go

agent.py

@@ -30,8 +30,8 @@ class RandomAgent():
         pass
 
 class qLearningAgent:
-    def __init__(self, mat, TD_lambda = 0.8, alpha = 0.05, gamma = 0.9, epsilon = 0.01):
-        self.state_per_tile = 11
+    def __init__(self, mat, TD_lambda = 0.8, alpha = 0.05, gamma = 0.95, epsilon = 0.005):
+        self.state_per_tile = 12
         self.alpha = alpha
         self.gamma = gamma
         self.epsilon = epsilon # e-greedy
@@ -161,7 +161,7 @@ class qLearningAgent:
                 n[k] += tr[k]
                 index = np.where(n[k]!=0)# can't divide by zeros :/
                 # print(n[k])
-                w[k][index] += target*tr[k][index]/n[k][index]
+                w[k][index] += target*tr[k][index]/np.sqrt(n[k][index])
                 # w[k] += target*tr[k]
                 # print(w[k])
         # print(self._target(next_state,reward) \