lmdk_bgt.incremental: Initial commit

code/lib/lmdk_bgt.py

@@ -467,6 +467,67 @@ def discount(seq, lmdks, epsilon):
   return rls_data, bgts, skipped
 
 
+def incremental(seq, lmdks, epsilon, diff_rt):
+  '''
+    Incremental sampling.
+
+    Parameters:
+      seq - The point sequence.
+      lmdks - The landmarks.
+      epsilon - The available privacy budget.
+      diff_rt - Sampling rate difference.
+    Returns:
+      rls_data - The perturbed data.
+      bgts - The privacy budget allocation.
+      skipped - The number of skipped releases.
+  '''
+  # Uniform budget allocation
+  bgts = uniform(seq, lmdks, epsilon)
+  # Released
+  rls_data = [None]*len(seq)
+  # The sampling rate
+  # (publish with 50% chance)
+  samp_rt = .5
+  # Track landmarks
+  lmdk_cur = 0
+  # Track consecutive landmarks
+  lmdk_con = 0
+  # Track skipped releases
+  skipped = 0
+  for i, p in enumerate(seq):
+    # Check if current point is a landmark
+    if lmdk_lib.is_landmark(p, lmdks):
+      lmdk_cur += 1
+      # Increase sampling rate
+      samp_rt += (1 - samp_rt)*diff_rt
+    else:
+      # Decrease sampling rate
+      samp_rt -= samp_rt*diff_rt
+    # Get coordinates
+    loc = (p[1], p[2])
+    if i == 0 or lmdk_lib.should_sample(samp_rt):
+      # Add noise to original data
+      new_loc = lmdk_lib.add_polar_noise(loc, bgts[i])
+      rls_data[i] = [p[0], new_loc[0], new_loc[1], p[3]]
+      # Adjust sampling rate
+      if lmdk_lib.is_landmark(p, lmdks):
+        # Adjust consecutive landmarks tracking
+        lmdk_con += 1
+    else:
+      skipped += 1
+      # Skip current release and approximate with previous
+      rls_data[i] = rls_data[i - 1]
+      if lmdk_lib.is_landmark(p, lmdks):
+        # Reset consecutive landmarks tracking
+        lmdk_con = 0
+        # Allocate the current budget to the following releases uniformly
+        for j in range(i + 1, len(seq)):
+          bgts[j] += bgts[i]/(len(lmdks) - lmdk_cur + 1)
+      # No budget was spent
+      bgts[i] = 0
+  return rls_data, bgts, skipped
+
+
 def uniform_r(seq, lmdks, epsilon):
   # Released
   rls_data = [None]*len(seq)