一、softmax
参考博文'Tensorflow 实现 MNIST 手写数字识别'http://blog.csdn.net/u010858605/article/details/69830657
# coding=utf-8
#!/usr/bin/env python
import tensorflow as tf
# 获取数据,MNIST数据集包含55000样本的训练集,5000样本的验证集,10000样本的测试集
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("input_data", one_hot=True)
# 显示图像和类标的形状
print('训练集信息:')
print(mnist.train.images.shape,mnist.train.labels.shape)
print('测试集信息:')
print(mnist.test.images.shape,mnist.test.labels.shape)
print('验证集信息:')
print(mnist.validation.images.shape,mnist.validation.labels.shape)
# 实现模型 y=softmax(wx+b)
# placeholder:输入数据的地方,None 代表不限条数的输入,每条是784维的向量
# Variable:存储模型参数,持久化的
sess = tf.InteractiveSession()
x = tf.placeholder(tf.float32, [None, 784])
W = tf.Variable(tf.zeros([784,10]))
b = tf.Variable(tf.zeros([10]))
y = tf.nn.softmax(tf.matmul(x,W) + b)
# 定义一个交叉熵作为loss函数cross_entropy,其中y是我们预测的概率分布, y_是实际的分布
y_ = tf.placeholder(tf.float32, [None,10])
cross_entropy = tf.reduce_mean(-tf.reduce_sum(y_ * tf.log(y),reduction_indices=[1]))
# 采用随机梯度下降法,步长为0.5进行训练
train_step = tf.train.GradientDescentOptimizer(0.5).minimize(cross_entropy)
# 让模型循环训练1000次,每次随机train100条样本
tf.global_variables_initializer().run()
for i in range(1000):
batch_xs, batch_ys = mnist.train.next_batch(100)
train_step.run({x: batch_xs, y_: batch_ys})
# 模型评估
correct_prediction = tf.equal(tf.argmax(y,1), tf.argmax(y_,1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
print('MNIST手写图片准确率:')
print(accuracy.eval({x: mnist.test.images, y_: mnist.test.labels}))
实现的流程:
- 定义算法公式,也就是神经网络前向传播时的计算。
- 定义 loss ,选定优化器,并指定优化器优化 loss。
- 迭代地对数据进行训练。
- 在测试集或验证集上对准确率进行评测。
运行结果如下:
二、CNN
参考博文'用TensorFlow构造CNN进行手写数字识别' http://blog.csdn.net/clcwcxfwf/article/details/72854204
和‘TensorFlow学习笔记(3)----CNN识别MNIST手写数字’http://blog.csdn.net/PhDat101/article/details/52403127
构建两层的卷积神经网络来进行手写数字的识别,数据集是MNIST,开发平台是TensorFlow。
网络的结构为:
input layer => convolutional layer => pooling layer => convolutional layer => pooling layer => fully connected layer => fully connected layer
# coding=utf-8
# !/usr/bin/env python
from tensorflow.examples.tutorials.mnist import input_data
import tensorflow as tf
# 用于构建2层的卷积神经网络
def deepnn(x):
# 把图像向量还原成28*28的图像
x_image = tf.reshape(x, [-1, 28, 28, 1])
# 第一层卷积
# [卷积核的高度,卷积核的宽度,RGB通道数量,输出的通道数量]
W_conv1 = weight_variable([5, 5, 1, 32])
# bias变量,长度为32的向量
b_conv1 = bias_variable([32])
# relu函数,即f(x)=max(0,x)
h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1)
h_pool1 = max_pool_2x2(h_conv1)
# 第二层卷积
W_conv2 = weight_variable([5, 5, 32, 64])
b_conv2 = bias_variable([64])
h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2) + b_conv2)
h_pool2 = max_pool_2x2(h_conv2)
# 全连接层,共1024个神经元
W_fc1 = weight_variable([7 * 7 * 64, 1024])
b_fc1 = bias_variable([1024])
h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1)
h_pool2_flat = tf.reshape(h_pool2, [-1, 7 * 7 * 64])
# dropout是将某一些神经元的输出变为0,这是为了防止过拟合
keep_prob = tf.placeholder(tf.float32)
h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)
# softmax层
W_fc2 = weight_variable([1024, 10])
b_fc2 = bias_variable([10])
y_conv = tf.matmul(h_fc1_drop, W_fc2) + b_fc2
return y_conv, keep_prob
# 卷积函数,待操作的数据x,模板W,tensor不同维度上的步长,强制与原tensor等大
def conv2d(x, W):
return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME')
# pooling函数,平面数据的pool模板2*2,平面数据滑动步长2*2(非重叠的pool)
def max_pool_2x2(x):
return tf.nn.max_pool(x, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')
# 用于分配系数
def weight_variable(shape):
# 均值0标准方差0.1,剔除2倍标准方差之外的随机数据
initial = tf.truncated_normal(shape, stddev=0.1)
return tf.Variable(initial)
# 用于分配偏置
def bias_variable(shape):
# 统一值0.1
initial = tf.constant(0.1, shape=shape)
return tf.Variable(initial)
# 读取数据
mnist = input_data.read_data_sets('input_data', one_hot=True)
# 定义图片和标签的占位符
# x是输入的图像,y_是对应的标签
x = tf.placeholder(tf.float32, [None, 784])
y_ = tf.placeholder(tf.float32, [None, 10])
y_conv, keep_prob = deepnn(x)
# 用交叉熵来计算loss
cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y_, logits=y_conv))
# AdamOptimizer调参
train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
# 模型评估
correct_prediction = tf.equal(tf.argmax(y_conv, 1), tf.argmax(y_, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
with tf.Session() as sess:
# 初始化
sess.run(tf.global_variables_initializer())
# 开始训练
for i in range(20000):
# 每次50张图片
batch = mnist.train.next_batch(50)
# 每100次迭代输出一次日志
if i % 100 == 0:
train_accuracy = accuracy.eval(feed_dict={x: batch[0], y_: batch[1], keep_prob: 1.0})
print('step %d, training accuracy %g' % (i, train_accuracy))
train_step.run(feed_dict={x: batch[0], y_: batch[1], keep_prob: 0.5})
# 验证最终的准确率
print('test accuracy %g' % accuracy.eval(feed_dict={x: mnist.test.images, y_:mnist.test.labels,keep_prob:1.0}))
实现的流程:
- 构建卷积网络
- 定义 loss ,选定优化器,并指定优化器优化 loss。
- 迭代地对数据进行训练。
- 在测试集或验证集上对准确率进行评测。