Question

阅读许多类似的问题，其中大多数提到您不应该尝试序列化不可序列化的对象。我无法理解该问题。我能够将模型另存为.h5文件，但这不能达到我想要做的目的。请帮忙！

    def image_generator(train_data_dir, test_data_dir):
        train_datagen = ImageDataGenerator(rescale=1/255,
                                          rotation_range = 30,  
                                          zoom_range = 0.2, 
                                          width_shift_range=0.1,  
                                          height_shift_range=0.1,
                                          validation_split = 0.15)
      
        test_datagen = ImageDataGenerator(rescale=1/255)
        
        train_generator = train_datagen.flow_from_directory(train_data_dir,
                                      target_size = (160,160),
                                      batch_size = 32,
                                      class_mode = 'categorical',
                                      subset='training')
        
        val_generator = train_datagen.flow_from_directory(train_data_dir,
                                      target_size = (160,160),
                                      batch_size = 32,
                                      class_mode = 'categorical',
                                      subset = 'validation')
        
        test_generator = test_datagen.flow_from_directory(test_data_dir,
                                     target_size=(160,160),
                                     batch_size = 32,
                                     class_mode = 'categorical')
        return train_generator, val_generator, test_generator
    
    def model_output_for_TL (pre_trained_model, last_output):    
        x = Flatten()(last_output)
        
        # Dense hidden layer
        x = Dense(512, activation='relu')(x)
        x = Dropout(0.2)(x)
        
        # Output neuron. 
        x = Dense(2, activation='softmax')(x)
        
        model = Model(pre_trained_model.input, x)
        
        return model
    
    train_generator, validation_generator, test_generator = image_generator(train_dir,test_dir)
    
    pre_trained_model = InceptionV3(input_shape = (160, 160, 3), 
                                    include_top = False, 
                                    weights = 'imagenet')
    for layer in pre_trained_model.layers:
      layer.trainable = False
    last_layer = pre_trained_model.get_layer('mixed5')
    last_output = last_layer.output
    model_TL = model_output_for_TL(pre_trained_model, last_output)
    
    model_TL.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])
    
    history_TL = model_TL.fit(
          train_generator,
          steps_per_epoch=10,  
          epochs=10,
          verbose=1,
          validation_data = validation_generator)
    
    pickle.dump(model_TL,open('img_model.pkl','wb'))

Answer 1

我能够使用Google Colab在TF 2.3.0中复制您的问题

import pickle
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense

model = Sequential()
model.add(Dense(1, input_dim=42, activation='sigmoid'))
model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])

with open('model.pkl', 'wb') as f:
    pickle.dump(model, f)

输出：

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-1-afb2bf58a891> in <module>()
      8 
      9 with open('model.pkl', 'wb') as f:
---> 10     pickle.dump(model, f)

TypeError: can't pickle _thread.RLock objects

@adriangb，有关github中有关此问题的建议的热修复程序，请参阅this

import pickle

from tensorflow.keras.models import Sequential, Model
from tensorflow.keras.layers import Dense
from tensorflow.python.keras.layers import deserialize, serialize
from tensorflow.python.keras.saving import saving_utils


def unpack(model, training_config, weights):
    restored_model = deserialize(model)
    if training_config is not None:
        restored_model.compile(
            **saving_utils.compile_args_from_training_config(
                training_config
            )
        )
    restored_model.set_weights(weights)
    return restored_model

# Hotfix function
def make_keras_picklable():

    def __reduce__(self):
        model_metadata = saving_utils.model_metadata(self)
        training_config = model_metadata.get("training_config", None)
        model = serialize(self)
        weights = self.get_weights()
        return (unpack, (model, training_config, weights))

    cls = Model
    cls.__reduce__ = __reduce__

# Run the function
make_keras_picklable()

# Create the model
model = Sequential()
model.add(Dense(1, input_dim=42, activation='sigmoid'))
model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])

# Save
with open('model.pkl', 'wb') as f:
    pickle.dump(model, f)

Answer 2

可以在 here 找到已接受答案中链接的修补程序的改进版本。虽然它有点复杂，但它更有可能在 TensorFlow 的未来版本中工作。此版本还修复了 tensorflow/tensorflow#44670。

来源：我是上面链接的修补程序以及此改进版本的作者。

收到TypeError：无法腌制_thread.RLock对象

2 个答案: