GitList
Browse code

make loss functions accessible from learner namespace

The default loss functions will now become public API (in case
anyone wants to use them in their own loss functions) and the
non-standard loss functions will be accessed as
`adaptive.learner.learner2D.uniform` etc.

Joseph Weston authored on 19/02/2018 18:37:32
Showing 3 changed files
adaptive/learner/__init__.py
History View file @ 4965811
... ... 
@@ -2,7 +2,7 @@
2 2 
 from .average_learner import AverageLearner
3 3 
 from .base_learner import BaseLearner
4 4 
 from .balancing_learner import BalancingLearner
5 
-from .learner1D import Learner1D, uniform_sampling_1d
6 
-from .learner2D import Learner2D, uniform_sampling_2d
5 
+from .learner1D import Learner1D
6 
+from .learner2D import Learner2D
7 7 
 from .integrator_learner import IntegratorLearner
8 8 
 from .data_saver import DataSaver
adaptive/learner/learner1D.py
History View file @ 4965811
... ... 
@@ -11,7 +11,7 @@ import scipy.interpolate
11 11 
 from ..notebook_integration import ensure_holoviews
12 12 
 from .base_learner import BaseLearner
13 13
14 
-def uniform_sampling_1d(interval, scale, function_values):
14 
+def uniform_loss(interval, scale, function_values):
15 15 
     """Loss function that samples the domain uniformly.
16 16
17 17 
     Works with `~adaptive.Learner1D` only.
... ... 
@@ -32,7 +32,7 @@ def uniform_sampling_1d(interval, scale, function_values):
32 32 
     return dx
33 33
34 34
35 
-def _default_loss_per_interval(interval, scale, function_values):
35 
+def default_loss(interval, scale, function_values):
36 36 
     """Calculate loss on a single interval
37 37
38 38 
     Currently returns the rescaled length of the interval. If one of the
... ... 
@@ -87,7 +87,7 @@ class Learner1D(BaseLearner):
87 87
88 88 
     def __init__(self, function, bounds, loss_per_interval=None):
89 89 
         self.function = function
90 
-        self.loss_per_interval = loss_per_interval or _default_loss_per_interval
90 
+        self.loss_per_interval = loss_per_interval or default_loss
91 91
92 92 
         # A dict storing the loss function for each interval x_n.
93 93 
         self.losses = {}
adaptive/learner/learner2D.py
History View file @ 4965811
... ... 
@@ -42,7 +42,7 @@ def areas(ip):
42 42 
     return areas
43 43
44 44
45 
-def uniform_sampling_2d(ip):
45 
+def uniform_loss(ip):
46 46 
     """Loss function that samples the domain uniformly.
47 47
48 48 
     Works with `~adaptive.Learner2D` only.
... ... 
@@ -86,7 +86,7 @@ def resolution_loss(ip, min_distance=0, max_distance=1):
86 86 
     A = areas(ip)
87 87 
     dev = np.sum(deviations(ip), axis=0)
88 88
89 
-    # similar to the _default_loss_per_triangle
89 
+    # similar to the default_loss
90 90 
     loss = np.sqrt(A) * dev + A
91 91
92 92 
     # Setting areas with a small area to zero such that they won't be chosen again
... ... 
@@ -99,7 +99,7 @@ def resolution_loss(ip, min_distance=0, max_distance=1):
99 99 
     return loss
100 100
101 101
102 
-def _default_loss_per_triangle(ip):
102 
+def default_loss(ip):
103 103 
     dev = np.sum(deviations(ip), axis=0)
104 104 
     A = areas(ip)
105 105 
     losses = dev * np.sqrt(A) + 0.1 * A
... ... 
@@ -207,7 +207,7 @@ class Learner2D(BaseLearner):
207 207 
     def __init__(self, function, bounds, loss_per_triangle=None):
208 208 
         self.ndim = len(bounds)
209 209 
         self._vdim = None
210 
-        self.loss_per_triangle = loss_per_triangle or _default_loss_per_triangle
210 
+        self.loss_per_triangle = loss_per_triangle or default_loss
211 211 
         self.bounds = tuple((float(a), float(b)) for a, b in bounds)
212 212 
         self.data = OrderedDict()
213 213 
         self._stack = OrderedDict()