Add further todos

diffprivlib/mechanisms/snapping.py

@@ -56,7 +56,8 @@ def __init__(self, *, epsilon, sensitivity, lower, upper):
         super().__init__(epsilon=epsilon, delta=0.0, sensitivity=sensitivity)
         TruncationAndFoldingMixin.__init__(self, lower=lower, upper=upper)
         # scale and center bounds around 0
-        self.bound = (self.upper - self.lower) / 2.0 / self.sensitivity
+        # todo: bound should be updated if upper, lower, or sensitivity changed
+        self._bound = (self.upper - self.lower) / 2.0 / self.sensitivity
 
     def _check_all(self, value):
         super()._check_all(value)
@@ -68,10 +69,10 @@ def _check_all(self, value):
         return True
 
     def _truncate(self, value):
-        if value > self.bound:
-            return self.bound
-        if value < -self.bound:
-            return -self.bound
+        if value > self._bound:
+            return self._bound
+        if value < -self._bound:
+            return -self._bound
 
         return value
 
@@ -89,7 +90,7 @@ def effective_epsilon(self):
         # as per section 5.2 of Mironov
         # 1/λ + 12Bη/λ + 2η
         machine_epsilon = np.finfo(float).epsneg
-        return self.epsilon + 12.0 * self.bound * self.epsilon + 2.0 * machine_epsilon
+        return self.epsilon + 12.0 * self._bound * self.epsilon + 2.0 * machine_epsilon
 
     def _scale_and_offset_value(self, value):
         """
@@ -98,10 +99,10 @@ def _scale_and_offset_value(self, value):
         :return: value offset to be centered on 0 and scaled to sensitivity 1
         """
         value_scaled = value / self.sensitivity
-        return value_scaled - self.bound - (self.lower / self.sensitivity)
+        return value_scaled - self._bound - (self.lower / self.sensitivity)
 
     def _reverse_scale_and_offset_value(self, value):
-        return (value + self.bound) * self.sensitivity + self.lower
+        return (value + self._bound) * self.sensitivity + self.lower
 
     def _get_nearest_power_of_2(self, x):
         # todo: extract from class?
@@ -124,6 +125,7 @@ def _round_to_nearest_power_of_2(self, value):
         return value - remainder
 
     def _sample_laplace(self):
+        # todo: only produces integer multiples of 2^-53; replace with _sample_uniform
         unif_rv = self._rng.random() - 0.5
         # Using crlibm, as suggested in Mironov paper
         # no need for sensitivity already scaled down to sensitivity 1