code: Added functions for HUE

code/lib/lmdk_bgt.py

@@ -398,6 +398,61 @@ def adaptive_cont(seq, lmdks, epsilon, inc_rt, dec_rt):
   return rls_data, bgts, skipped
 
 
+def adaptive_cons(seq, lmdks, epsilon, inc_rt, dec_rt):
+  '''
+    Adaptive budget allocation.
+
+    Parameters:
+      seq - The point sequence.
+      lmdks - The landmarks.
+      epsilon - The available privacy budget.
+      inc_rt - Sampling rate increase rate.
+      dec_rt - Sampling rate decrease rate.
+    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
+  samp_rt = 1
+  # Track landmarks
+  lmdk_cur = 0
+  # Track skipped releases
+  skipped = 0
+  for i, p in enumerate(seq):
+    # Check if current point is a landmark
+    is_landmark = any((lmdks[:]==p).all(1))
+    if is_landmark:
+      lmdk_cur += 1
+    if lmdk_lib.should_sample(samp_rt) or i == 0:
+      # Add noise to original data
+      o = lmdk_lib.add_laplace_noise(p[1], 1, bgts[i])
+      rls_data[i] = [p[0], o]
+      # Adjust sampling rate
+      if i > 0:
+        if abs(rls_data[i - 1][1] - o) < 1/bgts[i]:
+          # Decrease
+          samp_rt -= samp_rt*dec_rt
+        else:
+          # Increase
+          samp_rt += (1 - samp_rt)*inc_rt
+    else:
+      skipped += 1
+      # Skip current release and approximate with previous
+      rls_data[i] = rls_data[i - 1]
+      if is_landmark:
+        # 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 skip(seq, lmdks, epsilon):
   '''
     Skip landmarks.
@@ -459,6 +514,36 @@ def skip_cont(seq, lmdks, epsilon):
   return rls_data, bgts
 
 
+def skip_cons(seq, lmdks, epsilon):
+  '''
+    Skip landmarks.
+
+    Parameters:
+      seq - The point sequence.
+      lmdks - The landmarks.
+      epsilon - The available privacy budget.
+    Returns:
+      rls_data - The perturbed data.
+      bgts - The privacy budget allocation.
+  '''
+  # Event-level budget allocation
+  bgts = np.array(len(seq)*[epsilon])
+  # Released
+  rls_data = [None]*len(seq)
+  for i, p in enumerate(seq):
+    # Check if current point is a landmark
+    is_landmark = any((lmdks[:]==p).all(1))
+    # Add noise
+    o = [p[0], lmdk_lib.add_laplace_noise(p[1], 1, bgts[i])]
+    if is_landmark:
+      if i > 0:
+        # Approximate with previous
+        o = rls_data[i - 1]
+      bgts[i] = 0
+    rls_data[i] = o
+  return rls_data, bgts
+
+
 def sample(seq, lmdks, epsilon):
   '''
     Publish randomly.
@@ -642,6 +727,18 @@ def uniform_cont(seq, lmdks, epsilon):
   return rls_data, bgts
 
 
+def uniform_cons(seq, lmdks, epsilon):
+  # Released
+  rls_data = [None]*len(seq)
+  # Budgets
+  bgts = uniform(seq, lmdks, epsilon)
+  for i, p in enumerate(seq):
+    is_landmark = any((lmdks[:]==p).all(1))
+    # [timestamp, perturbed consumption]
+    rls_data[i] = [p[0], lmdk_lib.add_laplace_noise(p[1], 1, bgts[i])]
+  return rls_data, bgts
+
+
 def utility_analysis(seq, lmdks, o, epsilon):
   '''
     Analyze the utility.
@@ -688,3 +785,10 @@ def mae_cont(o):
     if p[0] != p[1]:
       mae += 1/len(o)
   return mae
+
+
+def mae_cons(seq, o):
+  mae = 0
+  for i, p in enumerate(seq):
+    mae += abs(p[1] - o[i][1])/len(seq)
+  return mae