MNIST手写数字识别问题的多层神经网络模型实践
（10分）
MNIST手写数字识别问题的多层神经网络模型实践
按课程案例，动手完成编码实践。
自行设计一种神经网络模型，并尝试采用不同超参数，让模型的准确率达到97.5%。
提交要求：
1、你认为最优的一次带运行结果的源代码文件（.ipynb 格式）
2、作为附件上传
评分标准：
1、完成MNIST手写数字识别的神经网络建模与应用，有完整的代码，模型能运行，准确率达97%以上；得6分；每下降一个百分点，少得1分；
2、准确率达97.3%以上；再得2分，否则再得0分；
3、准确率到97.5%以上；再得2分，否则再得0分；
源代码：
import tensorflow.compat.v1 as tf
import matplotlib.pyplot as plt #图像的显示
import numpy as np
from tensorflow.examples.tutorials.mnist import input_data
from time import time
tf.disable_eager_execution()
#读取mnist数据
mnist = input_data.read_data_sets(r"D:\编程代码\python程序\mnist_data",one_hot = True) #r表示原始字符串
#构建输入层
x = tf.placeholder(tf.float32,[None,784],name="X") #每张图片28*28个像素点
y = tf.placeholder(tf.float32,[None,10],name="Y")
#构建隐藏层
H1_NN = 256 #第1隐藏层 神经元为256个
H2_NN = 64 #第2隐藏层神经元为64个
H3_NN = 32 #第3隐藏层神经元为32个
#定义全连接层函数
def fcn_layer(inputs,input_dim,output_dim,activation=None):
"""
inputs:输入数据
input_dim:输入神经元数量
output_dim:输出神经元数量
activation:激活函数
"""
W = tf.Variable(tf.truncated_normal([input_dim,output_dim],stddev=0.1))
b = tf.Variable(tf.zeros([output_dim]))
XWb = tf.matmul(inputs,W) + b #矩阵相乘
outputs = XWb if (activation is None) else activation(XWb)
return outputs
h1 = fcn_layer(inputs=x,input_dim=784,output_dim=H1_NN,activation=tf.nn.relu)
h2 = fcn_layer(inputs=h1,input_dim=H1_NN,output_dim=H2_NN,activation=tf.nn.relu)
h3 = fcn_layer(inputs=h2,input_dim=H2_NN,output_dim=H3_NN,activation=tf.nn.relu)
forward = fcn_layer(inputs=h3,input_dim=H3_NN,output_dim=10,activation=None)
pred = tf.nn.softmax(forward) #Softmax 分类
#设置训练参数
train_epochs = 40 #训练轮数
batch_size = 50 #单次训练样本数
total_batch = int(mnist.train.num_examples/batch_size) #一轮训练有多少批次
display_step = 1 #显示粒度
learning_rate = 0.008 #学习率
#定义损失函数
loss_function = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=forward,labels=y))
#选择AdamOptimizer优化器
optimizer = tf.train.AdamOptimizer(learning_rate).minimize(loss_function)
#定义准确率
correct_prediction = tf.equal(tf.argmax(y,1),tf.argmax(pred,1)) #argmax将最大标签取出来
#准确率,将布尔型转换为浮点型，并计算平均值
accuracy = tf.reduce_mean(tf.cast(correct_prediction,tf.float32))
startTime = time()
sess = tf.Session() #声明回话
init = tf.global_variables_initializer() #变量的初始化
sess.run(init)
#开始训练
for epoch in range(train_epochs):
for batch in range(total_batch):
xs,ys = mnist.train.next_batch(batch_size) #读取批次数据
sess.run(optimizer,feed_dict={x:xs,y:ys}) #执行批次训练
#使用验证集计算误差和准确率
loss,acc = sess.run([loss_function,accuracy],feed_dict={x:mnist.validation.images,y:mnist.validation.labels})
#打印训练中的详细信息
if (epoch+1) % display_step == 0:
print("Train Epoch:{:02d} Loss={:.9f} Accuracy={:.4f}".format(epoch+1,loss,acc))
print("Train Finished！")
#显示运行时间
duration = time() - startTime
print("Train Finished takes:{:.2f}".format(duration))
print("\n\n")
#完成训练后,在测试集上评估模型的准确率
accu_test = sess.run(accuracy,feed_dict={x:mnist.test.images,y:mnist.test.labels})
print("Test Accuracy：",accu_test)
#完成训练后,在验证集上评估模型的准确率
accu_validation = sess.run(accuracy,feed_dict={x:mnist.validation.images,y:mnist.validation.labels})
print("Validation Accuracy：",accu_validation)
#完成训练后,在测试集上评估模型的准确率
accu_train = sess.run(accuracy,feed_dict={x:mnist.train.images,y:mnist.train.labels})
print("Train Accuracy：",accu_train)
print("\n\n")
#预测结果
prediction_result = sess.run(tf.argmax(pred,1),feed_dict={x:mnist.test.images})
#找出预测错误
compare_lists = prediction_result==np.argmax(mnist.test.labels,1)
#列表推导式
err_list = [i for i in range(len(compare_lists)) if compare_lists[i] ==False]
#定义可视化函数
def plot_images_labels_prediction(images,labels,prediction,num=10):
#函数文档
"""
images:图像列表
labels:标签列表
prediction:预测值列表
index:从第index个开始显示
num=10：缺省依次显示10幅
"""
j=-1
fig = plt.gcf() #获取当前图标
fig.set_size_inches(10,12)
if num > 25:
num = 25 #最多显示25个子图
for i in range(num):
j += 1
index = err_list[j]
ax = plt.subplot(5,5,i+1) #获取当前需要处理的子图
#显示第index个图像
ax.imshow(np.reshape(images[index],(28,28)),cmap="binary")
title = "label=" + str(np.argmax(labels[index])) #标题
if len(prediction) > 0:
title += (",predict=" + str(prediction[index]))
ax.set_title(title,fontsize=10) #显示标题信息字体大小10号
ax.set_xticks([]) #不显示坐标轴
ax.set_yticks([])
plt.show()
print("可视化查看错误的样本：")
#显示15幅错误预测值与标签值对比图
plot_images_labels_prediction(mnist.test.images,mnist.test.labels,prediction_result,15)
输出结果：
Extracting D:\编程代码\python程序\mnist_data\train-images-idx3-ubyte.gz
Extracting D:\编程代码\python程序\mnist_data\train-labels-idx1-ubyte.gz
Extracting D:\编程代码\python程序\mnist_data\t10k-images-idx3-ubyte.gz
Extracting D:\编程代码\python程序\mnist_data\t10k-labels-idx1-ubyte.gz
Train Epoch:01 Loss=0.154160351 Accuracy=0.9578
Train Epoch:02 Loss=0.170846000 Accuracy=0.9598
Train Epoch:03 Loss=0.124862306 Accuracy=0.9692
Train Epoch:04 Loss=0.140269428 Accuracy=0.9720
Train Epoch:05 Loss=0.121404067 Accuracy=0.9696
Train Epoch:06 Loss=0.132848248 Accuracy=0.9702
Train Epoch:07 Loss=0.144610047 Accuracy=0.9708
Train Epoch:08 Loss=0.143921122 Accuracy=0.9728
Train Epoch:09 Loss=0.132440180 Accuracy=0.9728
Train Epoch:10 Loss=0.143663749 Accuracy=0.9738
Train Epoch:11 Loss=0.133612305 Accuracy=0.9756
Train Epoch:12 Loss=0.149417579 Accuracy=0.9718
Train Epoch:13 Loss=0.157808796 Accuracy=0.9738
Train Epoch:14 Loss=0.167589828 Accuracy=0.9746
Train Epoch:15 Loss=0.158174515 Accuracy=0.9724
Train Epoch:16 Loss=0.136991411 Accuracy=0.9760
Train Epoch:17 Loss=0.150212899 Accuracy=0.9784
Train Epoch:18 Loss=0.170966819 Accuracy=0.9768
Train Epoch:19 Loss=0.414516211 Accuracy=0.9698
Train Epoch:20 Loss=0.205433667 Accuracy=0.9722
Train Epoch:21 Loss=0.157262698 Accuracy=0.9784
Train Epoch:22 Loss=0.186451778 Accuracy=0.9730
Train Epoch:23 Loss=0.194627523 Accuracy=0.9726
Train Epoch:24 Loss=0.143057838 Accuracy=0.9764
Train Epoch:25 Loss=0.222550884 Accuracy=0.9750
Train Epoch:26 Loss=0.172464162 Accuracy=0.9766
Train Epoch:27 Loss=0.191078886 Accuracy=0.9756
Train Epoch:28 Loss=0.175143555 Accuracy=0.9778
Train Epoch:29 Loss=0.239344627 Accuracy=0.9736
Train Epoch:30 Loss=0.196491301 Accuracy=0.9744
Train Epoch:31 Loss=0.160425201 Accuracy=0.9758
Train Epoch:32 Loss=0.229437336 Accuracy=0.9792
Train Epoch:33 Loss=0.224512666 Accuracy=0.9766
Train Epoch:34 Loss=0.229335472 Accuracy=0.9776
Train Epoch:35 Loss=0.274593800 Accuracy=0.9712
Train Epoch:36 Loss=0.193944454 Accuracy=0.9772
Train Epoch:37 Loss=0.268351167 Accuracy=0.9706
Train Epoch:38 Loss=0.258617491 Accuracy=0.9780
Train Epoch:39 Loss=0.270832181 Accuracy=0.9766
Train Epoch:40 Loss=0.263729692 Accuracy=0.9768
Train Finished！
Train Finished takes:179.20
作者：Darlen、阳仔
