horoiwa

import gym
from gym import wrappers


env = gym.make(ENV_NAME)
env = wrappers.Monitor(env, "./log", force=True,
                       video_callable=(lambda ep: ep % 25 == 0))

agent = DQNAgent(env=env)
agent.play(episodes=400)

horoiwa

class DQNAgent:
    
    """ ==== 中略 ==== """
    
    def play(self, episodes):

        total_rewards = []

        for n in range(episodes):

horoiwa

import collections

@dataclass
class Experience:

    state: np.ndarray

    action: int

    reward: float

horoiwa

import numpy as np
import tensorflow as tf
import tensorflow.keras.layers as kl


class QNetwork(tf.keras.Model):

    def __init__(self, action_space, lr=0.001):

        super(QNetwork, self).__init__()

horoiwa

class DQNAgent:
  
    """"略：その他のメソッド""""
    
    def update_qnetwork(self):
        
        (states, actions, rewards,
         next_states, dones) = self.get_minibatch(self.BATCH_SIZE)

        next_Qs = np.max(self.target_network.predict(next_states), axis=1)

horoiwa

import threading

import tensorflow as tf
import gym


class GlobalCounter:
  def __init__(self):
    n = 0

horoiwa

import tensorflow as tf
import tensorflow.keras.layers as kl
import tensorflow_probability as tfp
import numpy as np


class ActorCriticNet(tf.keras.Model):

    def __init__(self, action_space=2):

horoiwa

class A3CAgent:
  
  """ 中略 """
  
  def play(self, coord):

        self.total_reward = 0

        self.state = self.env.reset()

horoiwa

class A3CAgent:
  
    """ 中略 """

    def compute_loss(self, states, actions, discounted_rewards):

        states = tf.convert_to_tensor(
            np.vstack(states), dtype=tf.float32)

        values, logits = self.local_ACNet(states)

horoiwa

import gym


def envfunc():
    env = gym.make("BreakoutDeterministic-v4")
    return env

  
class A2CAgent: