I want to make an AI playing my custom environment, unfortunately, when I run my code, following error accrues:
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.
The error says that you souldn't use len() and you should use .shape istead, unfortunately this seems to be an error inside tensorflow My full code is:
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')
I use Tensorflow: 2.8.0. This seems to be an error in Tensorflow's code but I have no idea what to do
CodePudding user response：
As mentioned here, you need to install a newer version of keras-rl:
!pip install keras-rl2
You also need to add an extra dimension to your input shape and a Flatten layer at the end, since Keras expects this when working with the DQN agent:
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
Lastly, your step method in your custom environment must also return an info dictionary (I just created an empty one):
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, {}
If you make these changes, it should work fine. Here is the full working code:
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')
For more information, see the docs.