tensorflow——MNIST for beginner

MNIST for beginner

实质：单层softmax进行10分类

单张图片大小为28x28，拉成784维的列向量。对应的标签为10维的列向量，表示该张图片所属的类别（如果图片表示数字9，属于第十类，则第十维为1，其他九维都为0）。然后我们对输入图片的每个像素点都进行线性加权并加上偏置。所有参数及对应的维度如下所示：

• 输入：$x$(784, 1)
• 权重：$W$(10, 784)
• 偏置：$b$(10, 1)
• 标签：$y^{'}$(10, 1)

具体实现细节

对于单张图片，表示数字i的证据为

$$evidence_i=\sum_j{W_{i,j}x_j}+b_i$$

将所有证据转化成概率

$$y=softmax(Wx+b)$$

其中softmax定义为 $softmax(x)=\frac{exp(x_i)}{\sum_jexp(x_j)}$

误差表示——交叉熵

$$\textrm{cross_entropy}=-\sum_i{y_i^{'}logy_i}$$

batch loss(多样本)

$$loss=mean(\sum_j\textrm{cross_entropy}^j)$$

最后采用随机梯度下降法训练，代码如下：

import tensorflow as tf

# import data
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)

# model
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)
y_ = tf.placeholder(tf.float32, [None, 10])

# loss function & training method
cross_entropy = tf.reduce_mean(-tf.reduce_sum(y_ * tf.log(y), reduction_indices = [1]))
train_step = tf.train.GradientDescentOptimizer(0.5).minimize(cross_entropy)

# test trained model
correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))

# train
sess = tf.Session()  # launch the session
sess.run(tf.initialize_all_variables())  # initialize
for i in range(1000):
    batch_xs, batch_ys = mnist.train.next_batch(100)
    sess.run(train_step, feed_dict = {x: batch_xs, y_: batch_ys})
print(sess.run(accuracy, feed_dict={x: mnist.test.images, y_: mnist.test.labels}))

训练环境

name: TITAN X (Pascal)
major: 6 minor: 1 memoryClockRate (GHz) 1.531
pciBusID 0000:02:00.0
Totamory: 11.90GiB
Free memory: 11.75GiB

结果（测试准确率）

0.9165