Update run.py

neuralkey/run.py

@@ -14,44 +14,53 @@
 update_rules = ['hebbian', 'anti_hebbian', 'random_walk']
 update_rule = update_rules[0]
 
-#Create 2 machines
+#Create 3 machines : Alice, Bob and Eve. Eve will try to intercept the communication between
+#Alice and Bob.
 print("Creating machines : k=" + str(k) + ", n=" + str(n) + ", l=" + str(n))
 print("Using " + update_rule + " update rule.")
-machine1 = Machine(k, n, l)
-machine2 = Machine(k, n, l)
+Alice = Machine(k, n, l)
+Bob = Machine(k, n, l)
+Eve = Machine(k, n, l)
 
 #Random number generator
 def random():
 	return np.random.randint(-l, l + 1, [k, n])
 
 #Function to evaluate the synchronization score between two machines.
 def sync_score(m1, m2):
-	return 1.0 - np.average(1.0 * np.abs(m1.W - m2.W)/(2 * l - 1))
+	return 1.0 - np.average(1.0 * np.abs(m1.W - m2.W)/(2 * l))
 
 #Synchronize weights
 
 sync = False # Flag to check if weights are sync
 nb_updates = 0 # Update counter
+nb_eve_updates = 0 # To count the number of times eve updated
 start_time = time.time() # Start time
 sync_history = [] # to store the sync score after every update
 
 while(not sync):
 
 	X = random() # Create random vector of dimensions [k, n]
 
-	tauA = machine1(X) # Get output from machine1
-	tauB = machine2(X) # Get output from machine2
+	tauA = Alice(X) # Get output from Alice
+	tauB = Bob(X) # Get output from Bob
+	tauE = Eve(X) # Get output from Eve
 
-	machine1.update(tauB, update_rule) # Update machine1 with machine2's output
-	machine2.update(tauA, update_rule) # Update machine2 with machine1's output
+	Alice.update(tauB, update_rule) # Update Alice with Bob's output
+	Bob.update(tauA, update_rule) # Update Bob with Alice's output
+
+	#Eve would update only if tauA = tauB = tauE
+	if tauA == tauB == tauE:
+		Eve.update(tauA, update_rule)
+		nb_eve_updates += 1
 
 	nb_updates += 1
 
-	score = 100 * sync_score(machine1, machine2) # Calculate the synchronization of the 2 machines
+	score = 100 * sync_score(Alice, Bob) # Calculate the synchronization of the 2 machines
 
 	sync_history.append(score) # Add sync score to history, so that we can plot a graph later.
 
-	sys.stdout.write('\r' + "Synchronization = " + str(int(score)) + "%   /  Updates = " + str(nb_updates)) 
+	sys.stdout.write('\r' + "Synchronization = " + str(int(score)) + "%   /  Updates = " + str(nb_updates) + " / Eve's updates = " + str(nb_eve_updates)) 
 	if score == 100: # If synchronization score is 100%, set sync flag = True
 		sync = True
 
@@ -63,6 +72,13 @@ def sync_score(m1, m2):
 print ('Time taken = ' + str(time_taken)+ " seconds.")
 print ('Updates = ' + str(nb_updates) + ".")
 
+#See if Eve got what she wanted:
+eve_score = 100 * int(sync_score(Alice, Eve))
+if eve_score > 100:
+	print("Oops! Nosy Eve synced her machine with Alice's and Bob's !")
+else:
+	print("Eve's machine is only " + str(eve_score) + " % " + "synced with Alice's and Bob's and she did " + str(nb_eve_updates) + " updates.") 
+
 #Plot graph 
 import matplotlib.pyplot as mpl
 mpl.plot(sync_history)