构建一个神经网络对mnist手写数据集进行识别

构建一个神经网络对mnist手写数据集进行识别

代码

import tensorflow as tf
import numpy as np
import tensorflow.examples.tutorials.mnist.input_data as input_data
# 先导入mnist数据集
# 以one_hot=True的方式读取图片，那么标签值就是长度为10的向量，表示0～9的概率
mnist = input_data.read_data_sets("E:\pycharm\work\cookbook\项目\data\mnist", one_hot=True)


def add_layer1(input):    # 第一层神经网络
    W1 = tf.Variable(tf.random_normal([784,512]))        # 权重值
    b1 = tf.Variable(tf.random_normal([512]))           # 偏置值
    Wx_plus_b_1 = tf.matmul(input,W1) + b1         # 计算输入值
    outputs = tf.nn.sigmoid(Wx_plus_b_1)   # 激活函数sigmoid激活
    return outputs


def add_layer2(input):    # 第二层神经网络
    W2 = tf.Variable(tf.random_normal([512, 256]))        # 权重值
    b2 = tf.Variable(tf.random_normal([256]))           # 偏置值
    Wx_plus_b_2 = tf.matmul(input,W2) + b2        # 计算输入值
    outputs= tf.nn.sigmoid(Wx_plus_b_2)   # 激活函数sigmoid激活
    return outputs


def add_layer3(input):    # 第三层神经网络
    W2 = tf.Variable(tf.random_normal([256, 10]))        # 权重值
    b2 = tf.Variable(tf.random_normal([10]))           # 偏置值
    Wx_plus_b_2 = tf.matmul(input,W2) + b2        # 计算输入值
    outputs = tf.nn.softmax(Wx_plus_b_2)   # 激活函数softmax激活
    return outputs

#占位
x = tf.placeholder(tf.float32,shape=[None,784])
y_actual = tf.placeholder(tf.float32,shape=[None,10])
y_1 = add_layer1(x)
y_2 = add_layer2(y_1)
y_predict = add_layer3(y_2)
loss = tf.reduce_mean(-tf.reduce_sum(y_actual*tf.log(y_predict),reduction_indices=1))   # 损失函数 交叉熵函数
train_step = tf.train.GradientDescentOptimizer(0.2).minimize(loss)     # tensorflow优化器 实现梯度下降算法   降低损失值 学习率0.2
correct_prediction = tf.equal(tf.argmax(y_predict, 1), tf.argmax(y_actual, 1))    # 正确率
accuracy = tf.reduce_mean(tf.cast(correct_prediction,'float'))         # 算平均正确率
init = tf.global_variables_initializer()
sess = tf.Session()                                               # 创建会话
sess.run(init)
for i in range(20000):                      # 循环
    batch_xs, batch_ys=mnist.train.next_batch(200)             # 每批喂入200个
    sess.run(train_step,feed_dict = {x:batch_xs,y_actual:batch_ys})            # 这里喂入数据
    if(i%200 == 0):                                                           # 两百个一次循环    输出正确率
        print("accuracy:",sess.run(accuracy,feed_dict={x:mnist.test.images, y_actual:mnist.test.labels}))
sess.close()

运行结果