[tensorflow、神经网络] - 使用tf和mnist训练一个识别手写数字模型,并测试

• 参考

• 包含: 1.层级的计算、2.训练的整体流程、3.tensorboard画图、4.保存/使用模型、5.总体代码(含详细注释)

1. 层级的计算

如上图,mnist手写数字识别的训练集提供的图片是 28 * 28 * 1的手写图像,初始识别的时候,并不知道一次要训练多少个数据,因此输入的规模为 [None, 784]. 由于最终的标签输出的是10个数据,因此输出的规模为[None, 10], 中间采取一个简单的全连接层作为隐藏层,规模为[784, 10]

2. 训练的整体流程

• 1.首先定义占位符:

# 训练集数据
x = tf.placehodler(tf.float32,	[None, 784])
# 训练集标签
y_true = tf.placeholder(rf.int32, [None, 10])

• 2.建立模型

# 随机生成权重矩阵和偏置
# 权重
weight = tf.Variable(tf.random_normal([784, 10], mean =0.0, stddev=1.0), name="weight")
# 偏置
bias = tf.Variable(tf.constant(0.0, shape=[10]))
# 预测
y_predict = tf.matmul(x, weight) + bias

• 3.计算平均损失

loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y_true, logits=y_predict))

• 4.优化方案(梯度下降)

train_op = tf.train.GradientDescentOptimizer(0.1).minimize(loss)

• 5.计算损失率

equal_list = tf.equal(tf.argmax(y_true, 1), tf.argmax(y_predict, 1))
accuracy = tf.reduce_mean(tf.cast(equal_list, tf.float32))

3. tensorboard使用

# 按作用域命名
with tf.variable_scope("data"):
	pass

with tf.variable_scope("full_layer"):
	pass
# 收集变量(单维度)
tf.summary.scalar("losses", loss)
tf.summary.scalar("acc", accuracy)

# 收集变量(多维度)
tf.summary.histogram("weightes", weight)
tf.summary.histogram("biases", bias)

# 将训练的每一步写入
with tf.Session() as sess:
	# 建立events文件,然后写入
	filewriter = tf.summary.FileWriter("./tmp/", graph=sess.graph)
	for i in range(5000):
		# 写入每步训练的值
		summary = sess.run(merged, feed_dict={x: mnist_x, y_true: mnist_y})
		filewriter.add_summary(summary, i)

• 4.模型的保存/使用

# 模型的初始化(一般写在Session上面)
saver = tf.train.Saver()

# Session中为模型保存分配资源
with tf.Session() as sess:
	# 保存模型
	saver.save(sess, "./tmp/ckpt/fc_model")
	
	# 加载模型
	saver.restore(sess, "./tmp/ckpt/fc_model")
	# 预测
	for i in range(100):
		x, y = mnist.test.next_batch(1)
		predict = tf.argmax(sess.run(y_predict, feed_dict={x: x_test, y_true: y_test}), 1).eval()

5.总体代码

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

FLAGS = tf.app.flags.FLAGS

tf.app.flags.DEFINE_integer("is_train", 1, "0: 预测, 1: 训练")

"""
    单层(全连接层)实现手写数字识别
      特征值[None, 784]  目标值[None, 10]
    
    1、 定义数据占位符
      特征值[None, 784]  目标值[None, 10]
      
    2、 建立模型
      随机初始化权重和偏置
      w[784, 10]  b
      y_predict = tf.matmul(x, w) + b
    
    3、 计算损失
      loss: 平均样本的损失  
    
    4、 梯度下降优化
    
    
    5、 准确率计算:
        equal_list = tf.equal(tf.argmax(y, 1), tf.argmax(y_label, 1))    
        accuracy = tf.reduce_mean(tf.cast(equal_list, tf.float32))
"""


def ful_connected():
    # 读取数据
    mnist = input_data.read_data_sets("./data/mnist/input_data/", one_hot=True)

    # 1、 建立数据的占位符 x [None, 784]  y_true  [None, 10]
    with tf.variable_scope("data"):
        x = tf.placeholder(tf.float32, [None, 784])

        y_true = tf.placeholder(tf.int32, [None, 10])

    # 2、 建立一个全连接层的神经网络 w [784, 10]  b [10]
    with tf.variable_scope("full_layer"):
        # 随机初始化权重和偏置
        weight = tf.Variable(tf.random_normal([784, 10], mean=0.0, stddev=1.0), name="weight")

        bias = tf.Variable(tf.constant(0.0, shape=[10]))

        # 预测None个样本的输出结果  [None, 784] * [784, 10]  + [10] = [None, 10]
        y_predict = tf.matmul(x, weight) + bias

    # 3、 求出所有样本的损失,然后求平均值
    with tf.variable_scope("softmax"):
        # 求平均交叉熵损失
        loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y_true, logits=y_predict))

    # 4、 梯度下降求出损失
    with tf.variable_scope("optimizer"):
        train_op = tf.train.GradientDescentOptimizer(0.1).minimize(loss)

    # 5、 计算准确率
    with tf.variable_scope("count_acc"):
        equal_list = tf.equal(tf.argmax(y_true, 1), tf.argmax(y_predict, 1))

        # equal_list  None个样本  [1, 0, 1, 0, 1, 1, ....]
        accuracy = tf.reduce_mean(tf.cast(equal_list, tf.float32))

    # 收集变量(单维度)
    tf.summary.scalar("losses", loss)
    tf.summary.scalar("acc", accuracy)
    # 收集变量(高维度)
    tf.summary.histogram("weightes", weight)
    tf.summary.histogram("biases", bias)

    # 定义一个初始化变量的op
    init_op = tf.global_variables_initializer()
    # 定义合并变量
    merged = tf.summary.merge_all()

    # 保存模型
    saver = tf.train.Saver()

    # 开启会话训练
    with tf.Session() as sess:
        # 初始化变量
        sess.run(init_op)

        # 建立events文件,然后写入
        filewriter = tf.summary.FileWriter("./tmp/", graph=sess.graph)

        if FLAGS.is_train == 0:

            # 迭代步骤去训练,更新参数预测
            for i in range(5000):
                # 取出真实存在的特征值 和 目标值
                mnist_x, mnist_y = mnist.train.next_batch(50)

                # 运行train_op训练
                sess.run(train_op, feed_dict={x: mnist_x, y_true: mnist_y})

                # 写入每步训练的值
                summary = sess.run(merged, feed_dict={x: mnist_x, y_true: mnist_y})

                filewriter.add_summary(summary, i)

                # 打印损失
                print("训练第%d步,准确率为:%f" % (i, sess.run(accuracy, feed_dict={x: mnist_x, y_true: mnist_y})))

            # 保存模型
            saver.save(sess, "./tmp/ckpt/fc_model")
        else:
            # 加载模型
            saver.restore(sess, "./tmp/ckpt/fc_model")

            # 预测
            for i in range(100):
                # 每次测试一张图片
                x_test, y_test = mnist.test.next_batch(1)

                print("第%d张图片是: %d,预测结果是:%d" % (
                    i,
                    tf.argmax(y_test, 1).eval(),
                    tf.argmax(sess.run(y_predict, feed_dict={x: x_test, y_true: y_test}), 1).eval()
                ))

    return None


if __name__ == "__main__":
    ful_connected()

6. cnn版本的mnist

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

"""
    使用卷积神经网络实现 mnist的手写数据集识别
"""

"""
    input: [None, 784]
    output: [784, 10]
    
    进入卷积时,首先需要改变图片的形状 [None, 784] --> [None, 28, 28, 1]
    卷积网络设计:
        · 第一层卷积层: 32 * core(5*5)、strides(1)、padding="SAME"
            · 此时大小为: [None, 28, 28, 32]
            · 激活
            · 池化: 2*2、 strides(2)、 padding="SAME"
                · 此时大小为: [None, 14, 14, 32]
        · 第二层卷积层: 64 * core(5*5)、 strides(1)、 padding="SAME"
            · 此时大小为: [None, 14, 14, 64]
            · 激活
            · 池化: 2*2、 strides(2)、 padding="SAME"
                · 此时大小为: [None, 7, 7, 64]
        · 全连接层: [None, 7*7*64] * [7*7*64, 10] + bias = [None, 10]
"""


# 定义个初始化权重的函数
def weight_variable(shape):
    w = tf.Variable(tf.random_normal(shape=shape, mean=0.0, stddev=1.0))
    return w


# 定义一个初始化偏置的函数
def bias_variables(shape):
    b = tf.Variable(tf.constant(0.0, shape=shape))
    return b


def model():
    """
    自定义的卷积模型
    :return:
    """
    # 1、准备数据的占位符 x [None, 784] 、 y_true [None, 10]
    with tf.variable_scope("data"):
        x = tf.placeholder(tf.float32, [None, 784])
        y_true = tf.placeholder(tf.int32, [None, 10])

    # 2、一卷积层 卷积: 5*5*1, 32个, strides = 1 、激活、池化
    with tf.variable_scope("conv1"):
        # 随机初始化权重,偏置[32]
        w_conv1 = weight_variable([5, 5, 1, 32])
        b_conv1 = bias_variables([32])

        # 对x进行形状的改变 [None, 784] -> [None, 28, 28, 1]
        x_reshape = tf.reshape(x, [-1, 28, 28, 1])

        # [None, 28, 28, 1] -> [None, 28, 28, 32]
        x_relu1 = tf.nn.relu(tf.nn.conv2d(x_reshape, w_conv1, strides=[1, 1, 1, 1], padding="SAME") + b_conv1)

        # 池化 2*2, strides2 [None, 28, 28, 32] -> [None, 14, 14, 32]
        x_pool1 = tf.nn.max_pool(x_relu1, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding="SAME")

    # 3、二卷积层  5*5*32, 64个filter, strides= 1
    with tf.variable_scope("conv2"):
        w_conv2 = weight_variable([5, 5, 32, 64])
        b_conv2 = bias_variables([64])

        # 卷积、激活、池化计算
        # [None, 14, 14, 32] -> [None, 14, 14, 64]
        x_relu2 = tf.nn.relu(tf.nn.conv2d(x_pool1, w_conv2, strides=[1, 1, 1, 1], padding="SAME") + b_conv2)

        # 池化 2*2, strides2 [None, 14, 14, 64] -> [None, 7, 7, 64]
        x_pool2 = tf.nn.max_pool(x_relu2, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding="SAME")

        # 4、全连接层 [None, 7, 7, 64] --> [None, 7*7*64] * [7*7*64, 10] + [10] = [None, 10]

        # 随机初始化权重和偏置
        w_fc = weight_variable([7 * 7 * 64, 10])
        b_fc = bias_variables([10])

        # 修改形状: [None, 7, 7, 64] -> [None, 7*7*64]
        x_fc_reshape = tf.reshape(x_pool2, [-1, 7 * 7 * 64])

        # 矩阵运算,得出每个样本的10个结果
        y_predict = tf.matmul(x_fc_reshape, w_fc) + b_fc

    return x, y_true, y_predict


def conf_fc():
    # 1、 读取数据
    mnist = input_data.read_data_sets("./data/mnist/input_data/", one_hot=True)

    # 2、 定义模型,得出输出
    x, y_true, y_predict = model()

    # 3、 求出所有的损失,然后求平均值
    with tf.variable_scope("soft_cross"):
        # 求平均交叉熵损失
        loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y_true, logits=y_predict))

    # 4、 梯度下降求出损失
    with tf.variable_scope("optimizer"):
        train_op = tf.train.GradientDescentOptimizer(0.00005).minimize(loss)

    # 5、 计算准确率
    with tf.variable_scope("acc"):
        equal_list = tf.equal(tf.argmax(y_true, 1), tf.argmax(y_predict, 1))

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

    # 定义一个初始变量op
    init_op = tf.global_variables_initializer()

    # 开启会话运行
    with tf.Session() as sess:
        sess.run(init_op)

        # 循环去训练
        for i in range(1000):
            # 取出真实存在的特征值和目标值
            mnist_x, mnist_y = mnist.train.next_batch(50)

            # 运行train_op训练
            sess.run(train_op, feed_dict={x: mnist_x, y_true: mnist_y})

            # 打印损失
            print("训练第%d步,准确率为:%f" % (i, sess.run(accuracy, feed_dict={x: mnist_x, y_true: mnist_y})))

    return None


if __name__ == "__main__":
    conf_fc()