tensorflow2.0保存和恢复模型

方法1&＃xff1a;只保存模型的权重和偏置

这种方法不会保存整个网络的结构&＃xff0c;只是保存模型的权重和偏置&＃xff0c;所以在后期恢复模型之前&＃xff0c;必须手动创建和之前模型一模一样的模型&＃xff0c;以保证权重和偏置的维度和保存之前的相同。

tf.keras.model类中的save_weights方法和load_weights方法&＃xff0c;参数解释我就直接搬运官网的内容了。

save_weights(filepath,overwrite&＃61;True,save_format&＃61;None
)

 Arguments:

• filepath: String, path to the file to save the weights to. When saving in TensorFlow format, this is the prefix used for checkpoint files (multiple files are generated). Note that the &＃39;.h5&＃39; suffix causes weights to be saved in HDF5 format.
• overwrite: Whether to silently overwrite any existing file at the target location, or provide the user with a manual prompt.
• save_format: Either &＃39;tf&＃39; or &＃39;h5&＃39;. A filepath ending in &＃39;.h5&＃39; or &＃39;.keras&＃39; will default to HDF5 if save_format is None. Otherwise None defaults to &＃39;tf&＃39;.

load_weights(filepath,by_name&＃61;False
)

 实例1&＃xff1a;

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import datasets, layers, optimizers# step1 加载训练集和测试集合
mnist &＃61; tf.keras.datasets.mnist
(x_train, y_train),(x_test, y_test) &＃61; mnist.load_data()
x_train, x_test &＃61; x_train / 255.0, x_test / 255.0# step2 创建模型
def create_model():return tf.keras.models.Sequential([tf.keras.layers.Flatten(input_shape&＃61;(28, 28)),tf.keras.layers.Dense(512, activation&＃61;&＃39;relu&＃39;),tf.keras.layers.Dropout(0.2),tf.keras.layers.Dense(10, activation&＃61;&＃39;softmax&＃39;)])
model &＃61; create_model()# step3 编译模型 主要是确定优化方法&＃xff0c;损失函数等
model.compile(optimizer&＃61;&＃39;adam&＃39;,loss&＃61;&＃39;sparse_categorical_crossentropy&＃39;,metrics&＃61;[&＃39;accuracy&＃39;])# step4 模型训练 训练一个epochs
model.fit(x&＃61;x_train,y&＃61;y_train,epochs&＃61;1,)# step5 模型测试
loss, acc &＃61; model.evaluate(x_test, y_test)
print("train model, accuracy:{:5.2f}%".format(100 * acc))# step6 保存模型的权重和偏置
model.save_weights(&＃39;./save_weights/my_save_weights&＃39;)# step7 删除模型
del model# step8 重新创建模型
model &＃61; create_model()
model.compile(optimizer&＃61;&＃39;adam&＃39;,loss&＃61;&＃39;sparse_categorical_crossentropy&＃39;,metrics&＃61;[&＃39;accuracy&＃39;])# step9 恢复权重
model.load_weights(&＃39;./save_weights/my_save_weights&＃39;)# step10 测试模型
loss, acc &＃61; model.evaluate(x_test, y_test)
print("Restored model, accuracy:{:5.2f}%".format(100 * acc))

train model, accuracy:96.55%

Restored model, accuracy:96.55%

可以看到在模型的权重和偏置恢复之后&＃xff0c;在测试集合上同样达到了训练之前相同的准确率。

方法2&＃xff1a;直接保存整个模型

这种方法会将网络的结构&＃xff0c;权重和优化器的状态等参数全部保存下来&＃xff0c;后期恢复的时候就没必要创建新的网络了。

tf.keras.model类中的save方法和load_model方法

save(filepath,overwrite&＃61;True,include_optimizer&＃61;True,save_format&＃61;None
)

Arguments:

• filepath: String, path to SavedModel or H5 file to save the model.
• overwrite: Whether to silently overwrite any existing file at the target location, or provide the user with a manual prompt.
• include_optimizer: If True, save optimizer&＃39;s state together.
• save_format: Either &＃39;tf&＃39; or &＃39;h5&＃39;, indicating whether to save the model to Tensorflow SavedModel or HDF5. The default is currently &＃39;h5&＃39;, but will switch to &＃39;tf&＃39; in TensorFlow 2.0. The &＃39;tf&＃39; option is currently disabled (use tf.keras.experimental.export_saved_model instead).

实例2&＃xff1a;

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import datasets, layers, optimizers# step1 加载训练集和测试集合
mnist &＃61; tf.keras.datasets.mnist
(x_train, y_train),(x_test, y_test) &＃61; mnist.load_data()
x_train, x_test &＃61; x_train / 255.0, x_test / 255.0# step2 创建模型
def create_model():return tf.keras.models.Sequential([tf.keras.layers.Flatten(input_shape&＃61;(28, 28)),tf.keras.layers.Dense(512, activation&＃61;&＃39;relu&＃39;),tf.keras.layers.Dropout(0.2),tf.keras.layers.Dense(10, activation&＃61;&＃39;softmax&＃39;)])
model &＃61; create_model()# step3 编译模型 主要是确定优化方法&＃xff0c;损失函数等
model.compile(optimizer&＃61;&＃39;adam&＃39;,loss&＃61;&＃39;sparse_categorical_crossentropy&＃39;,metrics&＃61;[&＃39;accuracy&＃39;])# step4 模型训练 训练一个epochs
model.fit(x&＃61;x_train,y&＃61;y_train,epochs&＃61;1,)# step5 模型测试
loss, acc &＃61; model.evaluate(x_test, y_test)
print("train model, accuracy:{:5.2f}%".format(100 * acc))# step6 保存模型的权重和偏置
model.save(&＃39;my_model.h5&＃39;) # creates a HDF5 file &＃39;my_model.h5&＃39;# step7 删除模型
del model # deletes the existing model# step8 恢复模型
# returns a compiled model
# identical to the previous one
restored_model &＃61; tf.keras.models.load_model(&＃39;my_model.h5&＃39;)# step9 测试模型
loss, acc &＃61; restored_model.evaluate(x_test, y_test)
print("Restored model, accuracy:{:5.2f}%".format(100 * acc))

train model, accuracy:96.94%

Restored model, accuracy:96.94%

方法3&＃xff1a;使用tf.keras.callbacks.ModelCheckpoint方法在训练过程中保存模型

该方法继承自tf.keras.callbacks类&＃xff0c;一般配合mode.fit函数使用

ref&＃xff1a;

https://tensorflow.google.cn/versions/r2.0/api_docs/python/tf/keras/Model#save

https://blog.csdn.net/pan15125284/article/details/93605559

https://blog.csdn.net/qq_31456593/article/details/88829202

https://tensorflow.google.cn/beta/guide/keras/saving_and_serializing

https://tensorflow.google.cn/versions/r2.0/api_docs/python/tf/keras/callbacks/ModelCheckpoint