Minor changes

1 files changed, 25 insertions, 4 deletions

diff --git a/qlearning.py b/qlearning.py
index e35d778..60775e3 100755
--- a/qlearning.py
+++ b/qlearning.py
@@ -38,7 +38,12 @@ def isWall(h,game):
         return(True)
     return(False)
 
+
+last_state=None
+last_action=None
 def event_handler(game,event):
+    global last_state,last_action
+    
     h=game.snake[0]
     left=(h[0]-1,h[1])
     right=(h[0]+1,h[1])
@@ -70,14 +75,30 @@ def event_handler(game,event):
     # This come from me I do not now if it is the best way to identify a state
     state=2**11*snake_go_up+2**10*snake_go_right+2**9*snake_go_down+2**8*snake_go_left+2**7*apple_up+2**6*apple_right+2**5*apple_down+2**4*apple_left+2**3*obstacle_up+2**2*obstacle_right+2**1*obstacle_down+obstacle_left
 
-
+    # Choose an action
     if np.max(qtable[state]) > 0:
         action = np.argmax(qtable[state])
     else:
-        action=random.choice((12,3,6,9))
-
-    game.direction=action
+        action=random.choice((0,1,2,3))
+
+    # Update current state Q
+    if last_state != None:
+        qtable[last_state,last_action]=qtable[last_state,last_action]+0.5*(reward+0.5*qtable[state])
+    last_state=state
+    last_action=action
+
+    # Apply the action
+    snake_action=12
+    if action==1:
+        snake_action=3
+    elif action==2:
+        snake_action=6
+    elif action==3:
+        snake_action=9
+    game.direction=snake_action
 
 for i in range(0,10):
+    last_state=None
+    last_action=None
     score=game.run(event_handler=event_handler)
     print("Game ended with "+str(score))
\ No newline at end of file