使用Keras训练模型时遇TypeError: Protocols cannot be instantiated错误求助

问题重现 #

用户实现的MNIST分类模型代码如下：

(x_train,y_train),(x_test,y_test) = mnist.load_data()
x_train = x_train.reshape(60000,784)
x_test = x_test.reshape(10000,784)

x_train = x_train.astype('float32')
x_test = x_test.astype('float32')

#Normalize
x_train /= 255
x_test /= 255

#convert label to one hot
y_train = np_utils.to_categorical(y_train)
y_test = np_utils.to_categorical(y_test)

myModel = Sequential()
myModel.add(Dense(500, activation='relu',input_shape=(784,)))
myModel.add(Dense(100,activation='relu'))
myModel.add(Dense(10,activation='softmax'))

myModel.compile(optimizer=SGD(lr=0.001),loss=categorical_crossentropy,metrics=['accuracy'])

network_history = myModel.fit(x_train,y_train,batch_size=128,epochs=2)

调用model.fit()时触发报错：

network_history = myModel.fit(x_train,y_train,batch_size=(128))
Traceback (most recent call last):

  File "C:\Users\Parisima\AppData\Local\Temp/ipykernel_13876/4252018919.py", line 1, in <module>
    network_history = myModel.fit(x_train,y_train,batch_size=(128))

  File "C:\Users\Parisima\anaconda3\envs\tf_envs\lib\site-packages\keras\utils\traceback_utils.py", line 67, in error_handler
    raise e.with_traceback(filtered_tb) from None

  File "C:\Users\Parisima\anaconda3\envs\tf_envs\lib\typing.py", line 1083, in _no_init
    raise TypeError('Protocols cannot be instantiated')

TypeError: Protocols cannot be instantiated

错误原因分析 #

这个错误核心原因是模块导入不规范或版本兼容性问题：

1. 代码中未显式导入Sequential、Dense、SGD、categorical_crossentropy、np_utils等组件，可能导致Python解析时误将这些标识符当成了typing模块中的Protocol类，从而触发实例化错误。
2. 旧版本TensorFlow/Keras与Python 3.8+的typing模块存在适配问题，Protocol类被错误调用。
3. 报错堆栈中出现batch_size=(128)，这是将参数写成了元组，虽然用户代码里是batch_size=128，但测试时的写法也可能触发异常。

解决步骤 #

1. 添加正确的导入语句
   确保所有Keras组件都从对应模块导入，建议使用TensorFlow集成的Keras（tf.keras）以避免版本冲突：

   import tensorflow as tf
   from tensorflow.keras.models import Sequential
   from tensorflow.keras.layers import Dense
   from tensorflow.keras.optimizers import SGD
   from tensorflow.keras.losses import categorical_crossentropy
   from tensorflow.keras.utils import to_categorical
   from tensorflow.keras.datasets import mnist
   

   注意：原代码中的np_utils.to_categorical可以替换为tf.keras.utils.to_categorical，无需额外导入numpy的utils模块。

2. 修正参数写法
   确保batch_size是整数类型，不要写成元组（即batch_size=128而非batch_size=(128)）。

3. 版本兼容性检查
   如果使用Python 3.8及以上版本，建议升级TensorFlow到2.5+版本，避免旧版本typing模块的Bug。

修正后的完整代码 #

import tensorflow as tf
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense
from tensorflow.keras.optimizers import SGD
from tensorflow.keras.losses import categorical_crossentropy
from tensorflow.keras.datasets import mnist

# 加载数据
(x_train,y_train),(x_test,y_test) = mnist.load_data()
x_train = x_train.reshape(60000,784)
x_test = x_test.reshape(10000,784)

# 类型转换与归一化
x_train = x_train.astype('float32') / 255.0
x_test = x_test.astype('float32') / 255.0

# 标签转为one-hot编码
y_train = tf.keras.utils.to_categorical(y_train, 10)
y_test = tf.keras.utils.to_categorical(y_test, 10)

# 构建模型
myModel = Sequential()
myModel.add(Dense(500, activation='relu', input_shape=(784,)))
myModel.add(Dense(100, activation='relu'))
myModel.add(Dense(10, activation='softmax'))

# 编译模型
# 注意：SGD的lr参数在新版本中建议用learning_rate
myModel.compile(optimizer=SGD(learning_rate=0.001), loss=categorical_crossentropy, metrics=['accuracy'])

# 训练模型
network_history = myModel.fit(x_train, y_train, batch_size=128, epochs=2)

内容的提问来源于stack exchange，提问作者paricma