code: Several fixes

code/expt/copenhagen-sel.py

@@ -71,13 +71,11 @@ def main(args):
       lmdks_sel, eps_out = lmdk_sel.find_lmdks(seq, lmdks, epsilon)
 
       # Skip
-      rls_data_s, bgts_s = lmdk_bgt.skip_cont(seq, lmdks_sel, eps_out)
+      rls_data_s, _ = lmdk_bgt.skip_cont(seq, lmdks_sel, eps_out)
-      # lmdk_bgt.validate_bgts(seq, lmdks, epsilon, bgts_s)
       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_sel, eps_out)
+      rls_data_u, _ = lmdk_bgt.uniform_cont(seq, lmdks_sel, eps_out)
-      # lmdk_bgt.validate_bgts(seq, lmdks, epsilon, bgts_u)
       mae_u[i] += (lmdk_bgt.mae_cont(rls_data_u)/args.iter)*100
 
       # Adaptive

code/expt/hue-sel.py

@@ -64,19 +64,18 @@ def main(args):
 
     for _ in range(args.iter):
 
-      lmdks, eps_out = lmdk_sel.find_lmdks(seq, lmdks, epsilon)
+      lmdks_sel, eps_out = lmdk_sel.find_lmdks(seq, lmdks, epsilon)
 
       # Skip
-      rls_data_s, bgts_s = lmdk_bgt.skip_cons(seq, lmdks, eps_out)
+      rls_data_s, _ = lmdk_bgt.skip_cons(seq, lmdks_sel, eps_out)
-      # lmdk_bgt.validate_bgts(seq, lmdks, epsilon, bgts_s)
       mae_s[i] += lmdk_bgt.mae_cons(seq, rls_data_s)/args.iter
 
       # Uniform
-      rls_data_u, bgts_u = lmdk_bgt.uniform_cons(seq, lmdks, eps_out)
+      rls_data_u, _ = lmdk_bgt.uniform_cons(seq, lmdks_sel, eps_out)
       mae_u[i] += lmdk_bgt.mae_cons(seq, rls_data_u)/args.iter
 
       # Adaptive
-      rls_data_a, _, _ = lmdk_bgt.adaptive_cons(seq, lmdks, eps_out, .5, .5)
+      rls_data_a, _, _ = lmdk_bgt.adaptive_cons(seq, lmdks_sel, eps_out, .5, .5)
       mae_a[i] += lmdk_bgt.mae_cons(seq, rls_data_a)/args.iter
 
       # Calculate once

code/expt/lmdk_sel_cmp.py

@@ -57,8 +57,6 @@ def main(args):
     print('(%d/%d) %s... ' %(d_i + 1, len(dist_type), title), end='', flush=True)
     mae = np.zeros(len(lmdk_n))
     for n_i, n in enumerate(lmdk_n):
-      if n == lmdk_n[-1]:
-        break
       for r in range(args.iter):
         lmdks = lmdk_lib.get_lmdks(seq, n, d)
         hist, h = lmdk_lib.get_hist(seq, lmdks)
@@ -67,20 +65,11 @@ def main(args):
         res, _ = exp_mech.exponential(hist, opts, exp_mech.score, delta, epsilon)
         mae[n_i] += lmdk_lib.get_norm(hist, res)/args.iter  # Euclidean
         # mae[n_i] += lmdk_lib.get_emd(hist, res)/args.iter  # Wasserstein
-    mae = mae/21  # Euclidean
+    # Rescaling (min-max normalization)
-    # mae = mae/11.75  # Wasserstein
+    # https://en.wikipedia.org/wiki/Feature_scaling#Rescaling_(min-max_normalization)
+    mae = (mae - mae.min())/(mae.max() - mae.min())
     print('[OK]', flush=True)
-    # # Plot bar for current distribution
+    # Plot bar for current distribution
-    # plt.bar(
-    #   x_i + x_offset,
-    #   mae,
-    #   bar_width,
-    #   label=label,
-    #   linewidth=lmdk_lib.line_width
-    # )
-    # # Change offset for next bar
-    # x_offset += bar_width
-    # Plot line
     plt.plot(
       x_i,
       mae,

code/expt/t-drive-sel.py

@@ -87,18 +87,18 @@ def main(args):
       for bgt in bgt_conf:
         for _ in range(args.iter):
 
-          lmdks, eps_out = lmdk_sel.find_lmdks(seq, lmdks, bgt['epsilon'])
+          lmdks_sel, eps_out = lmdk_sel.find_lmdks(seq, lmdks, bgt['epsilon'])
 
           # Skip
-          rls_data_s, _ = lmdk_bgt.skip(seq, lmdks, eps_out)
+          rls_data_s, _ = lmdk_bgt.skip(seq, lmdks_sel, eps_out)
           mae_s[i] += lmdk_bgt.mae(seq, rls_data_s)/args.iter
 
           # Uniform
-          rls_data_u, _ = lmdk_bgt.uniform_r(seq, lmdks, eps_out)
+          rls_data_u, _ = lmdk_bgt.uniform_r(seq, lmdks_sel, eps_out)
           mae_u[i] += lmdk_bgt.mae(seq, rls_data_u)/args.iter
 
           # Adaptive
-          rls_data_a, _, _ = lmdk_bgt.adaptive(seq, lmdks, eps_out, .5, .5)
+          rls_data_a, _, _ = lmdk_bgt.adaptive(seq, lmdks_sel, eps_out, .5, .5)
           mae_a[i] += lmdk_bgt.mae(seq, rls_data_a)/args.iter
 
           # Calculate once

code/lib/lmdk_lib.py

@@ -189,8 +189,10 @@ def get_hist(seq, lmdks):
   #   lmdks_rel = np.append(lmdks_rel, end)
 
   # Dealing with zeros.
-  if len(seq) == 0 or len(lmdks) == 0:
+  if len(seq) == 0:
     return np.zeros(math.ceil(max(seq))), 1
+  elif len(lmdks) == 0:
+    return np.zeros(1), len(seq)
 
   # Interquartile range (IQR) is a measure of statistical dispersion, being equal to the difference between 75th and 25th percentiles, or between upper and lower quartiles.
   # https://en.wikipedia.org/wiki/Interquartile_range

code/lib/lmdk_sel.py

@@ -406,7 +406,7 @@ def find_lmdks_eps(seq, lmdks, epsilon):
   '''
   # The new landmarks
   lmdks_new = lmdks
-  if len(lmdks) > 0 and len(seq) != len(lmdks):
+  if len(seq) != len(lmdks):
     # Get landmarks timestamps in sequence
     lmdks_seq = find_lmdks_seq(seq, lmdks)
     # Turn landmarks to histogram
@@ -426,7 +426,7 @@ def find_lmdks_eps(seq, lmdks, epsilon):
       # 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)
+      size = int(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])