Path: news.mathworks.com!newsfeed-00.mathworks.com!newsfeed2.dallas1.level3.net!news.level3.com!postnews.google.com!57g2000hsv.googlegroups.com!not-for-mail
From:  Greg Heath <heath@alumni.brown.edu>
Newsgroups: comp.soft-sys.matlab
Subject: Re: help me to train the following in neuralnetwork
Date: Wed, 14 Nov 2007 05:14:23 -0800
Organization: http://groups.google.com
Lines: 61
Message-ID: <1195046063.927122.172720@57g2000hsv.googlegroups.com>
References: <fh9clb$ivr$1@fred.mathworks.com>
NNTP-Posting-Host: 69.141.182.127
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
X-Trace: posting.google.com 1195046064 19563 127.0.0.1 (14 Nov 2007 13:14:24 GMT)
X-Complaints-To: groups-abuse@google.com
NNTP-Posting-Date: Wed, 14 Nov 2007 13:14:24 +0000 (UTC)
In-Reply-To: <fh9clb$ivr$1@fred.mathworks.com>
User-Agent: G2/1.0
X-HTTP-UserAgent: Mozilla/4.0 (compatible; MSIE 7.0; Windows NT 5.1; (R1 1.5); .NET CLR 1.1.4322; .NET CLR 2.0.50727; .NET CLR 3.0.04506.30; Seekmo 10.0.341.0),gzip(gfe),gzip(gfe)
Complaints-To: groups-abuse@google.com
Injection-Info: 57g2000hsv.googlegroups.com; posting-host=69.141.182.127;
Xref: news.mathworks.com comp.soft-sys.matlab:437458



On Nov 12, 6:15 am, "Steve " <senthilkumaa...@yahoo.de> wrote:
> Hello all,
>
> I tried to train the foll. network i got an error,could
> anybodysuggest me,
>
> [P]=[p1 p2 p3 p4 p5 p6 p7 p8 p9 p10 p11 p12 p13 p14 p15 ];
>
> [T]=[t1 t2 t3 t4 t5 t6 t7 t8 t9 t10 t11 t12 t13 t14 t15 ];
>
> [NVarin Ncases]=size(P)
> [NVarin Ncases]=size(T)

 [NVarout Ncases]=size(T)

Nvarin = Nvarout = 1
Ncases = 15

>   net   = newff(minmax(P),[1 0 1],
> {'tansig' 'purelin','traingdm'});

[1 0 1] is an error it indicates you have

1 node in the 1st hidden layer
0 nodes in the second hidden layer!
1 output node

What configuration do you really want?

I recommend using one hidden layer via [H 1]
and finding a suitable value for H by trial and error.

>  Y = sim(net,P);
>  plot(P,T,P,Y,'o')
>  hold on
>  net.trainParam.epochs = 100;
>  net.trainParam.goal = 1e-6;

Why 1e-6? Why not some small fraction of var(T)?

>  net = train(net,P,T);
>  Y = sim(net,P);
>  plot(P,T,P,Y,'o')
>
> Where pi= 1 x 20
> corresponding target ti= 1000 x 2 (i= 1 to 15)

This makes no sense to me.

What are Nvarin, Nvarout and Ncases?

Hope this helps.

Greg     

P.S. Searching on

greg-heath pretraining advice

may help