code: Corrections and cleaning up

code/expt/copenhagen.py

@@ -27,7 +27,7 @@ def main(args):
   # The landmarks percentages
   lmdks_pct = [0, 20, 40, 60, 80, 100]
   # The privacy budget
-  epsilon = 1.0
+  epsilon = .1
 
   # Number of methods
   n = 3
@@ -68,32 +68,26 @@ def main(args):
       # Skip
       rls_data_s, bgts_s = lmdk_bgt.skip_cont(seq, lmdks, epsilon)
       # lmdk_bgt.validate_bgts(seq, lmdks, epsilon, bgts_s)
-      mae_s[i] += lmdk_bgt.mae_cont(rls_data_s)/args.iter
+      mae_s[i] += (lmdk_bgt.mae_cont(rls_data_s)/args.iter)*100
 
       # Uniform
       rls_data_u, bgts_u = lmdk_bgt.uniform_cont(seq, lmdks, epsilon)
       # lmdk_bgt.validate_bgts(seq, lmdks, epsilon, bgts_u)
-      mae_u[i] += lmdk_bgt.mae_cont(rls_data_u)/args.iter
+      mae_u[i] += (lmdk_bgt.mae_cont(rls_data_u)/args.iter)*100
 
       # Adaptive
       rls_data_a, _, _ = lmdk_bgt.adaptive_cont(seq, lmdks, epsilon, .5, .5)
-      mae_a[i] += lmdk_bgt.mae_cont(rls_data_a)/args.iter
+      mae_a[i] += (lmdk_bgt.mae_cont(rls_data_a)/args.iter)*100
 
       # Calculate once
       if i == 0:
         # Event
         rls_data_evt, _ = lmdk_bgt.uniform_cont(seq, lmdk_lib.find_lmdks_cont(lmdk_data, seq, uid, 0), epsilon)
-        mae_evt += lmdk_bgt.mae_cont(rls_data_evt)/args.iter
+        mae_evt += (lmdk_bgt.mae_cont(rls_data_evt)/args.iter)*100
         # User
         rls_data_usr, _ = lmdk_bgt.uniform_cont(seq, lmdk_lib.find_lmdks_cont(lmdk_data, seq, uid, 100), epsilon)
-        mae_usr += lmdk_bgt.mae_cont(rls_data_usr)/args.iter
+        mae_usr += (lmdk_bgt.mae_cont(rls_data_usr)/args.iter)*100
-
+  exit()
-  mae_u *= 100
-  mae_s *= 100
-  mae_a *= 100
-  mae_evt *= 100
-  mae_usr *= 100
-
   plt.axhline(
     y = mae_evt,
     color = '#212121',
@@ -172,7 +166,7 @@ if __name__ == '__main__':
     main(parse_args())
     end_time = time.time()
     print('##############################')
-    print('Time   : %.4fs' % (end_time - start_time))
+    print('Time elapsed: %s' % (time.strftime('%H:%M:%S', time.gmtime(end_time - start_time))))
     print('##############################')
   except KeyboardInterrupt:
     print('Interrupted by user.')

code/lib/lmdk_bgt.py

@@ -492,7 +492,8 @@ def skip_cont(seq, lmdks, epsilon):
       lmdks - The landmarks.
       epsilon - The available privacy budget.
     Returns:
-      rls_data - The perturbed data.
+      rls_data - The new data.
+                 [0: The true answer, 1: The perturbed answer]
       bgts - The privacy budget allocation.
   '''
   # Event-level budget allocation
@@ -501,15 +502,15 @@ def skip_cont(seq, lmdks, epsilon):
   rls_data = [None]*len(seq)
   for i, p in enumerate(seq):
     # Check if current point is a landmark
-    r = any((lmdks[:]==p).all(1))
+    is_landmark = any(np.equal(lmdks, p).all(1))
     # Add noise
-    o = lmdk_lib.randomized_response(r, bgts[i])
+    o = lmdk_lib.randomized_response(is_landmark, bgts[i])
-    if r:
+    if is_landmark:
       if i > 0:
         # Approximate with previous
         o = rls_data[i - 1][1]
-      bgts[i] = 0
+        bgts[i] = 0
-    rls_data[i] = [r, o]
+    rls_data[i] = [is_landmark, o]
   return rls_data, bgts
 
 
@@ -720,7 +721,7 @@ def uniform_cont(seq, lmdks, epsilon):
   # Budgets
   bgts = uniform(seq, lmdks, epsilon)
   for i, p in enumerate(seq):
-    r = any((lmdks[:]==p).all(1))
+    r = any(np.equal(lmdks, p).all(1))
     # [original, perturbed]
     rls_data[i] = [r, lmdk_lib.randomized_response(r, bgts[i])]
   return rls_data, bgts

code/lib/lmdk_lib.py

@@ -959,13 +959,13 @@ def find_lmdks_cont(lmdk_data, seq, uid, pct):
                   landmarks percentage.
                   0: uid, 1: lmdk_pct, 2: contacts
       seq       - The users' data.
-                  0: tim, 1: uid, 2: cont, 3: rssi 
+                  0: tim, 1: uid, 2: cont, 3: rssi
       uid       - The user's id that we are interested in.
       pct       - The landmarks percentage.
     Returns:
       lmdks - The user's landmarks contacts for the given 
               landmarks percentage.
-              0: uid_b
+              0: tim, 1: uid, 2: cont, 3: rssi
   '''
   # Initialize user's landmarks
   lmdks = np.empty(0).reshape(0,4)

code/lib/lmdk_sel.py

@@ -5,8 +5,8 @@ from lmdk_lib import *
 import exp_mech
 import numpy as np
 import random
+import scipy.stats as stats
 import time
-from scipy.spatial.distance import cdist
 
 
 '''
@@ -163,7 +163,7 @@ def get_opts_from_top_h(seq, lmdks):
   # The options to be returned
   hist_opts = []
   # Keep adding points until the maximum is reached
-  while np.sum(hist_cur) < max(seq):
+  while np.sum(hist_cur) < len(seq):
     # Track the minimum (best) evaluation
     diff_min = float('inf')
     # The candidate option
@@ -175,8 +175,8 @@ def get_opts_from_top_h(seq, lmdks):
         hist_tmp = np.copy(hist_cur)
         hist_tmp[i] += 1
         # Find difference from original
-        # diff_cur = get_norm(hist, hist_tmp)  # Euclidean
+        diff_cur = get_norm(hist, hist_tmp)  # Euclidean
-        diff_cur = get_emd(hist, hist_tmp)  # Wasserstein
+        # diff_cur = get_emd(hist, hist_tmp)  # Wasserstein
         # Remember if it is the best that you've seen
         if diff_cur < diff_min:
           diff_min = diff_cur
@@ -326,19 +326,37 @@ def find_lmdks(seq, lmdks, epsilon):
     opts = get_opts_from_top_h(get_seq(1, len(seq)), lmdks_seq)
     # Landmarks selection budget
     eps_sel = epsilon/(len(lmdks_seq) + 1)
-    # Get private landmarks timestamps
+    # Get landmarks histogram with dummy landmarks
-    lmdks_seq, _ = exp_mech.exponential(hist, opts, exp_mech.score, 1.0, eps_sel)
+    hist_new, _ = exp_mech.exponential(hist, opts, exp_mech.score, 1.0, eps_sel)
+    # Split sequence in parts of size h 
+    pt_idx = []
+    for idx in range(1, len(seq), h):
+      pt_idx.append([idx, idx + h - 1])
+    pt_idx[-1][1] = len(seq)
+    # Get new landmarks indexes
+    lmdks_seq_new = np.array([], dtype=int)
+    for i, pt in enumerate(pt_idx):
+      # Already landmarks
+      lmdks_seq_pt = lmdks_seq[(lmdks_seq >= pt[0]) & (lmdks_seq <= pt[1])]
+      # Sample randomly from the rest of the sequence
+      size = hist_new[i] - len(lmdks_seq_pt)
+      rglr = np.setdiff1d(np.arange(pt[0], pt[1] + 1), lmdks_seq_pt)
+      # Add already landmarks
+      lmdks_seq_new = np.concatenate([lmdks_seq_new, lmdks_seq_pt])
+      # Add new landmarks
+      if size > 0 and len(rglr) > size:
+        lmdks_seq_new = np.concatenate([lmdks_seq_new,
+          np.random.choice(
+            rglr, 
+            size = size, 
+            replace = False
+          )
+        ])
     # Get actual landmarks values
-    lmdks_new = seq[lmdks_seq]
+    lmdks_new = seq[lmdks_seq_new - 1]
   return lmdks_new, epsilon - eps_sel
 
 def test():
-
-  A = np.array([1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1])
-  B = np.array([0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0])
-  print(get_norm(A, B))
-  exit()
-
   # Start and end points of the sequence
   # # Nonrandom
   # start = 1