Decrease testing epsilon

0.05 allows opponent to score on Pong

Author: Kaixhin

agent.py

@@ -43,7 +43,7 @@ def act(self, state):
       return (self.online_net(state.unsqueeze(0)) * self.support).sum(2).argmax(1).item()
 
   # Acts with an ε-greedy policy
-  def act_e_greedy(self, state, epsilon=0.05):
+  def act_e_greedy(self, state, epsilon=0.001):
     return random.randrange(self.action_space) if random.random() < epsilon else self.act(state)
 
   def learn(self, mem):