tensorflow保存模型到本地

tensorflow保存模型到本地

#import
import tensorflow as tf #导入框架

import numpy as np #导入科学计算库

import os #调用路径

# #train, test
# mnist = tf.keras.datasets.mnist #加载手写识别数据集
#
# (x_train, y_train), (x_test, y_test) = mnist.load_data() #分配训练集和测试集的输入特征和标签
#
# x_train, x_test = x_train / 255.0, x_test / 255.0 #输入特征进行归一化处理

#第二步，train, test
# mnist = tf.keras.datasets.mnist #导入mnist数据集
# (x_train, y_train), (x_test, y_test) = mnist.load_data() #分别配置训练集和测试集的输入和标签
path = './mnist.npz' #数据路径
f = np.load(path) #加载数据
x_train, y_train = f['x_train'], f['y_train'] #导入训练集的输入和标签
x_test, y_test = f['x_test'], f['y_test']  #导入测试集的输入和标签
f.close()

x_train, x_test = x_train/255.0, x_test/255.0 #把输入特征做归一化处理，让值在0-1之间，更容易让神经网络吸收

#model.Sequential()
model = tf.keras.models.Sequential([ #实现前向传播
    tf.keras.layers.Flatten(), #拉成一维
    tf.keras.layers.Dense(128, activation = 'relu'), #全连接网络，128个神经元，激活函数为relu
    tf.keras.layers.Dense(10, activation = 'softmax') #全连接网络，10个神经元，激活函数为softmax
])

#model.compile()
model.compile( #配置训练方式
    optimizer = 'adam', #使用adam优化器
    loss = tf.keras.losses.SparseCategoricalCrossentropy(from_logits = False), #配置损失函数，输出是概率分布
    metrics = ['sparse_categorical_accuracy'] #评测指标方法
)

checkpoint_save_path = './checkpoint/mnist.ckpt' #模型保存地址
if os.path.exists(checkpoint_save_path + '.index'): #判断模型是否存在
    print('---------------加载模型------------------') #如果模型存在就打印加载模型
    model.load_weights(checkpoint_save_path) #加载模型

#保存模型
cp_callback = tf.keras.callbacks.ModelCheckpoint( #保存模型
    filepath = checkpoint_save_path, #模型保存地址
    save_weights_only = True,
    save_best_only = True
)

histroy = model.fit( #执行训练过程
    x_train,
    y_train,
    batch_size = 32,
    epochs = 5,
    validation_data = (x_test, y_test),
    validation_freq = 1,
    callbacks = [cp_callback]
)

#model.summary
model.summary #打印参数

 

 结果为：(这里我的模型已经存在了，所以出现如下结果)

E:\Anaconda3\envs\TF2\python.exe C:/Users/Administrator/PycharmProjects/untitled8/保存模型.py
---------------加载模型------------------
Train on 60000 samples, validate on 10000 samples
Epoch 1/5

   32/60000 [..............................] - ETA: 55:02 - loss: 0.1279 - sparse_categorical_accuracy: 0.9688
  256/60000 [..............................] - ETA: 7:03 - loss: 0.0318 - sparse_categorical_accuracy: 0.9961 
  448/60000 [..............................] - ETA: 4:08 - loss: 0.0272 - sparse_categorical_accuracy: 0.9978
  640/60000 [..............................] - ETA: 2:58 - loss: 0.0310 - sparse_categorical_accuracy: 0.9969
  832/60000 [..............................] - ETA: 2:20 - loss: 0.0284 - sparse_categorical_accuracy: 0.9964
 1024/60000 [..............................] - ETA: 1:57 - loss: 0.0261 - sparse_categorical_accuracy: 0.9961
 

省略.......

59296/60000 [============================>.] - ETA: 0s - loss: 0.0130 - sparse_categorical_accuracy: 0.9962
59648/60000 [============================>.] - ETA: 0s - loss: 0.0130 - sparse_categorical_accuracy: 0.9962
60000/60000 [==============================] - 9s 156us/sample - loss: 0.0131 - sparse_categorical_accuracy: 0.9961 - val_loss: 0.0823 - val_sparse_categorical_accuracy: 0.9793