- improved

git-svn-id: http://moon:8086/svn/projects/RL-lab@344 fda53097-d464-4ada-af97-ba876c37ca34

rl_lab.py

@@ -32,7 +32,8 @@ def to_this_door(door_last_room):
 	q, r = divmod(door_last_room + 2, 4)
 	return r
 
-def choose_door(Pn):
+def choose_door(P):
+	Pn = P / np.sum(P)
 	Pnorm = np.cumsum(Pn)
 	size = len(Pn)
 	Z = np.random.uniform(0,1)
@@ -60,7 +61,7 @@ def forget(forgetting_factor):
 			pos = (x,y)
 			P_door[pos,:] *= forgetting_factor
 
-def run(pos, N_trials, learning_rate, penalty_factor=0.99, forgetting_factor=0.999):
+def run(pos, N_trials, learning_rate, penalty_factor=0.99, forgetting_factor=0.99):
 	lab_visited = np.zeros((lab_size_x, lab_size_y))
 	moves_needed = 0
 	door_last = -1
@@ -71,12 +72,10 @@ def run(pos, N_trials, learning_rate, penalty_factor=0.99, forgetting_factor=0.9
 
 #		forget(forgetting_factor)
 
-		Pn = P_door[pos]/np.sum(P_door[pos])
-
 		P_door[pos,:] *= forgetting_factor
 		while True:
 			fail = False
-			door = choose_door(Pn)
+			door = choose_door(P_door[pos])
 			if door_last >= 0:
 				this_door = to_this_door(door_last)
 				if this_door == door:
@@ -124,7 +123,6 @@ print('Visited map after {} moves:'.format(moves_needed))
 print(lab_visited)
 print ('P_door after {} moves:'.format(moves_needed))
 
-P = normalize(P_door)
 for i in range(0, num_doors_per_room):
-	print (P[:,:,i])
+	print (P_door[:,:,i])