gradboard 0.1.5__tar.gz → 0.1.7__tar.gz

{gradboard-0.1.5 → gradboard-0.1.7}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: gradboard
-Version: 0.1.5
+Version: 0.1.7
 Summary: Easily snowboard down gnarly loss gradients
 License: MIT
 Author: Nicholas Bailey

{gradboard-0.1.5 → gradboard-0.1.7}/gradboard/scheduler.py

@@ -28,12 +28,9 @@ class PASS:
         max_lr: float = None,
         cool_point: float = None,
     ):
-        assert (max_lr is not None) == (cool_point is not None)
-        assert (
-            ((max_lr is not None) and (cool_point is not None))
-            != range_test
-            is not None
-        )
+        if not range_test:
+            assert max_lr is not None
+            assert cool_point is not None
 
         self.model = model
         self.optimiser = optimiser
@@ -142,7 +139,7 @@ class PASS:
         learning_rates = [t[0] for t in range_test_results]
         losses = [t[1] for t in self.range_test_results]
         losses = losses[:-1] + [10 * max(losses)]
-        smoothed_losses = gaussian_filter1d([t[1] for t in range_test_results][:-1], 3)
+        smoothed_losses = gaussian_filter1d([t[1] for t in losses], 3)
         return list(zip(learning_rates, smoothed_losses, strict=True))
 
     def _plot_range_test(self, range_test_results):
@@ -161,8 +158,6 @@ class PASS:
         ...
         """
         range_test_results = self._smoothed_range_test(self.range_test_results)
-        self._plot_range_test(range_test_results)
-
         minimum = min(range_test_results, key=lambda x: x[1])
         points_left_of_min = [p for p in range_test_results if p[0] < minimum[0]]
         highest_point_left_of_min = max(points_left_of_min, key=lambda x: x[1])

{gradboard-0.1.5 → gradboard-0.1.7}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "gradboard"
-version = "0.1.5"
+version = "0.1.7"
 description = "Easily snowboard down gnarly loss gradients"
 authors = [
     {name = "Nicholas Bailey"}