深度学习实例之基于mnist的手写数字识别
本文主要是介绍基于mnist数据集的手写数字识别.
一 数据集
mnist 数据集：包含 7 万张黑底白字手写数字图片， 其中 55000 张为训练集，5000 张为验证集， 10000 张为测试集。每张图片大小为 28*28 像素，图片中纯黑色像素值为 0， 纯白色像素值为 1。数据集的标签是长度为 10 的一维数组，数组中每个元素索引号表示对应数字出现的概率。在将 mnist 数据集作为输入喂入神经网络时，需先将数据集中每张图片变为长度784 一维数组，将该数组作为神经网络输入特征喂入神经网络。
1. 使用tensorflow提供的数据集mnist,具体的加载方法为:
from tensorflow.examples.tutorials.mnist import input_datamnist = input_data(data_path,one_hot=True)2. 数据集分为train,validation,test三个数据集.
① 返回数据集train样本数 mnist.train.num_examples
② 返回数据集validation样本数 mnist.validation.num_examples
③ 返回数据集test样本数 mnist.test.num_examples
3. 使用mnist.train.images返回train数据集中的所有图片的像素值
4. 使用mnist.train.labels返回train数据集中的所有图片的标签
5. 使用mnist.train.next_batch()将数据输入神经网络
二 前向计算(得到预测值)
废话不说了,直接看代码.(mnist_forward.py)
# _*_coding:utf-8_*_
import tensorflow as tf
input_node = 784
output_node = 10
layer1_node = 500
def get_weight(shape, regularizer):
# 表示要求产生的数据服从正态分布, 并且每个值与均值之间的差值均小于两倍的标准差
w = tf.Variable(tf.truncated_normal(shape, stddev=0.1))
# 判断是否进行正则化操作
if regularizer is not None:
tf.add_to_collection('losses', tf.contrib.layers.l2_regularizer(regularizer)(w))
return w
def get_bias(shape):
b = tf.Variable(tf.zeros(shape))
return b
def forward(x, regulaizer):
# 输入层到Layer1层
w1 = get_weight(shape=[input_node, layer1_node], regularizer=regulaizer)
b1 = get_bias(shape=[layer1_node])
y1 = tf.nn.relu(tf.add(tf.matmul(x, w1), b1))
# 从Layer1层到输出层
w2 = get_weight(shape=[layer1_node, output_node], regularizer=regulaizer)
b2 = get_bias(shape=[output_node])
y = tf.add(tf.matmul(y1, w2), b2)
return y三 反向计算(参数更新)
( mnist_backward.py)# _*_coding:utf-8_*_
import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
import mnist_forward
import os
batch_size = 200 # 每轮训练的图片数量
learning_rate_base = 0.1 # 初始学习率
learning_rate_decay = 0.99 # 学习率衰减率
regularizer = 0.0001 # 正则化系数
total_steps = 50000 # 训练轮数
moving_average_decay = 0.99
model_save_path = './model/'
model_name = 'mnist_model'
def backward(mnist):
# 设置x,y
x = tf.placeholder(tf.float32, [None, mnist_forward.input_node])
y = tf.placeholder(tf.float32, [None, mnist_forward.output_node])
y_hat = mnist_forward.forward(x, regularizer) # 获取forward的返回值
global_step = tf.Variable(0, trainable=False) # 当前轮数值初始化
# 损失函数(softmax和交叉熵共同组成的loss,在加上正则化损失的总和)
ce = tf.nn.sparse_softmax_cross_entropy_with_logits(labels=tf.argmax(y, 1), logits=y_hat)
cem = tf.reduce_mean(ce) # 求所有元素的均值
loss = cem + tf.add_n(tf.get_collection('losses')) # 得到包含所有参数损失的损失函数
# 学习率梯度衰减的模型
'''
decayed_learning_rate = learning_rate *decay_rate ^ (global_step / decay_steps)
'''
learning_rate = tf.train.exponential_decay(learning_rate_base, global_step, mnist.train.num_examples / batch_size,
learning_rate_decay, staircase=True)
# 定义参数优化方法
train_step = tf.train.GradientDescentOptimizer(learning_rate).minimize(loss, global_step)
# 定义参数的滑动平均
ema = tf.train.ExponentialMovingAverage(moving_average_decay, global_step)
ema_op = ema.apply(tf.trainable_variables())
# 定义参数的控制依赖
with tf.control_dependencies([train_step, ema_op]):
train_op = tf.no_op(name='train')
# 创建模型保存的实例化对象
saver = tf.train.Saver()
with tf.Session() as sess:
# 参数初始化
sess.run(tf.global_variables_initializer())
ckpt = tf.train.get_checkpoint_state(model_save_path)
# 判断是否有ckpt模型,有则恢复模型(这种方法简单便捷,不进行重复的训练)
if ckpt and ckpt.model_checkpoint_path:
# 恢复会话,继续训练模型
saver.restore(sess, ckpt.model_checkpoint_path)
# 模型训练,迭代
for i in range(total_steps):
xs, ys = mnist.train.next_batch(batch_size)
_, loss_value, step = sess.run([train_op, loss, global_step], feed_dict={x: xs, y: ys})
if i % 100 == 0:
print('After {} train step(s), loss on training batch is {}'.format(step, loss_value))
# 将当前会话加载到指定路径
saver.save(sess, os.path.join(model_save_path, model_name), global_step=global_step)
def main():
mnist = input_data.read_data_sets('./mnist', one_hot=True)
backward(mnist)
if __name__ == '__main__':
main()四 测试代码
( mnist_test.py)# _*_coding:utf-8_*_
import tensorflow as tf
import time
from tensorflow.examples.tutorials.mnist import input_data
import mnist_forward
import mnist_backward
sleep_time = 5
def mnist_test(mnist):
with tf.Graph().as_default() as g:
x = tf.placeholder(tf.float32, shape=[None, mnist_forward.input_node], name='x')
y = tf.placeholder(tf.float32, shape=[None, mnist_forward.output_node], name='y')
y_hat = mnist_forward.forward(x, None)
ema = tf.train.ExponentialMovingAverage(mnist_backward.moving_average_decay)
ema_restore = ema.variables_to_restore()
# 创建模型保存的实例化对象
saver = tf.train.Saver(ema_restore)
# 模型的准确率计算
correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_hat, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
# 循环计算
while True:
with tf.Session() as sess:
ckpt = tf.train.get_checkpoint_state(mnist_backward.model_save_path)
# 判断是否有ckpt模型,有则恢复模型
if ckpt and ckpt.model_checkpoint_path:
# 恢复会话
saver.restore(sess, ckpt.model_checkpoint_path)
global_step = ckpt.model_checkpoint_path.split('/')[-1].split('-')[-1]
accuracy_score = sess.run(accuracy, feed_dict={x: mnist.test.images, y: mnist.test.labels})
print('After {} training steps, test accuracy is {}'.format(global_step, accuracy_score))
else:
print('No checkpoint file found')
return
time.sleep(sleep_time)
def main():
mnist = input_data.read_data_sets('./mnist', one_hot=True)
mnist_test(mnist)
if __name__ == '__main__':
main()五 小结
笔者觉得文章代码中的一个亮点(较为实用,不需要每次运行代码都从头开始,节约时间!!!):
# 判断是否有ckpt模型,有则恢复模型(这种方法简单便捷,不进行重复的训练)
if ckpt and ckpt.model_checkpoint_path:
# 恢复会话,继续训练模型
saver.restore(sess, ckpt.model_checkpoint_path)