在Karpathy寫的下面的代碼中，為什么我們有這一行（為什么我們需要比較均勻分布來選擇一個動作，而策略函數是這樣做的）  # forward the policy network and sample an action from the returned probability  aprob, h = policy_forward(x)  action = 2 if np.random.uniform() < aprob else 3 # roll the dice!而不僅僅是 # forward the policy network and sample an action from the returned probability  aprob, h = policy_forward(x)  action = 2 if 0.5 < aprob else 3 # roll the dice!....Karpathy 的完整代碼來自：https://gist.github.com/karpathy/a4166c7fe253700972fcbc77e4ea32c5