[GRASS5] Re: (grass) Neural network tool for grass

[benducke at compuserve.de]

Yes, SNNS has lots of different NN architectures,
maybe a few too many, actually. Most users will
probably have difficulties chosing from so many
different designs.

As far as I remember, back propagation networks are
still very popular because they are so well understood.
Radial basis functions I think are quite popular as well
and have greater classification power.
Kohonen maps are something I never really got acquainted
with.

Is there anybody on this list who could give us a brief
overview which architectures and features are the most important
and would have to be included in a GRASS extension
geared towards practical application?

Cheers,

Benjamin

----- Originalnachricht -----
Von: Paulick Consult <stefan.paulick at urbeli.com>
Datum: Dienstag, 7. Juni 2005 5:01 pm
Betreff: [GRASS5] Re: (grass) Neural network tool for grass

> Maybe it is worth to have a look at SNNS: 
> 
> http://www-ra.informatik.uni-tuebingen.de/SNNS/ 
> 
> /Stefan 
> 
> 
> benducke at compuserve.de schrieb: 
> 
> > I would definitely be interested in porting some NN
> > classification functionality to GRASS 6.
> > Have been thinking about how to do this for some
> > time, as well ...
> > I have got some background in NNs and porting older
> > modules to GRASS 6.
> > I guess the best way would be to take an open source
> > NN library like fann and start from there.
> > Unfortunately, I am tied up with my GRASS extensions
> > installer tool right now.
> > But let's keep in touch about this! 
> > 
> > Benjamin 
> > 
> > ----- Originalnachricht -----
> > Von: Request Tracker <grass-bugs at intevation.de>
> > Datum: Dienstag, 7. Juni 2005 12:36 pm
> > Betreff: [GRASS5] [bug #3327] (grass) Neural network tool for 
> grass 
> > 
> >> this bug's URL: http://intevation.de/rt/webrt?serial_num=3327
> >> -----------------------------------------------------------------
> ---
> >> ----- 
> >> 
> >> Subject: Neural network tool for grass 
> >> 
> >> Platform: GNU/Linux/i386
> >> grass obtained from: Trento Italy site
> >> grass binary for platform: Compiled from Sources
> >> GRASS Version: GRASS 6.1.cvs (2005) 
> >> 
> >> Hallo developers,
> >> I'm clasifying some raster maps, and the result is not the best 
> :-/ 
> >> I use SMAP
> >> classification module. 
> >> 
> >> Many people say, that neural networks should give better 
> results, 
> >> Bayesian Image Segmentation. 
> >> 
> >> I have no idea, how to port r.nntool from GRASS 4.x->6.x. Is 
> there 
> >> someone, who
> >> would do this? I can help with some work.  
> >> 
> >> Would it not be easier to write new r.nntool, using some library 
> >> (e.g. fann)? I do not know, how are neural networks working, but 
> I 
> >> can offer all my skills and time. 
> >> 
> >> Jachym
> >> -- 
> >> Jachym Cepicky
> >> e-mail: jachym.cepicky at centrum.cz
> >> URL: http://les-ejk.cz
> >> GPG: http://les-ejk.cz/gnupg_public_key/ 
> >> 
> >> -------------------------------------------- Managed by Request 
> >> Tracker
> >> _______________________________________________
> >> grass5 mailing list
> >> grass5 at grass.itc.it
> >> http://grass.itc.it/mailman/listinfo/grass5 
> >> 
> > 
> > _______________________________________________
> > grass5 mailing list
> > grass5 at grass.itc.it
> > http://grass.itc.it/mailman/listinfo/grass5
> 
> _______________________________________________
> grass5 mailing list
> grass5 at grass.itc.it
> http://grass.itc.it/mailman/listinfo/grass5
>