TensorFlow MNIST LeNet 模型持久化

前向传播过程mnist_inference.py

import tensorflow as tf

# 定义神经网络相关的参数
INPUT_NODE = 784
OUTPUT_NODE = 10


def inference(inputs, dropout_keep_prob):
    x_image = tf.reshape(inputs, [-1, 28, 28, 1])

    # 第一层：卷积层
    conv1_weights = tf.get_variable("conv1_weights", [5, 5, 1, 32], initializer=tf.truncated_normal_initializer(stddev=0.1))  # 过滤器大小为5*5, 当前层深度为1， 过滤器的深度为32
    conv1 = tf.nn.conv2d(x_image, filter=conv1_weights, strides=[1, 1, 1, 1], padding='SAME')  # 移动步长为1, 使用全0填充
    conv1_biases = tf.get_variable("conv1_biases", [32], initializer=tf.constant_initializer(0.0))
    relu1 = tf.nn.relu(tf.nn.bias_add(conv1, conv1_biases))  # 激活函数Relu去线性化

    # 第二层：最大池化层
    # 池化层过滤器的大小为2*2, 移动步长为2，使用全0填充
    pool1 = tf.nn.max_pool(relu1, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME') #输出14*14*32


    # 第三层：卷积层
    conv2_weights = tf.get_variable("conv2_weights", [5, 5, 32, 64], initializer=tf.truncated_normal_initializer(stddev=0.1))  # 过滤器大小为5*5, 当前层深度为32， 过滤器的深度为64
    conv2 = tf.nn.conv2d(pool1, conv2_weights, strides=[1, 1, 1, 1], padding='SAME')  # 移动步长为1, 使用全0填充
    conv2_biases = tf.get_variable("conv2_biases", [64], initializer=tf.constant_initializer(0.0))
    relu2 = tf.nn.relu(tf.nn.bias_add(conv2, conv2_biases))

    # 第四层：最大池化层
    # 池化层过滤器的大小为2*2, 移动步长为2，使用全0填充
    pool2 = tf.nn.max_pool(relu2, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME') #输出7*7*64

    # 第五层：全连接层
    pool2_vector = tf.reshape(pool2, [-1, 7 * 7 * 64])
    fc1_weights = tf.get_variable("fc1_weights", [7 * 7 * 64, 1024], initializer=tf.truncated_normal_initializer(stddev=0.1))  # 7*7*64=3136把前一层的输出变成特征向量
    fc1_baises = tf.get_variable("fc1_baises", [1024], initializer=tf.constant_initializer(0.1))
    fc1 = tf.nn.relu(tf.matmul(pool2_vector, fc1_weights) + fc1_baises)

    # 为了减少过拟合，加入Dropout层
    fc1_dropout = tf.nn.dropout(fc1, dropout_keep_prob)

    # 第六层：全连接层
    fc2_weights = tf.get_variable("fc2_weights", [1024, 10], initializer=tf.truncated_normal_initializer(stddev=0.1))  # 神经元节点数1024, 分类节点10
    fc2_biases = tf.get_variable("fc2_biases", [10], initializer=tf.constant_initializer(0.1))
    fc2 = tf.matmul(fc1_dropout, fc2_weights) + fc2_biases

    return fc2

训练mnist_train.py

import os
import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
import mnist_inference

#
BATCH_SIZE = 100
#学习率
LEARN_RATE = 0.001
MODEL_SAVE_PATH = "model/"
MODEL_NAME = "model.ckpt"
EPOCH = 2

def train(mnist):
    inputs = tf.placeholder(tf.float32, [None, mnist_inference.INPUT_NODE])
    labels = tf.placeholder(tf.float32, [None, mnist_inference.OUTPUT_NODE])
    dropout_keep_prob = tf.placeholder(tf.float32)

    logits = mnist_inference.inference(inputs, dropout_keep_prob)


    global_step = tf.Variable(0, trainable=False)

    cross_entropy = tf.nn.softmax_cross_entropy_with_logits(logits=logits, labels=labels)
    #tf.nn.sparse_softmax_cross_entropy_with_logits
    cost = tf.reduce_mean(cross_entropy)
    train_op = tf.train.AdamOptimizer(LEARN_RATE).minimize(cost, global_step=global_step)

    saver = tf.train.Saver()

    with tf.Session() as sess:
        tf.global_variables_initializer().run()
        print(mnist.train.images.shape)
        for i in range(20000):
            batch_inputs, batch_labels = mnist.train.next_batch(BATCH_SIZE)
            _, cost_value, step = sess.run([train_op, cost, global_step], feed_dict={inputs: batch_inputs, labels: batch_labels, dropout_keep_prob:0.5})

            if i % 1000 == 0:
                print("After %d training step(s), loss on training batch is %f." % (step, cost_value))
                saver.save(sess, os.path.join(MODEL_SAVE_PATH, MODEL_NAME), global_step=global_step)

def main(argv=None):
    mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)
    train(mnist)

if __name__ == '__main__':
    tf.app.run()

评估mnis_eval.py

import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
import mnist_inference
import mnist_train

def evaluate(mnist):
    inputs = tf.placeholder(tf.float32, [None, 784])
    labels = tf.placeholder(tf.float32, [None, 10])
    dropout_keep_prob = tf.placeholder(tf.float32)

    logits = mnist_inference.inference(inputs, dropout_keep_prob)
    print(logits)



    correct_prediction = tf.equal(tf.argmax(logits, 1), tf.argmax(labels, 1))

    accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))


    saver = tf.train.Saver()

    with tf.Session() as sess:
        ckpt = tf.train.get_checkpoint_state(mnist_train.MODEL_SAVE_PATH)
        if ckpt and ckpt.model_checkpoint_path:
            saver.restore(sess, ckpt.model_checkpoint_path)
            global_step = ckpt.model_checkpoint_path.split('/')[-1].split('-')[-1]
            accuracy_score = sess.run(accuracy, feed_dict={inputs: mnist.test.images, labels: mnist.test.labels, dropout_keep_prob:1.0})

            print("After %s training step(s), validation accuracy = %f" % (global_step, accuracy_score))
        else:
            print("No checkpoint file found")
            return



def main(argv=None):
    mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)
    evaluate(mnist)


if __name__ == '__main__':
    tf.app.run()