = np.maximum(self._max_value, new_output)
+
+        scale_multiplier = 1 / self._scale
+        if isinstance(scale_multiplier, float):
+            scale_multiplier = np.array([scale_multiplier], dtype=float)
+
+        # the maximum absolute value that is in the range. Because this is the
+        # largest number, this also has the largest absolute numerical error.
+        max_absolute_value_in_range = np.max(np.abs([self._min_value, self._max_value]), axis=0)
+        # since a float has a relative error of 1e-15, the absolute error is the value * 1e-15
+        abs_err = 1e-15 * max_absolute_value_in_range
+        # when scaling the floats, the error gets increased.
+        scaled_err = abs_err * scale_multiplier
+
+        allowed_numerical_error = 1e-2
+
+        # do not scale along the axis if the numerical error gets too big
+        scale_multiplier[scaled_err > allowed_numerical_error] = 1
+
+        self._output_multiplier = scale_multiplier
+
+        scale_factor = np.max(np.nan_to_num(self._scale / self._old_scale))
+        if scale_factor > self._recompute_losses_factor:
+            self._old_scale = self._scale
+            self.recompute_all_losses()
+            return True
+        return False
+
     def losses(self):
         """Get the losses of each simplex in the current triangulation, as dict