PyInstaller打包TensorFlow窗口程序时model.predict报AttributeError求助

问题背景 #

在Windows环境下，使用PyInstaller打包基于CustomTkinter的TensorFlow模型训练GUI程序时，代码在IDE中运行完全正常，但通过-w参数（无控制台窗口）打包生成的.exe文件执行时，预测功能出现停滞（GUI界面仍保持活跃）。去掉-w参数打包的程序则能正常运行，排查后发现model.predict调用时触发报错：AttributeError: 'NoneType' object has no attribute 'write'。

环境信息 #

• python==3.11.5
• pyinstaller==6.1.0
• tensorflow==2.12.0

可复现的最小示例代码 #

import numpy as np 
import tensorflow as tf
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense
from sys import exc_info

X_train = np.array([75, 36, -45, 127, 117, 97, 70, 120])
y_train = np.array([23.9, 2.2, -42.8, 52.8, 47.2, 36.1, 21.1, 48.9])

neural_model = Sequential()
neural_model.add(Dense(2, input_dim = 1, activation= 'sigmoid'))
neural_model.add(Dense(2, activation= 'sigmoid'))
neural_model.add(Dense(1, activation= 'linear'))

optimizer = tf.keras.optimizers.RMSprop(0.001)
neural_model.compile(
    optimizer = optimizer,
    loss = 'mse',
    metrics = ['mae', 'mse']
    )
history = neural_model.fit(X_train, y_train,
                        epochs=3,
                        batch_size=2,
                        verbose=False)
try:
    prediction_test = neural_model.predict([-49])
except Exception as e:
    file_name = '\somename.txt'
    file_path = r"somepath" + file_name
    with open(file_path, 'a') as txt:
        txt.write(f"An error occurred: {e}\n")
        error_type, error_value, traceback_info = exc_info()
        txt.write(f"Error type: {error_type}\n")
        txt.write(f"Error value: {error_value}\n")
        txt.write(f"Traceback: {traceback_info}\n")

打包命令 #

pyinstaller --onedir -w scriptfilename.py

问题原因 #

使用-w参数打包时，程序的标准输出（sys.stdout）和标准错误（sys.stderr）流会被设置为None，而TensorFlow的predict方法内部会尝试向这些流写入日志信息，导致触发AttributeError，进而阻塞预测操作的执行。

解决方案 #

1. 初始化标准输出/错误流 #

在导入TensorFlow之前，手动将标准流重定向到空对象或文件，避免因流为None引发异常：

import sys
import os

# 处理无控制台时的标准流问题
if sys.stdout is None:
    sys.stdout = open(os.devnull, 'w')
if sys.stderr is None:
    sys.stderr = open(os.devnull, 'w')

# 之后再导入TensorFlow和其他模块
import numpy as np 
import tensorflow as tf
# ... 其余代码

或者使用自定义的空写入类（无需创建文件）：

import sys

class NullWriter:
    def write(self, msg):
        pass

if sys.stdout is None:
    sys.stdout = NullWriter()
if sys.stderr is None:
    sys.stderr = NullWriter()

# 导入后续模块

2. 补充TensorFlow的隐藏依赖 #

PyInstaller可能无法自动检测TensorFlow的所有动态导入模块，打包时添加--hidden-import参数确保依赖被正确包含：

pyinstaller --onedir -w --hidden-import tensorflow --hidden-import tensorflow.keras scriptfilename.py

3. 异步执行预测操作 #

GUI程序为单线程模型，长时间的预测操作会导致GUI假死，建议将predict调用放在独立线程中执行：

import threading

def run_prediction():
    try:
        prediction_test = neural_model.predict([-49])
        # 此处添加预测结果的处理和GUI更新逻辑
        print(prediction_test)
    except Exception as e:
        # 异常处理逻辑
        file_name = '\somename.txt'
        file_path = r"somepath" + file_name
        with open(file_path, 'a') as txt:
            txt.write(f"An error occurred: {e}\n")
            error_type, error_value, traceback_info = exc_info()
            txt.write(f"Error type: {error_type}\n")
            txt.write(f"Error value: {error_value}\n")
            txt.write(f"Traceback: {traceback_info}\n")

# 触发预测时启动线程
threading.Thread(target=run_prediction, daemon=True).start()

内容的提问来源于stack exchange，提问作者NeuroGreen