基于TensorFlow的MNIST手写数字识别-初级

一：MNIST数据集

下载地址

MNIST是一个包含很多手写数字图片的数据集，一共4个二进制压缩文件

基于TensorFlow的MNIST手写数字识别-初级-LMLPHP

分别是test set images，test set labels，training set images，training set labels

training set包括60000个样本，test set包括10000个样本。

test set中前5000个样本来自原始的NISTtraining set，后5000个样本来自原始的NIST test set，因此，前5000个样本比后5000个样本更简单和干净。

每个样本是28*28像素的图片

基于TensorFlow的MNIST手写数字识别-初级-LMLPHP

二：tensorflow构建模型识别MNIST

导入数据：

from tensorflow.examples.tutorials.mnist import input_data

mnist = input_data.read_data_sets('MNIST_data', one_hot=True)

import tensorflow as tf

x = tf.placeholder(tf.float32, shape=[None, 784])

y_ = tf.placeholder(tf.float32, shape=[None, 10])  #真实值

W = tf.Variable(tf.zeros([784, 10]))

b = tf.Variable(tf.zeros([10]))

y = tf.nn.softmax(tf.matmul(x, w) + b) #预测值

softmax的目的：将输出转化为是每个数字的概率

#计算交叉熵

cross_entropy = tf.reduce_mean(-tf.reduce_sum(y_label *tf.log(y), reduction_indices=[1]))

train = tf.train.GradientDescentOptimizer(0.5).minimize(cross_entropy)

交叉熵：衡量预测值与真实值之间的差别，当然是越小越好

公式为：

基于TensorFlow的MNIST手写数字识别-初级-LMLPHP

其中y'是真实值，y为预测值

最后用梯度下降法优化参数即可

在Session中运行graph：

total_steps = 5000

batch_size = 100

with tf.Session() as sess:

    sess.run(tf.global_variables_initializer())

    for step in range(total_steps+1):

        batch_x, batch_y = mnist.train.next_batch(batch_size)

        sess.run(train,feed_dict={x: batch_x, y_label: batch_y})

预测正确率：

correct_prediction = tf.equal(tf.argmax(y, axis=1), tf.argmax(y_label, axis=1))

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

tf.argmax()函数返回axis轴上最大值的index

tf.equal()函数返回的是布尔值，需要用tf.cast()方法转为tf.float32类型

最后在test set上进行预测：

step_per_test = 100

        if step % step_per_test == 0:

            print(step, sess.run(accuracy, feed_dict={x: mnist.test.images, y_label: mnist.test.labels}))

完整代码如下：

from tensorflow.examples.tutorials.mnist import input_data

import tensorflow as tf

mnist = input_data.read_data_sets('MNIST_data/', one_hot=True)

x = tf.placeholder(tf.float32, [None, 784])

y_label = tf.placeholder(tf.float32, [None, 10])

w = tf.Variable(tf.zeros([784, 10]))

b = tf.Variable(tf.zeros([10]))

y = tf.nn.softmax(tf.matmul(x, w) + b)

#计算交叉熵

cross_entropy = tf.reduce_mean(-tf.reduce_sum(y_label *tf.log(y), reduction_indices=[1]))

train = tf.train.GradientDescentOptimizer(0.5).minimize(cross_entropy)

#eval

correct_prediction = tf.equal(tf.argmax(y, axis=1), tf.argmax(y_label, axis=1))

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

total_steps = 5000

batch_size = 100

with tf.Session() as sess:

    sess.run(tf.global_variables_initializer())

    for step in range(total_steps+1):

        batch_x, batch_y = mnist.train.next_batch(batch_size)

        sess.run(train,feed_dict={x: batch_x, y_label: batch_y})

        step_per_test = 100

        if step % step_per_test == 0:

            print(step, sess.run(accuracy, feed_dict={x: mnist.test.images, y_label: mnist.test.labels}))

运行结果：

基于TensorFlow的MNIST手写数字识别-初级-LMLPHP

准确率为0.92左右

后面我们会构建更好的模型达到更高的正确率。