yxb345 发表于 2007-4-18 17:51

求助:神经网络

我在做RBF神经网络插值的时候出现如下警告，虽然依然能得到结果,但是结果不正确.

NEWRB, neurons = 0, SSE = 9.37878
NEWRB, neurons = 5, SSE = 7.7613
NEWRB, neurons = 10, SSE = 7.17171
NEWRB, neurons = 15, SSE = 6.69804
NEWRB, neurons = 20, SSE = 6.50593
NEWRB, neurons = 25, SSE = 6.03001
NEWRB, neurons = 30, SSE = 3.84553
NEWRB, neurons = 35, SSE = 3.50285
NEWRB, neurons = 40, SSE = 3.14666
NEWRB, neurons = 45, SSE = 2.44182
NEWRB, neurons = 50, SSE = 1.38106
NEWRB, neurons = 55, SSE = 0.918209
NEWRB, neurons = 60, SSE = 0.523984
NEWRB, neurons = 65, SSE = 0.000166006
Warning: Matrix is close to singular or badly scaled.
         Results may be inaccurate. RCOND = 5.374890e-017.
(Type "warning off MATLAB:nearlySingularMatrix" to suppress this warning.)
> In D:\MATLAB6p5\toolbox\nnet\nnet\newrb.m (solvelin2) at line 243
In D:\MATLAB6p5\toolbox\nnet\nnet\newrb.m (designrb) at line 200
In D:\MATLAB6p5\toolbox\nnet\nnet\newrb.m at line 130
In D:\MATLAB6p5\work\rbf.m at line 15

perf =

4.2636e-008
请问这是怎么回事?

[ 本帖最后由 eight 于 2007-4-18 18:31 编辑 ]

eight 发表于 2007-4-18 18:33

Matrix is close to singular or badly scaled.
         Results may be inaccurate. RCOND = 5.374890e-017


你贴的错误已经给你提示了：矩阵接近奇异，结果也许不准确。原因我不懂，你自己找找相关资料吧

lxq 发表于 2007-4-18 21:08

只看看出错信息不好诊断呀
把整个问题贴出来大家讨论讨论

yxb345 发表于 2007-4-19 09:41

求助:RBF神经网络时的WARNING

Warning: Matrix is close to singular or badly scaled.
         Results may be inaccurate. RCOND = 5.374890e-017.
(Type "warning off MATLAB:nearlySingularMatrix" to suppress this warning.)
> In D:\MATLAB6p5\toolbox\nnet\nnet\newrb.m (solvelin2) at line 243
In D:\MATLAB6p5\toolbox\nnet\nnet\newrb.m (designrb) at line 200
In D:\MATLAB6p5\toolbox\nnet\nnet\newrb.m at line 130
In D:\MATLAB6p5\work\rbf.m at line 15

perf =

4.2636e-008
这是怎么回事?M文件和相关文件已贴在附件中.

[ 本帖最后由 yxb345 于 2007-4-19 09:44 编辑 ]

yxb345 发表于 2007-4-19 16:06

请教lxq,eight,急啊

为什么会有这样的WARNING啊?我的两个矩阵都没有非常小的值啊,怎么会近奇异呢

eight 发表于 2007-4-19 16:29

原帖由 yxb345 于 2007-4-19 16:06 发表
为什么会有这样的WARNING啊?我的两个矩阵都没有非常小的值啊,怎么会近奇异呢


我很久没有用神经网络了，大都忘记了，你查看一下本版关于神经网络的帖子吧，jimin在这方面颇有心得，你可以参考一下。他的签名档也有个链接，你自己上去逛逛吧

yxb345 发表于 2007-4-19 21:12

参数变化为什么会引起错误?

eg=0.0001;sc=1;me=100;df=5;
=newrb(pn,tn,eg,sc,me,df)
这时候正常运行,但是结果不正确.
我将eg改成0.01就出现在如下错误,这是为什么?
??? Error using ==> set
Invalid handle.

Error in ==> D:\MATLAB6p5\toolbox\nnet\nnet\plotperf.m
On line 133==> set(ud.TrainLine(4),...

Error in ==> D:\MATLAB6p5\toolbox\nnet\nnet\newrb.m (designrb)
On line 210==>   flag_stop=plotperf(tr,eg,'NEWRB',k);

Error in ==> D:\MATLAB6p5\toolbox\nnet\nnet\newrb.m
On line 130==> = designrb(p,t,goal,spread,mn,df);

Error in ==> D:\MATLAB6p5\work\rbf.m
On line 29==> =newrb(pt,tn(:,1:67),eg,sc,me,df);             %建立并训练网络
页: [1]
查看完整版本: 求助:神经网络