go

agent_afterstate.py

@@ -11,7 +11,7 @@ class Action(IntEnum):
     RIGHT = 4
 
 class afterstateAgent:
-    def __init__(self, mat, TD_lambda = 0.0, alpha = 0.0025, gamma = 0.95, epsilon = 0.01, verbose= True, symmetric=1, tuple = 2):
+    def __init__(self, mat, TD_lambda = 0.0, alpha = 0.0025, epsilon = 0.01, verbose= True, symmetric=1, tuple = 2):
         self.state_per_tile = 12
         self.commands = { Action.UP: up, Action.DOWN: down, Action.LEFT: left, Action.RIGHT: right}
         self.alpha = alpha
@@ -143,8 +143,8 @@ class afterstateAgent:
         else:
             self._reset_trace()
         for tr, ind in zip(self.trace, self.index):
-            v = tr.get(ind,0)
-            v += 1 
+            v = tr.get(ind,0.0)
+            v += 1.0
             tr[ind] = v
         # print(self.trace[0])
         # print(np.sum(self.trace,axis=1))
@@ -164,7 +164,7 @@ class afterstateAgent:
             return -1
         if done== 1:
             done = 0 #1 means no reward
-        phi_array = np.array([w[i] for w,i in zip(self.W, self._index(state))])
+        phi_array = np.array([w[i] for w,i in zip(self.W, self._index(afterstate))])
         if act:
             # in the act phase, explore when there is 0
             if 0 in phi_array:
@@ -203,7 +203,7 @@ class afterstateAgent:
             return
         s,r = self.test_next(self._action_index,self.state)
         n = next_state
-        if self.symmetric>0:
+        if self.symmetric>0.1:
             for i in range(4):
                 s = transpose(s)
                 n = transpose(n)