Python深度学习模型中TensorFlow/Keras配置及CartPole模型报错排查

一、ModuleNotFoundError: No module named 'keras.utils.generic_utils' 报错解决 #

报错详情 #

/home/nate/Desktop/00_Temp/02_Coding/DeepReinforcementLearningBalanceGame.py/.venv/bin/python /home/nate/Desktop/00_Temp/02_Coding/DeepReinforcementLearningBalanceGame.py/BalancingMl.py
2024-06-29 11:13:59.059897: I tensorflow/tsl/cuda/cudart_stub.cc:28] Could not find cuda drivers on your machine, GPU will not be used.
2024-06-29 11:13:59.101402: I tensorflow/tsl/cuda/cudart_stub.cc:28] Could not find cuda drivers on your machine, GPU will not be used.
2024-06-29 11:13:59.102021: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
To enable the following instructions: AVX2 FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.
2024-06-29 11:13:59.750624: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT
Traceback (most recent call last):
  File "/home/nate/Desktop/00_Temp/02_Coding/DeepReinforcementLearningBalanceGame.py/BalancingMl.py", line 6, in <module>
    from rl.agents import DQNAgent
  File "/home/nate/Desktop/00_Temp/02_Coding/DeepReinforcementLearningBalanceGame.py/.venv/lib/python3.8/site-packages/rl/agents/__init__.py", line 2, in <module>
    from .dqn import DQNAgent, NAFAgent, ContinuousDQNAgent
  File "/home/nate/Desktop/00_Temp/02_Coding/DeepReinforcementLearningBalanceGame.py/.venv/lib/python3.8/site-packages/rl/agents/dqn.py", line 8, in <module>
    from rl.core import Agent
  File "/home/nate/Desktop/00_Temp/02_Coding/DeepReinforcementLearningBalanceGame.py/.venv/lib/python3.8/site-packages/rl/core.py", line 8, in <module>
    from rl.callbacks import (
  File "/home/nate/Desktop/00_Temp/02_Coding/DeepReinforcementLearningBalanceGame.py/.venv/lib/python3.8/site-packages/rl/callbacks.py", line 12, in <module>
    from keras.utils.generic_utils import Progbar
ModuleNotFoundError: No module named 'keras.utils.generic_utils'

原因分析 #

你使用的rl包是旧版keras-rl，仅兼容TensorFlow 1.x和独立Keras包。当前环境安装的是TensorFlow 2.x，其中Keras以tf.keras形式集成，不再存在独立的keras.utils.generic_utils路径，导致模块找不到。

解决步骤 #

1. 卸载旧版keras-rl

pip uninstall -y keras-rl

2. 安装适配TensorFlow 2.x的keras-rl2

pip install keras-rl2

代码额外修正 #

你的代码还有几处小问题，修正后才能正常运行：

• BoltzmannQPolicy需要实例化，传入类对象会报错，改为policy=BoltzmannQPolicy()
• np.mea是拼写错误，应为np.mean
• 新版TensorFlow中，Adam优化器的学习率参数建议用learning_rate替代lr（兼容旧版但更规范）

修正后的完整代码：

import random
import gym
import numpy as np
import tensorflow as tf

from rl.agents import DQNAgent
from rl.policy import BoltzmannQPolicy
from rl.memory import SequentialMemory

env = gym.make("CartPole-v1")

states = env.observation_space.shape[0]
actions = env.action_space.n

model = tf.keras.models.Sequential()
model.add(tf.keras.layers.Flatten(input_shape=(1, states)))
model.add(tf.keras.layers.Dense(24, activation="relu"))
model.add(tf.keras.layers.Dense(24, activation="relu"))
model.add(tf.keras.layers.Dense(actions, activation="linear"))

agent = DQNAgent(
    model=model,
    memory=SequentialMemory(limit=50000, window_length=1),
    policy=BoltzmannQPolicy(),  # 修正：实例化策略
    nb_actions=actions,
    nb_steps_warmup=10,
    target_model_update=0.01
)

# 修正：使用learning_rate替代lr
agent.compile(tf.keras.optimizers.Adam(learning_rate=0.001), metrics=["mae"])
agent.fit(env, nb_steps=100000, visualize=False, verbose=1)

results = agent.test(env, nb_episodes=10, visualize=True)

# 修正：np.mean拼写正确
print(np.mean(results.history["episode_reward"]))

env.close()

二、Python深度学习中按需运行TensorFlow和Keras #

1. 延迟导入模块 #

将TensorFlow/Keras的导入语句放在具体使用的函数内部，而非脚本开头，避免程序启动时就加载整个框架：

def train_model():
    import tensorflow as tf
    from rl.agents import DQNAgent
    # 后续模型定义、训练代码

2. 控制TensorFlow内存使用 #

避免TensorFlow启动时占用全部GPU内存，开启内存增长模式：

import tensorflow as tf

gpus = tf.config.list_physical_devices('GPU')
if gpus:
    try:
        for gpu in gpus:
            tf.config.experimental.set_memory_growth(gpu, True)
    except RuntimeError as e:
        print(e)

如果使用CPU，可设置软设备放置，让TensorFlow自动选择可用设备：

tf.config.set_soft_device_placement(True)

3. 及时释放资源 #

训练/推理完成后，清理Keras会话释放内存：

tf.keras.backend.clear_session()

对于不再使用的模型，可手动删除对象并触发垃圾回收：

import gc

del model, agent
gc.collect()

4. 按需加载模型 #

如果模型已保存，仅在需要推理时加载：

def predict_with_model():
    model = tf.keras.models.load_model("saved_model.h5")
    # 推理逻辑
    model.predict(...)
    tf.keras.backend.clear_session()

内容的提问来源于stack exchange，提问作者ndh.py