Fix DRangedTree bounding on left-hand side of values

methods/utils/dranged_tree.py

@@ -140,7 +140,7 @@ def contains(self, point):
             return self.left.contains(point)
     def dump(self, space):
         print(space, f"split axis {self.axis} at {self.value}")
-        print(space + "  <")
+        print(space + "  <=")
         self.left.dump(space + "\t")
         print(space + "  >")
         self.right.dump(space + "\t")
@@ -149,6 +149,29 @@ def depth(self):
     def size(self):
         return 1 + self.left.size() + self.right.size()
 
+class SplitDLeanRightTree(DRangedTree):
+    """This is identical to a SplitDTree, except values equal to the value go into the right tree instead of the left"""
+    def __init__(self, left: DRangedTree, right: DRangedTree, axis: int, value: float):
+        self.left = left
+        self.right = right
+        self.axis = axis
+        self.value = value
+    def contains(self, point):
+        if point[self.axis] >= self.value:
+            return self.right.contains(point)
+        else:
+            return self.left.contains(point)
+    def dump(self, space):
+        print(space, f"split axis {self.axis} at {self.value}")
+        print(space + "  <")
+        self.left.dump(space + "\t")
+        print(space + "  >=")
+        self.right.dump(space + "\t")
+    def depth(self):
+        return 1 + max(self.left.depth(), self.right.depth())
+    def size(self):
+        return 1 + self.left.size() + self.right.size()
+
 class TreeState:
     def __init__(self, dimensions_or_tree, j: int = -1, drop: bool = False):
         if isinstance(dimensions_or_tree, TreeState):
@@ -234,7 +257,8 @@ def _make_tree_internal(rects, bounds, state: TreeState):
                 new_bounds = np.copy(bounds)
                 new_bounds[0, d] = left
                 subtree = _make_tree_internal(rects, new_bounds, state.descend(d))
-                return SplitDTree(EmptyTree(), subtree, d, left)
+                # We need to lean right, because we only want to exclude values strictly less than left.
+                return SplitDLeanRightTree(EmptyTree(), subtree, d, left)
             if bounds[1, d] == math.inf:
                 right = np.max(rects[:, 1, d])
                 new_bounds = np.copy(bounds)