参考学习博客：
# https://www.cnblogs.com/felixwang2/p/9190692.html
一、模型保存
1 # https://www.cnblogs.com/felixwang2/p/9190692.html 2 # TensorFlow(十三)：模型的保存与载入 3 4 import tensorflow as tf 5 from tensorflow.examples.tutorials.mnist import input_data 6 7 # 载入数据集 8 mnist = input_data.read_data_sets("MNIST_data", one_hot=True) 9 10 # 每个批次100张照片 11 batch_size = 100 12 # 计算一共有多少个批次 13 n_batch = mnist.train.num_examples // batch_size 14 15 # 定义两个placeholder 16 x = tf.placeholder(tf.float32, [None, 784]) 17 y = tf.placeholder(tf.float32, [None, 10]) 18 19 # 创建一个简单的神经网络，输入层784个神经元，输出层10个神经元 20 W = tf.Variable(tf.zeros([784, 10])) 21 b = tf.Variable(tf.zeros([10])) 22 prediction = tf.nn.softmax(tf.matmul(x, W) + b) 23 24 # 二次代价函数 25 # loss = tf.reduce_mean(tf.square(y-prediction)) 26 loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(labels=y, logits=prediction)) 27 # 使用梯度下降法 28 train_step = tf.train.GradientDescentOptimizer(0.2).minimize(loss) 29 30 # 初始化变量 31 init = tf.global_variables_initializer() 32 33 # 结果存放在一个布尔型列表中 34 correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(prediction, 1)) # argmax返回一维张量中最大的值所在的位置 35 # 求准确率 36 accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32)) 37 38 saver = tf.train.Saver() 39 40 gpu_options = tf.GPUOptions(allow_growth=True) 41 with tf.Session(config=tf.ConfigProto(gpu_options=gpu_options)) as sess: 42 sess.run(init) 43 for epoch in range(11): 44 for batch in range(n_batch): 45 batch_xs, batch_ys = mnist.train.next_batch(batch_size) 46 sess.run(train_step, feed_dict={x: batch_xs, y: batch_ys}) 47 48 acc = sess.run(accuracy, feed_dict={x: mnist.test.images, y: mnist.test.labels}) 49 print("Iter " + str(epoch) + ",Testing Accuracy " + str(acc)) 50 # 保存模型 51 saver.save(sess, 'net/my_net.ckpt')
输出结果：
Iter 0,Testing Accuracy 0.8629 Iter 1,Testing Accuracy 0.896 Iter 2,Testing Accuracy 0.9028 Iter 3,Testing Accuracy 0.9052 Iter 4,Testing Accuracy 0.9085 Iter 5,Testing Accuracy 0.9099 Iter 6,Testing Accuracy 0.9122 Iter 7,Testing Accuracy 0.9139 Iter 8,Testing Accuracy 0.9148 Iter 9,Testing Accuracy 0.9163 Iter 10,Testing Accuracy 0.9165
二、模型载入
1 # https://www.cnblogs.com/felixwang2/p/9190692.html 2 # TensorFlow(十三)：模型的保存与载入 3 4 import tensorflow as tf 5 from tensorflow.examples.tutorials.mnist import input_data 6 7 # 载入数据集 8 mnist = input_data.read_data_sets("MNIST_data", one_hot=True) 9 10 # 每个批次100张照片 11 batch_size = 100 12 # 计算一共有多少批次 13 n_batch = mnist.train.num_examples // batch_size 14 15 # 定义两个placeholder 16 x = tf.placeholder(tf.float32, [None, 784]) 17 y = tf.placeholder(tf.float32, [None, 10]) 18 19 # 创建一个简单的神经网络，输入层784个神经单元，输出层10个神经单元 20 W = tf.Variable(tf.zeros([784, 10])) 21 b = tf.Variable(tf.zeros([10])) 22 prediction = tf.nn.softmax(tf.matmul(x, W) + b) 23 24 # 二次代价函数 25 # loss = tf.reduce_mean(tf.square(y-prediction)) 26 loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(labels=y, logits=prediction)) 27 # 使用梯度下降法 28 train_step = tf.train.GradientDescentOptimizer(0.2).minimize(loss) 29 30 # 初始化变量 31 init = tf.global_variables_initializer() 32 33 # 结果存放在一个布尔值列表中 34 correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(prediction, 1)) # argmax返回一维张量中最大的值所在的位置 35 # 求准确率 36 accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32)) 37 38 saver = tf.train.Saver() 39 40 gpu_options = tf.GPUOptions(allow_growth=True) 41 with tf.Session(config=tf.ConfigProto(gpu_options=gpu_options)) as sess: 42 sess.run(init) 43 # 未载入模型时的识别率 44 print('未载入识别率', sess.run(accuracy, feed_dict={x: mnist.test.images, y: mnist.test.labels})) 45 saver.restore(sess, 'net/my_net.ckpt') 46 # 载入模型后的识别率 47 print('载入后识别率', sess.run(accuracy, feed_dict={x: mnist.test.images, y: mnist.test.labels}))
未载入识别率 0.098 载入后识别率 0.9178
程序输出如上结果。