用TensorFlow搭建一个全连接神经网络

说明

本例子利用TensorFlow搭建一个全连接神经网络，实现对MNIST手写数字的识别。

先上代码

from tensorflow.examples.tutorials.mnist import input_dataimport tensorflow as tf# prepare datamnist = input_data.read_data_sets('MNIST_data', one_hot=True)xs = tf.placeholder(tf.float32, [None, 784])ys = tf.placeholder(tf.float32, [None, 10])# the model of the fully-connected networkweights = tf.Variable(tf.random_normal([784, 10]))biases = tf.Variable(tf.zeros([1, 10]) + 0.1)outputs = tf.matmul(xs, weights) + biasespredictions = tf.nn.softmax(outputs)cross_entropy = tf.reduce_mean(-tf.reduce_sum(ys * tf.log(predictions),                                              reduction_indices=[1]))train_step = tf.train.GradientDescentOptimizer(0.5).minimize(cross_entropy)# compute the accuracycorrect_predictions = tf.equal(tf.argmax(predictions, 1), tf.argmax(ys, 1))accuracy = tf.reduce_mean(tf.cast(correct_predictions, tf.float32))with tf.Session() as sess:    init = tf.global_variables_initializer()    sess.run(init)    for i in range(1000):        batch_xs, batch_ys = mnist.train.next_batch(100)        sess.run(train_step, feed_dict={            xs: batch_xs,            ys: batch_ys        })        if i % 50 == 0:            print(sess.run(accuracy, feed_dict={                xs: mnist.test.images,                ys: mnist.test.labels            }))

代码解析

1. 读取MNIST数据

mnist = input_data.read_data_sets('MNIST_data', one_hot=True)

2. 建立占位符

xs = tf.placeholder(tf.float32, [None, 784])ys = tf.placeholder(tf.float32, [None, 10])

xs 代表图片像素数据, 每张图片(28×28)被展开成(1×784), 有多少图片还未定, 所以shape为None×784.

ys 代表图片标签数据, 0-9十个数字被表示成One-hot形式, 即只有对应bit为1, 其余为0.

3. 建立模型

weights = tf.Variable(tf.random_normal([784, 10]))biases = tf.Variable(tf.zeros([1, 10]) + 0.1)outputs = tf.matmul(xs, weights) + biasespredictions = tf.nn.softmax(outputs)cross_entropy = tf.reduce_mean(-tf.reduce_sum(ys * tf.log(predictions),                                              reduction_indices=[1]))train_step = tf.train.GradientDescentOptimizer(0.5).minimize(cross_entropy)

使用Softmax函数作为激活函数：

o u p u t = S o f t m a x (i n p u t \times w e i g h t + b i a s)

$ouput=Softmax(input \times weight + bias)$

4. 计算正确率

correct_predictions = tf.equal(tf.argmax(predictions, 1), tf.argmax(ys, 1))accuracy = tf.reduce_mean(tf.cast(correct_predictions, tf.float32))

5. 使用模型

with tf.Session() as sess:    init = tf.global_variables_initializer()    sess.run(init)    for i in range(1000):        batch_xs, batch_ys = mnist.train.next_batch(100)        sess.run(train_step, feed_dict={            xs: batch_xs,            ys: batch_ys        })        if i % 50 == 0:            print(sess.run(accuracy, feed_dict={                xs: mnist.test.images,                ys: mnist.test.labels            }))

运行结果

训练1000个循环, 准确率在87%左右.

Extracting MNIST_data/train-images-idx3-ubyte.gzExtracting MNIST_data/train-labels-idx1-ubyte.gzExtracting MNIST_data/t10k-images-idx3-ubyte.gzExtracting MNIST_data/t10k-labels-idx1-ubyte.gz0.10410.6320.73570.78370.79710.81470.82830.83760.84230.85010.85010.85330.85670.85970.85520.86470.86540.87010.87120.8712