1.單層神經(jīng)網(wǎng)絡(luò)

#從TensorFlow中導(dǎo)入MNIST數(shù)據(jù)集
import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("MNIST_data",one_hot=True) 
#定義變量
x = tf.placeholder(tf.float32,[None,784])
y_ = tf.placeholder(tf.float32,[None,10])
#創(chuàng)建神經(jīng)網(wǎng)格
W = tf.Variable(tf.zeros([784,10]))
b = tf.Variable(tf.zeros([10]))
y = tf.nn.softmax(tf.matmul(x,W) + b)
#二次代價(jià)函數(shù)
loss = tf.reduce_mean(tf.square(y_ - y))
#梯度下降法
train_step = tf.train.GradientDescentOptimizer(0.5).minimize(loss)
#初始化變量
init = tf.global_variables_initializer()
#求準(zhǔn)確率
correct_prediction = tf.equal(tf.argmax(y,1),tf.argmax(y_,1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction,tf.float32))
with tf.Session() as sess:
    sess.run(init)
#對(duì)55000個(gè)訓(xùn)練數(shù)據(jù)進(jìn)行21次訓(xùn)練
    for n in range(21):
        for i in range(550):
            batch_xs,batch_ys=mnist.train.next_batch(100)
            sess.run(train_step,feed_dict={x:batch_xs,y_:batch_ys})
            acc=sess.run(accuracy,feed_dict={x:mnist.test.images,y_:mnist.test.labels})
        print("Iter " + str(n)+",Testing Accuracy "+str(acc))

• 通過以上的代碼可以在測(cè)試數(shù)據(jù)上平均達(dá)到92% 的準(zhǔn)確率

2.增加一個(gè)隱藏層

import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("MNIST_data",one_hot=True) 
#每個(gè)批次的大小
batch_size = 100
#計(jì)算一共有多少個(gè)批次
n_batch = mnist.train.num_examples//batch_size
#定義幾個(gè)placeholder
x = tf.placeholder(tf.float32,[None,784])
y_ = tf.placeholder(tf.float32,[None,10])
#keep_prob = tf.placeholder(tf.float32)
#創(chuàng)建一個(gè)含隱藏層的神經(jīng)網(wǎng)絡(luò)
w1 = tf.Variable(tf.truncated_normal([784,300],stddev=0.1))
w2 = tf.Variable(tf.zeros([300,10]))
b1 =tf.Variable(tf.zeros([300]))
b2 =tf.Variable(tf.zeros([10]))
L1=tf.nn.relu(tf.matmul(x,w1)+b1)
y =tf.nn.softmax(tf.matmul(L1,w2)+b2)
#二次代價(jià)函數(shù)
loss=tf.reduce_mean(tf.square(y_ - y))
#使用梯度下降法
train_step=tf.train.GradientDescentOptimizer(0.5).minimize(loss)
#初始化變量
init=tf.global_variables_initializer()
#結(jié)果存放在一個(gè)布爾型列表中
correct_prediction=tf.equal(tf.argmax(y,1),tf.argmax(y_,1))
#求準(zhǔn)確率
accuracy=tf.reduce_mean(tf.cast(correct_prediction,tf.float32))
with tf.Session() as sess:
    sess.run(init)
    for epoch in range(21):
        for batch in range(n_batch):
            batch_xs,batch_ys=mnist.train.next_batch(batch_size)
            sess.run(train_step,feed_dict=({x:batch_xs,y_:batch_ys}))
        acc = sess.run(accuracy,feed_dict={x:mnist.test.images,y_:mnist.test.labels})
        print("Iter " + str(epoch)+",Testing Accuracy "+str(acc))

• 其它條件不變日戈，只增加一個(gè)隱藏層另凌，訓(xùn)練21次可以達(dá)到95%左右的準(zhǔn)確率谱轨，訓(xùn)練200次可以達(dá)到接近98% 的準(zhǔn)確率。
• 事實(shí)上吠谢，現(xiàn)在的Softmax Regression加入隱含層變成一個(gè)正統(tǒng)的神經(jīng)網(wǎng)絡(luò)后土童，再結(jié)合Dropout、Adagrad工坊、ReLU等技術(shù)準(zhǔn)確率可以達(dá)到98%献汗。引入卷積層、池化層后王污，也可以達(dá)到99%的正確率罢吃。而目前基于卷積神經(jīng)網(wǎng)絡(luò)的state-of-the-art的方法已經(jīng)可以達(dá)到99.8%的正確率。
  引用：《TensorFlow實(shí)戰(zhàn)》