（中国大学MOOC）《深度学习应用开发-TensorFlow实践》（第8讲---MNIST手写数字识别：多层神经网络与应用）

全连接单隐藏层网络建模实现

%matplotlib notebook

import tensorflow as tf

import tensorflow.examples.tutorials.mnist.input_data as input_data
from time import time
import matplotlib.pyplot as plt
import numpy as np

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

x = tf.placeholder(tf.float32, [None, 784], name="X")
y = tf.placeholder(tf.float32, [None, 10], name="Y")

H1_NN = 256

W1 = tf.Variable(tf.truncated_normal([784, H1_NN], stddev=0.1))
b1 = tf.Variable(tf.zeros([H1_NN]))

Y1 = tf.nn.relu(tf.matmul(x, W1) + b1)

W2 = tf.Variable(tf.truncated_normal([H1_NN, 10], stddev=0.1))
b2 = tf.Variable(tf.zeros([10]))

forward = tf.matmul(Y1, W2) + b2
pred = tf.nn.softmax(forward)

loss_function = tf.reduce_mean(
                    tf.nn.softmax_cross_entropy_with_logits(logits=forward, labels=y))

train_epochs = 4
batch_size = 50
total_batch = int(mnist.train.num_examples / batch_size)
display_step = 1
learning_rate = 0.01

optimizer = tf.train.AdamOptimizer(learning_rate).minimize(loss_function)

correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(pred, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))

startTime = time()

sess = tf.Session()
sess.run(tf.global_variables_initializer())

loss_list = []

for epoch in range(train_epochs):
    for batch in range(total_batch):
        xs, ys = mnist.train.next_batch(batch_size)
        sess.run(optimizer, feed_dict={x:xs, y:ys})
    loss, acc = sess.run([loss_function, accuracy], 
                         feed_dict={x:mnist.validation.images,
                                    y:mnist.validation.labels})
    loss_list.append(loss)
    if(epoch+1) % display_step == 0:
        print("Train Epoch:", "%02d" % (epoch+1),
              "Loss=", "{:.9f}".format(loss), "Accuracy=", "{:.4f}".format(acc))

duration = time() - startTime
print("Train Finished takes:", "{:.2f}".format(duration))

plt.plot(loss_list)


acc_test = sess.run(accuracy, feed_dict={x:mnist.test.images,
                                                y:mnist.test.labels})
print("Test Accuarcy:", acc_test)

prediction_result = sess.run(tf.argmax(pred, 1), feed_dict={x:mnist.test.images})
print(prediction_result[0:10])

compare_list = prediction_result == np.argmax(mnist.test.labels, 1)
print(compare_list)

err_list = [i for i in range(len(compare_list)) if compare_list[i]==False]
print(err_list, len(err_list))

def print_predict_errs(labels, prediction):
    cnt = 0
    compare_list = prediction_result == np.argmax(mnist.test.labels, 1)
    err_list = [i for i in range(len(compare_list)) if compare_list[i]==False]
    for x in err_list:
        print("index="+str(x)+
              "标签值=", np.argmax(labels[x]),
              "预测值=",prediction_result[x])
        cnt += 1
    print("总计："+str(cnt))
    
print_predict_errs(labels=mnist.test.labels, prediction=prediction_result)

def plot_images_lables_prediction(images,
                                  labels,
                                  prediction,
                                  index,
                                  num=10):
    fig = plt.gcf()
    fig.set_size_inches(10, 15)
    if num > 25:
        num = 25
    for i in range(0, num):
        ax = plt.subplot(5, 5, i+1)
        ax.imshow(np.reshape(images[index],(28,28)), 
                  cmap="binary")
        title = "label=" + str(np.argmax(labels[index]))
        if len(prediction) > 0:
            title += ", predict=" + str(prediction[index])
        ax.set_title(title, fontsize=10)
        ax.set_xticks([])
        ax.set_yticks([])
        index += 1
    plt.show()
    
plot_images_lables_prediction(mnist.test.images,
                              mnist.test.labels,
                              prediction_result,
                              0, 10)

增加到两层神经网络

## 两层

%matplotlib notebook

import tensorflow as tf

import tensorflow.examples.tutorials.mnist.input_data as input_data
from time import time
import matplotlib.pyplot as plt
import numpy as np

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

x = tf.placeholder(tf.float32, [None, 784], name="X")
y = tf.placeholder(tf.float32, [None, 10], name="Y")

H1_NN = 256
H2_NN = 128

W1 = tf.Variable(tf.truncated_normal([784, H1_NN], stddev=0.1))
b1 = tf.Variable(tf.zeros([H1_NN]))

W2 = tf.Variable(tf.truncated_normal([H1_NN, H2_NN], stddev=0.1))
b2 = tf.Variable(tf.zeros([H2_NN]))

W3 = tf.Variable(tf.truncated_normal([H2_NN, 10], stddev=0.1))
b3 = tf.Variable(tf.zeros([10]))

Y1 = tf.nn.relu(tf.matmul(x, W1) + b1)
Y2 = tf.nn.relu(tf.matmul(Y1, W2) + b2)

forward = tf.matmul(Y2, W3) + b3
pred = tf.nn.softmax(forward)

loss_function = tf.reduce_mean(
                    tf.nn.softmax_cross_entropy_with_logits(logits=forward, labels=y))

train_epochs = 40
batch_size = 50
total_batch = int(mnist.train.num_examples / batch_size)
display_step = 1
learning_rate = 0.01

optimizer = tf.train.AdamOptimizer(learning_rate).minimize(loss_function)

correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(pred, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))

startTime = time()

sess = tf.Session()
sess.run(tf.global_variables_initializer())

loss_list = []

for epoch in range(train_epochs):
    for batch in range(total_batch):
        xs, ys = mnist.train.next_batch(batch_size)
        sess.run(optimizer, feed_dict={x:xs, y:ys})
    loss, acc = sess.run([loss_function, accuracy], 
                         feed_dict={x:mnist.validation.images,
                                    y:mnist.validation.labels})
    loss_list.append(loss)
    if(epoch+1) % display_step == 0:
        print("Train Epoch:", "%02d" % (epoch+1),
              "Loss=", "{:.9f}".format(loss), "Accuracy=", "{:.4f}".format(acc))

duration = time() - startTime
print("Train Finished takes:", "{:.2f}".format(duration))
plt.plot(loss_list)


acc_test = sess.run(accuracy, feed_dict={x:mnist.test.images,
                                                y:mnist.test.labels})
print("Test Accuarcy:", acc_test)

prediction_result = sess.run(tf.argmax(pred, 1), feed_dict={x:mnist.test.images})
print(prediction_result[0:10])

compare_list = prediction_result == np.argmax(mnist.test.labels, 1)
print(compare_list)

err_list = [i for i in range(len(compare_list)) if compare_list[i]==False]
print(err_list, len(err_list))

def plot_images_lables_prediction(images,
                                  labels,
                                  prediction,
                                  index,
                                  num=10):
    fig = plt.gcf()
    fig.set_size_inches(10, 15)
    if num > 25:
        num = 25
    for i in range(0, num):
        ax = plt.subplot(5, 5, i+1)
        ax.imshow(np.reshape(images[index],(28,28)), 
                  cmap="binary")
        title = "label=" + str(np.argmax(labels[index]))
        if len(prediction) > 0:
            title += ", predict=" + str(prediction[index])
        ax.set_title(title, fontsize=10)
        ax.set_xticks([])
        ax.set_yticks([])
        index += 1
    plt.show()
    
plot_images_lables_prediction(mnist.test.images,
                              mnist.test.labels,
                              prediction_result,
                              0, 10)

发现预测准确率反而下降了！！

保存训练好的模型

## 保存模型

%matplotlib notebook

import tensorflow as tf

import tensorflow.examples.tutorials.mnist.input_data as input_data
from time import time
import matplotlib.pyplot as plt
import numpy as np
import os

tf.reset_default_graph()
logDir = "C:\\Users\\20191027\\Documents\\log" # 输出日志，用于TensorBoard可视化


## 1、准备数据
mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)

x = tf.placeholder(tf.float32, [None, 784], name="X")
y = tf.placeholder(tf.float32, [None, 10], name="Y")

## 2、构建模型

H1_NN = 256

W1 = tf.Variable(tf.truncated_normal([784, H1_NN], stddev=0.1))
b1 = tf.Variable(tf.zeros([H1_NN]))

Y1 = tf.nn.relu(tf.matmul(x, W1) + b1)

W2 = tf.Variable(tf.truncated_normal([H1_NN, 10], stddev=0.1))
b2 = tf.Variable(tf.zeros([10]))

forward = tf.matmul(Y1, W2) + b2
pred = tf.nn.softmax(forward)

loss_function = tf.reduce_mean(
                    tf.nn.softmax_cross_entropy_with_logits(logits=forward, labels=y))

## 3、训练模型
train_epochs = 40
batch_size = 40
total_batch = int(mnist.train.num_examples / batch_size)
display_step = 1
learning_rate = 0.001

save_step = 5 # 保存模型粒度（每几轮保存一次）。只会保存最近的5次模型
ckpt_dir = "./ckpt_dir/" # 保存模型的路径
if not os.path.exists(ckpt_dir):
    os.makedirs(ckpt_dir)

optimizer = tf.train.AdamOptimizer(learning_rate).minimize(loss_function)

correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(pred, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))

# 声明完所有变量后，再调用Saver，用于保存模型
saver = tf.train.Saver()

startTime = time()

sess = tf.Session()
sess.run(tf.global_variables_initializer())

loss_list = [] # 统计训练每轮的损失
acc_list = [] # 统计训练每轮的准确率

for epoch in range(train_epochs):
    for batch in range(total_batch):
        xs, ys = mnist.train.next_batch(batch_size)
        sess.run(optimizer, feed_dict={x:xs, y:ys})
    loss, acc = sess.run([loss_function, accuracy], 
                         feed_dict={x:mnist.validation.images,
                                    y:mnist.validation.labels})
    loss_list.append(loss)
    acc_list.append(acc)
    
    if(epoch+1) % display_step == 0:
        print("Train Epoch:", "%02d" % (epoch+1),
              "Loss=", "{:.9f}".format(loss), "Accuracy=", "{:.4f}".format(acc))

    # 保存模型
    if (epoch+1) % save_step == 0:
        saver.save(sess, os.path.join(ckpt_dir,'mnist_h256_model_{:06d}.ckpt'.format(epoch+1)))
        print('mnist_h256_model_{:06d}.ckpt saved'.format(epoch+1))
        
# 保存最终的模型
saver.save(sess, os.path.join(ckpt_dir, 'mnist_h256_model.ckpt'))
print("Model saved!")
    
duration = time() - startTime
print("Train Finished takes:", "{:.2f}".format(duration))

plt.plot(loss_list) # 打印损失随训练轮数的变化曲线
plt.plot(acc_list) # 打印准确率随训练轮数的变化曲线

## 4.预测
acc_test = sess.run(accuracy, feed_dict={x:mnist.test.images,
                                                y:mnist.test.labels})
print("Test Accuarcy:", acc_test)

# 输出日志，用于TensorBoard可视化显示
writer = tf.summary.FileWriter(logDir, tf.get_default_graph())
writer.close()

用于tensorboard可视化计算图的日志文件:

用于模型恢复的文件:

恢复保存的模型

遇到了很多问题！！按照老师讲解的视频无法恢复。。。怎么破？

1. tensorboard无法显示计算图怎么办？
   Win+R，输入cmd进入DOS黑框框，输入tensorboard --logdir=日志路径
   （计算图可视化方法见这篇博客）
   
2. 奇葩的事情是，笔者搜索很久都没找到答案，搞了好久它自己好了。。StackOverflow

## 恢复保存的模型（有奇葩错误bug，需要先带上tf.reset_default_graph()执行，再去掉它执行。。）
# (奇葩的是，我搜索了那么久，都没找到解决方案，它自己好了。。。)

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

import numpy as np
import matplotlib.pyplot as plt
import cv2

tf.reset_default_graph() # 要有这句话
mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)

x = tf.placeholder(tf.float32, [None, 784], name="X")
y = tf.placeholder(tf.float32, [None, 10], name="Y")

H1_NN = 256

W1 = tf.Variable(tf.truncated_normal([784, H1_NN], stddev=0.1))
b1 = tf.Variable(tf.zeros([H1_NN]))

Y1 = tf.nn.relu(tf.matmul(x, W1) + b1)

W2 = tf.Variable(tf.truncated_normal([H1_NN, 10], stddev=0.1))
b2 = tf.Variable(tf.zeros([10]))

forward = tf.matmul(Y1, W2) + b2
pred = tf.nn.softmax(forward)

correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(pred, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))


ckpt_dir = "./ckpt_dir/" # 保存模型的路径

saver = tf.train.Saver()

sess = tf.Session()
sess.run(tf.global_variables_initializer())

ckpt = tf.train.get_checkpoint_state(ckpt_dir)

if ckpt and ckpt.model_checkpoint_path:
    saver.restore(sess, ckpt.model_checkpoint_path)
    print("Restore model from "+ckpt.model_checkpoint_path)

# 预测
print("Test Accuracy:", accuracy.eval(session=sess, 
                                      feed_dict={x:mnist.test.images, y:mnist.test.labels}))

特此说明

本文参考中国大学MOOC官方课程《深度学习应用开发-TensorFlow实践》吴明晖、李卓蓉、金苍宏