def main():
# 初始化游戏
env = flappyBird.GameState()
# 图像输入形状和动作维度
obs_dim = resize_shape[0]
action_dim = env.action_dim
# 创建模型
model = Model(obs_dim, action_dim)
model.load_dict(paddle.load(save_model_path))
model.eval()
# 开始游戏
obs = env.reset()
episode_reward = 0
done = False
# 游戏未结束执行一直执行游戏
while not done:
obs = preprocess(obs)
obs = np.expand_dims(obs, axis=0)
obs = paddle.to_tensor(obs, dtype='float32')
action = model(obs)
action = paddle.argmax(action).numpy()[0]
obs, reward, done, info = env.step(action, is_train=False)
episode_reward += reward
print("最终得分为：{:.2f}".format(episode_reward))
def main():
# 初始化游戏
env = flappyBird.GameState()
# 开始游戏
obs = env.reset()
# 游戏未结束执行一直执行游戏
while True:
# 游戏生成的随机动作，int类型数值
action = env.action_space()
# 执行游戏
obs, reward, done, info = env.step(action, is_train=False)
print("=" * 50)
print("action:", action)
print("obs shape:", obs.shape)
print("reward:", reward)
print("terminal:", done)
print("info:", info)
if done:
obs = env.reset()
from replay_memory import ReplayMemory
# 定义训练的参数
batch_size = 256 # batch大小
num_episodes = 10000 # 训练次数
memory_size = 20000 # 内存记忆
n_step = 3 # 往后的步数
learning_rate = 1e-4 # 学习率大小
e_greed = 0.1 # 探索初始概率
gamma = 0.99 # 奖励系数
e_greed_decrement = 1e-6 # 在训练过程中，降低探索的概率
update_num = 0 # 用于计算目标模型更新次数
resize_shape = (1, 36, 52) # 训练缩放的大小，减少模型计算，原大小（288, 512）
save_model_path = "models/model.pdparams" # 保存模型路径
env = flappyBird.GameState()
obs_dim = resize_shape[0]
action_dim = env.action_dim
policyQ = Model(obs_dim, action_dim)
targetQ = Model(obs_dim, action_dim)
targetQ.eval()
n_step_buffer = []
rpm = ReplayMemory(memory_size)
optimizer = paddle.optimizer.Adam(parameters=policyQ.parameters(),
learning_rate=learning_rate)
def preprocess(observation):
observation = observation[:observation.shape[0] - 100, :]