[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Bug-gnubg] Some idle musings re. ratings

From: kvandoel
Subject: Re: [Bug-gnubg] Some idle musings re. ratings
Date: Mon, 15 Sep 2003 18:21:07 +0200 (CEST)

On Mon, 15 Sep 2003, Nardy Pillads wrote:

> > > The binary gnubg.wd file that contains the weights are produced from
> > > an ASCII gnubg.weights. You can download the gnubg.weights file from
> > > Nardy's site (if I remember correctly). You can perturb that file --
> > > I can give you the specs later, or you can try to guess it from
> > > reading the source lib/neuralnet.c
> > >
> > > You can either run makeweights on the perturbed ASCII file or you
> > > can delete the old gnubg.wd, in which case gnubg will use the ASCII
> > > gnubg.weights file.
> >
> > I did not find the ascii version on Nardy's site. Can anyone tell me
> > where I can get it?
> >
> > Please tell me the file format. Thanks,

> http://users.skynet.be/bk228456/gnubg.weights.014.zip

Thanks Nardy.  Correct me if I'm wrong  but I can just  generate it with
"save weights"? The file format seems to be:

#input_units #hidden_units #out_units X X X (don't know what X is?)
i_h weights
h_o weights
"thresholds" (what are those for?)

By perturbing the  weights I can easily create  bots with a controllable
rating. I don't know if they  are any better models of human errors than
noise in the outputs; that seems to be a very good model anyways.


reply via email to

[Prev in Thread] Current Thread [Next in Thread]