使用cx_Freeze打包Python应用后出现keras模块加载失败问题求助

我用cx_Freeze打包了名为AI Hound的Python应用，启动后触发ModuleNotFoundError，提示找不到keras模块。打包生成的build文件夹中存在tensorflow/python/keras目录，但程序仍尝试调用本地Python环境和项目venv中的文件。

报错信息 #

ERROR:root:UNEXPECTED ERROR:
Traceback (most recent call last):
  File "ai_hound.py", line 78, in run
  File "ai_hound.py", line 51, in load_models
  File "D:\PyPrj\FP2\venv\lib\site-packages\tensorflow\python\util\lazy_loader.py", line 58, in __getattr__
    module = self._load()
  File "D:\PyPrj\PF2\venv\lib\site-packages\tensorflow\python\util\lazy_loader.py", line 41, in _load
    module = importlib.import_module(self.__name__)
  File "C:\Users\PGD\AppData\Local\Programs\Python\Python39\lib\importlib\__init__.py", line 127, in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
  File "<frozen importlib._bootstrap>", line 1030, in _gcd_import
  File "<frozen importlib._bootstrap>", line 1007, in _find_and_load
  File "<frozen importlib._bootstrap>", line 972, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 228, in _call_with_frames_removed
  File "<frozen importlib._bootstrap>", line 1030, in _gcd_import
  File "<frozen importlib._bootstrap>", line 1007, in _find_and_load
  File "<frozen importlib._bootstrap>", line 972, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 228, in _call_with_frames_removed
  File "<frozen importlib._bootstrap>", line 1030, in _gcd_import
  File "<frozen importlib._bootstrap>", line 1007, in _find_and_load
  File "<frozen importlib._bootstrap>", line 972, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 228, in _call_with_frames_removed
  File "<frozen importlib._bootstrap>", line 1030, in _gcd_import
  File "<frozen importlib._bootstrap>", line 1007, in _find_and_load
  File "<frozen importlib._bootstrap>", line 984, in _find_and_load_unlocked
ModuleNotFoundError: No module named 'keras'

setup.py配置 #

import sys, os
from cx_Freeze import setup, Executable

PYTHON_INSTALL_DIR = os.path.dirname(os.path.dirname(os.__file__))
os.environ['TCL_LIBRARA'] =     os.path.join(PYTHON_INSTALL_DIR, "tcl", "tcl8.6")
os.environ['TK_LIBRARA'] =     os.path.join(PYTHON_INSTALL_DIR, "tcl", "tk8.6")

build_exe_options = {"packages": ["time", "sys", "os", "base64", "cv2", "logging", "codecs", "shutil", "zipfile", "fitz", "pandas", "numpy", "tensorflow", "PIL", "io", "tkinter"],
                     "includes": ["tkinter", "tensorflow"],
                     "include_files": [ os.path.join(PYTHON_INSTALL_DIR, "DLLs", "tk86t.dll"), os.path.join(PYTHON_INSTALL_DIR, "DLLs", "tcl86t.dll") ],
                     "excludes": []}

base = None
if sys.platform == "win32":
    base = "Win32GUI"

setup(
    name = "AI Hound",
    version = "1.0.1",
    description = "AI Hound",
    options = {"build_exe": build_exe_options},
    executables = [Executable("ai_hound.py", icon="icon.ico", base = base)]
)

1. Keras导入路径不匹配：TensorFlow集成的Keras是tensorflow.keras，而非独立的keras模块。如果代码中直接用import keras，会尝试寻找独立安装的Keras包，但打包时仅包含了tensorflow，未处理该导入映射。
2. cx_Freeze未识别懒加载机制：报错栈显示问题来自tensorflow.python.util.lazy_loader，cx_Freeze可能无法识别TensorFlow的懒加载逻辑，导致tensorflow.keras相关模块未被正确打包或路径映射错误。
3. 本地环境路径残留：打包后的程序仍调用本地venv路径，说明cx_Freeze未完全隔离环境，可能是部分硬编码路径或懒加载模块的路径解析未被正确处理。

1. 修改代码中的Keras导入方式 #

将代码中所有import keras改为from tensorflow import keras或import tensorflow.keras as keras，确保使用TensorFlow内置的Keras模块，避免寻找独立包。

2. 调整setup.py配置，强制包含相关模块 #

在build_exe_options中添加tensorflow.keras到packages或includes列表，同时指定namespace_packages处理TensorFlow的命名空间：

build_exe_options = {
    "packages": ["time", "sys", "os", "base64", "cv2", "logging", "codecs", "shutil", "zipfile", "fitz", "pandas", "numpy", "tensorflow", "tensorflow.keras", "PIL", "io", "tkinter"],
    "includes": ["tkinter", "tensorflow", "tensorflow.keras", "tensorflow.keras.layers", "tensorflow.keras.models"], # 根据实际使用的子模块补充
    "include_files": [os.path.join(PYTHON_INSTALL_DIR, "DLLs", "tk86t.dll"), os.path.join(PYTHON_INSTALL_DIR, "DLLs", "tcl86t.dll")],
    "namespace_packages": ["tensorflow"],
    "excludes": []
}

3. 修复环境变量拼写错误 #

setup.py中TCL_LIBRARA和TK_LIBRARA存在拼写错误，应为TCL_LIBRARY和TK_LIBRARY，这可能导致打包时依赖处理异常：

os.environ['TCL_LIBRARY'] = os.path.join(PYTHON_INSTALL_DIR, "tcl", "tcl8.6")
os.environ['TK_LIBRARY'] = os.path.join(PYTHON_INSTALL_DIR, "tcl", "tk8.6")

4. 清理旧build目录后重新打包 #

删除原有的build和dist目录，重新运行打包命令，避免旧文件干扰：

rm -r build dist
python setup.py build

5. 手动添加打包后的TensorFlow路径到sys.path #

如果上述方法无效，可在代码启动时手动添加路径：

import sys, os
if getattr(sys, 'frozen', False):
    # 获取打包后的程序目录
    app_path = os.path.dirname(sys.executable)
    # 添加tensorflow路径到sys.path
    tensorflow_path = os.path.join(app_path, 'tensorflow')
    if tensorflow_path not in sys.path:
        sys.path.insert(0, tensorflow_path)

内容的提问来源于stack exchange，提问作者Markus Bauer