経験再生

exprience_replay.py

import collections

@dataclass
class Experience:

    state: np.ndarray

    action: int

    reward: float

    next_state: np.ndarray

    done: bool

class DQNAgent:
      
      def __init__(self):
        
        self.experiences = collections.deque(maxlen=20000)
        
        """略：その他の処理"""
      
      """略：その他のメソッド"""
      
      def get_minibatch(self, batch_size):
        
        indices = np.random.choice(len(self.experiences),
                                   size=batch_size, replace=False)

        selected_experiences = [self.experiences[i] for i in indices]

        states = [exp.state for exp in selected_experiences]
        actions = [exp.action for exp in selected_experiences]
        rewards = [exp.reward for exp in selected_experiences]
        next_states = [exp.next_state for exp in selected_experiences]
        dones = [exp.done for exp in selected_experiences]

        return (states, actions, rewards, next_states, dones)