调用seaborn.heatmap()遇AttributeError: 'NoneType'无canvas属性问题

在Google Colab的Jupyter Notebook中执行Python代码调用seaborn.heatmap()绘制混淆矩阵时，触发AttributeError，提示'NoneType' object has no attribute 'canvas'。相同代码在Intellij IDEA中可正常运行，且两端使用的matplotlib（3.5.2）和seaborn（0.11.2）版本完全一致。

报错代码 #

ax = sns.heatmap(conf_matrices[j], annot=True, cmap='Blues')

完整报错栈 #

Traceback (most recent call last):
  File "/content/IMUSensorsRecognizer/source-files/Main.py", line 334, in <module>
    show_and_save_images(save, scores, confusion_matrices, params, test_sizes,
  File "/content/IMUSensorsRecognizer/source-files/Main.py", line 93, in show_and_save_images
    ax = sns.heatmap(conf_matrices[j], annot=True, cmap='Blues')
  File "/usr/local/envs/py39/lib/python3.9/site-packages/seaborn/_decorators.py", line 46, in inner_f
    return f(**kwargs)
  File "/usr/local/envs/py39/lib/python3.9/site-packages/seaborn/matrix.py", line 550, in heatmap
    ax = plt.gca()
  File "/usr/local/envs/py39/lib/python3.9/site-packages/matplotlib/pyplot.py", line 2274, in gca
    return gcf().gca(**kwargs)
  File "/usr/local/envs/py39/lib/python3.9/site-packages/matplotlib/pyplot.py", line 867, in gcf
    return figure()
  File "/usr/local/envs/py39/lib/python3.9/site-packages/matplotlib/pyplot.py", line 812, in figure
    fig = manager.canvas.figure
AttributeError: 'NoneType' object has no attribute 'canvas'

混淆矩阵示例 #

conf_matrices[j]的结构示例（由sklearn的confusion_matrix累加生成）：

[[3091 2152  581   70  276]
 [1741 3454  491  122  255]
 [ 721  716 4330   60  382]
 [ 231  344  261 3198  281]
 [ 878  429  539   67 1131]]

运行代码的Notebook Cell #

%%bash
ls
source activate py39
python --version
cd ~/../content/IMUSensorsRecognizer
pip install -r requirements.txt
pip install ipykernel 
cd source-files
ls

python Main.py rf 50 0.75 1 5 4 7 24 t f

排查过程与结论 #

• 尝试断开重连Colab会话、删除重新挂载数据，问题仍存在
• 更换Colab分配的机器后，问题完全解决

推测该问题是Colab分配的特定机器存在matplotlib后端环境异常，导致无法正常创建绘图上下文（figure/manager）。

临时规避方案 #

如果再次遇到同类问题，可尝试以下两种方法：

1. 手动创建绘图对象并传入heatmap
   显式创建figure和axes，避免依赖plt.gca()自动获取上下文：

   import matplotlib.pyplot as plt
   # 显式创建figure和axes
   fig, ax = plt.subplots(figsize=(8, 6))
   # 将ax传入heatmap
   sns.heatmap(conf_matrices[j], annot=True, cmap='Blues', ax=ax)
   

2. 强制指定matplotlib后端
   在代码开头添加以下配置，指定稳定的绘图后端：

   import matplotlib
   # 针对Colab使用inline后端，或用'Agg'作为无界面后端
   matplotlib.use('inline')
   

内容的提问来源于stack exchange，提问作者Federico Locatelli