python - Keras 符号输入/输出未实现 `__len__` 错误

Question

我想让 AI 玩我的自定义环境，不幸的是，当我运行我的代码时，出现以下错误:

  File "C:\Program Files\JetBrains\PyCharm Community Edition 2021.2\plugins\python-ce\helpers\pydev\_pydev_bundle\pydev_umd.py", line 198, in runfile
    pydev_imports.execfile(filename, global_vars, local_vars)  # execute the script
  File "C:\Program Files\JetBrains\PyCharm Community Edition 2021.2\plugins\python-ce\helpers\pydev\_pydev_imps\_pydev_execfile.py", line 18, in execfile
    exec(compile(contents+"\n", file, 'exec'), glob, loc)
  File "D:/PycharmProjects/Custom Enviroment AI/Enviroment.py", line 88, in <module>
    DQN = buildAgent(model, actions)
  File "D:/PycharmProjects/Custom Enviroment AI/Enviroment.py", line 82, in buildAgent
    dqn = DQNAgent(model, memory=memory, policy=policy, nb_actions=actions, nb_steps_warmup=10,
  File "D:\PycharmProjects\Custom Enviroment AI\venv\lib\site-packages\rl\agents\dqn.py", line 108, in __init__
    if hasattr(model.output, '__len__') and len(model.output) > 1:
  File "D:\PycharmProjects\Custom Enviroment AI\venv\lib\site-packages\keras\engine\keras_tensor.py", line 221, in __len__
    raise TypeError('Keras symbolic inputs/outputs do not '
TypeError: Keras symbolic inputs/outputs do not implement `__len__`. You may be trying to pass Keras symbolic inputs/outputs to a TF API that does not register dispatching, preventing Keras from automatically converting the API call to a lambda layer in the Functional Model. This error will also get raised if you try asserting a symbolic input/output directly.

错误说你不应该使用 len() 而你应该使用 .shape istead，不幸的是这似乎是 tensorflow 中的一个错误我的完整代码是:

from rl.memory import SequentialMemory
from rl.policy import BoltzmannQPolicy
from rl.agents.dqn import DQNAgent
from keras.layers import Dense
import tensorflow as tf
import numpy as np
import random
import pygame
import gym


class Env(gym.Env):
    def __init__(self):
        self.action_space = gym.spaces.Discrete(4)
        self.observation_space = gym.spaces.MultiDiscrete([39, 27])
        self.screen = pygame.display.set_mode((800, 600))
        self.PlayerX = 0
        self.PlayerY = 0
        self.FoodX = 0
        self.FoodY = 0
        self.state = [self.FoodX - self.PlayerX + 19, self.FoodY - self.PlayerY + 14]
        self.timeLimit = 1000

    def render(self, mode="human"):
        self.screen.fill((0, 0, 0))
        pygame.draw.rect(self.screen, (255, 255, 255), pygame.Rect(self.PlayerX * 40, self.PlayerY * 40, 40, 40))
        pygame.draw.rect(self.screen, (255, 0, 0), pygame.Rect(self.FoodX * 40, self.FoodY * 40, 40, 40))
        pygame.display.update()

    def reset(self):
        self.FoodX = random.randint(1, 19)
        self.FoodY = random.randint(1, 14)
        self.PlayerX = 0
        self.PlayerY = 0
        self.timeLimit = 1000
        return self.state

    def step(self, action):
        self.timeLimit -= 1
        reward = -1

        if action == 0 and self.PlayerY > 0:
            self.PlayerY -= 1
        if action == 1 and self.PlayerX > 0:
            self.PlayerX -= 1
        if action == 2 and self.PlayerY < 14:
            self.PlayerY += 1
        if action == 3 and self.PlayerX < 19:
            self.PlayerX += 1

        if self.PlayerX == self.FoodX and self.PlayerY == self.FoodY:
            reward += 30
            self.FoodX = random.randint(1, 19)
            self.FoodY = random.randint(1, 14)

        if self.timeLimit <= 0:
            done = True
        else:
            done = False

        self.state = [self.FoodX - self.PlayerX, self.FoodY - self.PlayerY]
        return self.state, reward, done


env = Env()

states = env.observation_space.shape
actions = env.action_space.n


def build_model(states, actions):
    model = tf.keras.Sequential()
    model.add(Dense(2, activation='relu', input_shape=states))
    model.add(Dense(4, activation='relu'))
    model.add(Dense(actions, activation='linear'))
    return model


def buildAgent(model, actions):
    policy = BoltzmannQPolicy()
    memory = SequentialMemory(limit=50000, window_length=1)
    dqn = DQNAgent(model, memory=memory, policy=policy, nb_actions=actions, nb_steps_warmup=10,
                   target_model_update=1e-2)
    return dqn


model = build_model(states, actions)
DQN = buildAgent(model, actions)
DQN.compile(tf.keras.optimizers.Adam(learning_rate=1e-3), metrics=['mae'])
DQN.fit(env, nb_steps=50000, visualize=False, verbose=1)
scores = DQN.test(env, nb_episodes=100, visualize=True)
print(np.mean(scores.history['episode_reward']))
pygame.quit()
model.save('model.h5')

我使用 Tensorflow:2.8.0。这似乎是 Tensorflow 代码中的一个错误，但我不知道该怎么做

Answer 1

如前所述here , 你需要安装更新版本的 keras-rl:

!pip install keras-rl2

您还需要为输入形状添加一个额外的维度，并在末尾添加一个Flatten 层，因为Keras 在使用DQN< 时需要这样做代理:

def build_model(states, actions):
    model = tf.keras.Sequential()
    model.add(Dense(2, activation='relu', input_shape=(1, states[0])))
    model.add(Dense(4, activation='relu'))
    model.add(Dense(actions, activation='linear'))
    model.add(Flatten())
    return model

最后，您自定义环境中的step 方法还必须返回一个info 字典(我刚刚创建了一个空字典):

    def step(self, action):
        self.timeLimit -= 1
        reward = -1

        if action == 0 and self.PlayerY > 0:
            self.PlayerY -= 1
        if action == 1 and self.PlayerX > 0:
            self.PlayerX -= 1
        if action == 2 and self.PlayerY < 14:
            self.PlayerY += 1
        if action == 3 and self.PlayerX < 19:
            self.PlayerX += 1

        if self.PlayerX == self.FoodX and self.PlayerY == self.FoodY:
            reward += 30
            self.FoodX = random.randint(1, 19)
            self.FoodY = random.randint(1, 14)

        if self.timeLimit <= 0:
            done = True
        else:
            done = False

        self.state = [self.FoodX - self.PlayerX, self.FoodY - self.PlayerY]
        return self.state, reward, done, {}

如果您进行这些更改，它应该可以正常工作。这是完整的工作代码:

from rl.memory import SequentialMemory
from rl.policy import BoltzmannQPolicy
from rl.agents.dqn import DQNAgent
from keras.layers import Dense, Flatten
import tensorflow as tf
import numpy as np
import random
import pygame
import gym


class Env(gym.Env):
    def __init__(self):
        self.action_space = gym.spaces.Discrete(4)
        self.observation_space = gym.spaces.MultiDiscrete([39, 27])
        self.screen = pygame.display.set_mode((800, 600))
        self.PlayerX = 0
        self.PlayerY = 0
        self.FoodX = 0
        self.FoodY = 0
        self.state = [self.FoodX - self.PlayerX + 19, self.FoodY - self.PlayerY + 14]
        self.timeLimit = 1000

    def render(self, mode="human"):
        self.screen.fill((0, 0, 0))
        pygame.draw.rect(self.screen, (255, 255, 255), pygame.Rect(self.PlayerX * 40, self.PlayerY * 40, 40, 40))
        pygame.draw.rect(self.screen, (255, 0, 0), pygame.Rect(self.FoodX * 40, self.FoodY * 40, 40, 40))
        pygame.display.update()

    def reset(self):
        self.FoodX = random.randint(1, 19)
        self.FoodY = random.randint(1, 14)
        self.PlayerX = 0
        self.PlayerY = 0
        self.timeLimit = 1000
        return self.state

    def step(self, action):
        self.timeLimit -= 1
        reward = -1

        if action == 0 and self.PlayerY > 0:
            self.PlayerY -= 1
        if action == 1 and self.PlayerX > 0:
            self.PlayerX -= 1
        if action == 2 and self.PlayerY < 14:
            self.PlayerY += 1
        if action == 3 and self.PlayerX < 19:
            self.PlayerX += 1

        if self.PlayerX == self.FoodX and self.PlayerY == self.FoodY:
            reward += 30
            self.FoodX = random.randint(1, 19)
            self.FoodY = random.randint(1, 14)

        if self.timeLimit <= 0:
            done = True
        else:
            done = False

        self.state = [self.FoodX - self.PlayerX, self.FoodY - self.PlayerY]
        return self.state, reward, done, {}


env = Env()

states = env.observation_space.shape
actions = env.action_space.n

def build_model(states, actions):
    model = tf.keras.Sequential()
    model.add(Dense(2, activation='relu', input_shape=(1, states[0])))
    model.add(Dense(4, activation='relu'))
    model.add(Dense(actions, activation='linear'))
    model.add(Flatten())
    return model

def buildAgent(model, actions):
    policy = BoltzmannQPolicy()
    memory = SequentialMemory(limit=50000, window_length=1)
    dqn = DQNAgent(model, memory=memory, policy=policy, nb_actions=actions, nb_steps_warmup=10,
                   target_model_update=1e-2)
    return dqn


model = build_model(states, actions)
DQN = buildAgent(model, actions)

DQN.compile(tf.keras.optimizers.Adam(learning_rate=1e-3), metrics=['mae'])
DQN.fit(env, nb_steps=50000, visualize=False, verbose=1)
scores = DQN.test(env, nb_episodes=100, visualize=True)
print(np.mean(scores.history['episode_reward']))
pygame.quit()
model.save('model.h5')

有关详细信息，请参阅 docs .