神经网络的收敛和权重问题

cccttt · 发表于 2006-3-20 00:23

马上注册，结交更多好友，享用更多功能，让你轻松玩转社区。

您需要登录才可以下载或查看，没有账号？我要加入

x

三层网络，10个输入，6个隐含，2个输出，原程序如下： p=[0.7,0.3,0.9,0.7,0.34,0.5,0.7,0.65,0.7,0.7;... 0.5,0.1,0.1,0.5,0.46,0.7,0.1,0.10,0.1,0.1;... 0.5,0.1,0.1,0.5,0.18,0.5,0.1,0.10,0.1,0.1;... 0.7,0.7,0.5,0.7,0.30,0.7,0.9,0.65,0.1,0.7;... 0.9,0.1,0.5,0.9,0.46,0.1,0.7,0.90,0.1,0.5;... 0.5,0.7,0.5,0.5,0.26,0.5,0.3,0.10,0.1,0.5;... 0.7,0.7,0.9,0.9,0.22,0.7,0.7,0.10,0.1,0.5;... 0.7,0.1,0.1,0.7,0.78,0.3,0.1,0.10,0.3,0.1;... 0.7,0.1,0.7,0.1,0.22,0.3,0.1,0.10,0.1,0.9;... 0.7,0.1,0.7,0.7,0.18,0.9,0.1,0.10,0.1,0.9;... 0.7,0.7,0.7,0.9,0.38,0.9,0.9,0.65,0.1,0.7;... 0.5,0.3,0.1,0.7,0.50,0.1,0.7,0.65,0.1,0.9;... 0.5,0.3,0.5,0.7,0.26,0.3,0.7,0.65,0.3,0.7;... 0.9,0.3,0.7,0.9,0.46,0.7,0.9,0.90,0.1,0.7;... 0.7,0.3,0.7,0.9,0.90,0.9,0.9,0.90,0.1,0.1;... 0.7,0.5,0.7,0.9,0.70,0.7,0.7,0.10,0.1,0.1;... 0.3,0.3,0.3,0.5,0.70,0.3,0.1,0.10,0.1,0.1;... 0.7,0.3,0.9,0.9,0.38,0.5,0.7,0.10,0.1,0.1;... 0.9,0.3,0.5,0.7,0.90,0.7,0.9,0.90,0.1,0.1;... 0.7,0.5,0.3,0.7,0.58,0.5,0.9,0.90,0.1,0.1]'; t=[0.85,0;1.10,1;1.36,1;0.94,0;0.75,0;1.57,1;0.87,0;1.45,1;1.78,1;1.38,1;0.91,0;1.35,1;... 1.04,1;0.78,0;0.75,0;0.97,0;2.1,1;0.95,0;0.90,0;0.94,0]'; net=newff(minmax(p),[10,6,2],{'logsig','logsig','logsig'},'trainlm') net.trainParam.epochs=10000; net.trainParam.goal=1e-5; net.trainParam.show=50; net.trainParam.lr=0.02; net.trainParam.lr_inc=1.04; net.trainParam.lr_dec=0.7; net=train(net,p,t) 
问题 1：为何不收敛？是传递函数问题还是输入样本的问题？ 
 2：需要对T进行归一化处理吗？如何归一化？ 
 3：欲获得十个输入指标在输入样本中的权重值，即欲先得到如下结果： 
隐含层单元

输　入　层　单　元

 输出层单元 1
2 3
4
5
6
7
8 9
10 1 0.843 0.792 0.022 0.216 0.831 0.103 0.799 0.949 0.835 2
0.414 0.056 0.592 0.279 0.226 0.999 0.281
0.976 - 0.119 3
0.844 0.910 0.453 0.806 0.655 0.674 0.046 0.401 -
0.301 4 0.791 0.365 0.298 0.825
0.981 0.016 0.296 0.278 4.364 5
0.375 0.529 0.624 0.592 0.248 0.099 0.381
0.159 0.113 6
0.956 0.754 0.643 0.978 0.521 0.096 0.294
0.152 0.45 
 再通过相应的公式求得各输入指标中各神经元权重系数，请问通过什么命令可以实现？由于我的输出结果是两个指标，如何同时显示两个输出层单元的神经元权重系数？ 
 请版主和各位大侠帮忙。

ericlin · 发表于 2006-3-20 12:07

1.隐含层神经元数太少或者目标误差太小，或训练函数改为trainrp； 2.要归一化， for i=1:2 t(i,:)=(t(i,:)-min(t(1,:)))/(max(t(1,:))-min(t(1,:))); end
<HR>
 其实你目标函数归一化以后第一个问题就解决了。

[此贴子已经被作者于2006-3-20 12:16:18编辑过]

ericlin · 发表于 2006-3-20 12:18

3.net.iw{1}
ans =
 Columns 1 through 8 
 -7.7223 0.0761 -13.4274 -10.7298 -9.2905 -7.1288 -7.3501 -1.2151 5.7457 5.7912 -0.2446 1.9923 8.7049 -5.4917 7.5222 -1.1844 -6.1592 -4.0895 2.9583 -3.5820 4.1414 -1.1677 -5.3087 -0.9751 6.4948 -2.9569 -0.5805 4.8782 3.7065 5.2725 2.4561 3.4636 0.2160 2.7347 0.9979 -5.8185 3.9658 -3.6653 -3.0706 -3.8463 6.0044 -0.1093 -5.5581 -6.7598 3.0519 -1.5270 -1.2021 -7.4134 -0.5029 2.0670 -1.3826 -0.5567 3.6972 -2.9306 1.2939 -2.0448 0.6290 4.0209 2.8574 4.6071 -1.9534 4.4892 -8.6203 -6.9671 -8.4123 -3.5283 -1.7248 -1.4303 -2.4048 -3.6906 -8.8747 -5.6814 -2.2191 -3.2355 -1.9795 -3.5009 1.4925 1.7258 -4.2359 -4.5635
 Columns 9 through 10 
 -3.0650 -2.0109 2.3326 -8.5651 5.4185 0.6421 8.5712 3.6259 2.6303 1.2828 -3.1021 -8.8550 1.0101 0.1241 -0.1368 -7.6489 -3.7087 2.1934 4.1976 1.1325

chenbing · 发表于 2006-3-20 16:48

我刚作个神经网络的问题，你可以增加第二层的神经元，试验一下，可能会收敛。建议你在输入和输出数据归一后再做。

binghe79 · 发表于 2006-3-23 11:02

net=newff(minmax(p),[10,6,2],{'logsig','logsig','logsig'},'trainlm') 三成网络是三个logsig 吗我怎么觉得应该是两个啊大家说说啊这个例子是一个隐含层还是两个隐含层啊

miniducky · 发表于 2006-3-27 20:10

一般所说的三层网络包含输入层、中间层和输出层。输入层和中间层也称为隐层。隐层通常用logsig作为传递函数，输出层通常用purelin作为传递函数。你所举的这个例子应是两个隐层。

ericlin · 发表于 2006-3-29 15:35

<DIV class=quote>以下是引用binghe79在2006-3-23 11:02:28的发言： net=newff(minmax(p),[10,6,2],{'logsig','logsig','logsig'},'trainlm') 三成网络是三个logsig 吗我怎么觉得应该是两个啊大家说说啊这个例子是一个隐含层还是两个隐含层啊</DIV>
net=newff(minmax(p),[10,6,2],{'logsig','logsig','logsig'},'trainlm') ＝net=newff(minmax(p),[6,2],{'logsig','logsig'},'trainlm')同样是三层网络，输入层是默认的

账号		自动登录	找回密码
密码			我要加入

神经网络的收敛和权重问题

马上注册，结交更多好友，享用更多功能，让你轻松玩转社区。

为何不收敛

神经网络隐层数的确定

回复：（binghe79）net=newff(minmax(p),[10,6,2],{'l...