Refactoring

valueIterationAgents.py

@@ -69,10 +69,7 @@ class ValueIterationAgent(ValueEstimationAgent):
                 rewards = util.Counter()
                 actions = self.mdp.getPossibleActions(state)
                 for action in actions:
-                    transitionStatesAndProbs = self.mdp.getTransitionStatesAndProbs(state, action)
-                    for stateProb in transitionStatesAndProbs:
-                        nextState, probability = stateProb
-                        rewards[action] += probability * ( self.mdp.getReward(state, action, nextState) + self.discount * self.values[nextState])
+                    rewards[action] = self.getQValue(state, action)
                 # Copied from util.Counter code
                 all = list(rewards.items())
                 values = [x[1] for x in all]
@@ -104,8 +101,6 @@ class ValueIterationAgent(ValueEstimationAgent):
         transitionStatesAndProbs = self.mdp.getTransitionStatesAndProbs(state, action)
         qValue = 0
         for nextState, probability in transitionStatesAndProbs:
-            #Stupid, but why does it only produce north q values????
-            #qValue += probability * self.mdp.getReward(state, action, nextState)
             qValue += probability * (self.discount * self.values[nextState] + self.mdp.getReward(state, action, nextState))
         return qValue
 
@@ -165,7 +160,23 @@ class AsynchronousValueIterationAgent(ValueIterationAgent):
         ValueIterationAgent.__init__(self, mdp, discount, iterations)
 
     def runValueIteration(self):
-        "*** YOUR CODE HERE ***"
+        states = self.mdp.getStates()
+        for i in range(self.iterations):
+            try:
+                state = states.pop(0)
+            except IndexError:
+                states = self.mdp.getStates()
+            rewards = util.Counter()
+            actions = self.mdp.getPossibleActions(state)
+            for action in actions:
+                rewards[action] = self.getQValue(state, action)
+            # Copied from util.Counter code
+            all = list(rewards.items())
+            values = [x[1] for x in all]
+            if not self.mdp.isTerminal(state):
+                self.values[state] = max(values)
+            else:
+                self.values[state] = 0
 
 class PrioritizedSweepingValueIterationAgent(AsynchronousValueIterationAgent):
     """