bug-gnubg
[Top][All Lists]
Advanced

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

RE: [Bug-gnubg] User training of the Neural Nets


From: Albert Silver
Subject: RE: [Bug-gnubg] User training of the Neural Nets
Date: Wed, 23 Aug 2006 16:10:16 -0300

Ok, so if I understand it, I can access and use some training, but cannot
change the fundamental aspects such as the inputs, or the nets (as in adding
new ones). 

You see, while scouring the archives, I saw from a discussion from not long
ago, that Ian Shaw had managed to get the plain TD training working, though
no one (at least I saw nothing in the archives) commented on the
effectiveness of his attempts. While looking at the CLI version, I saw two
functions linked to 'train': 

Train database - Train the network from a database of positions
Train td - Train the network from TD(0) zero-knowledge self-play

When I tried to execute 'Train Database' in order to see what happened, it
informed me this function was not compiled into the Windows build.

Although I'll ask if a build with this included could be provided (not just
to me but to anyone interested), I was hoping for some basics on how to use
this, and what to expect.

For example, if I had access to the 'Train DB' function, how could I use it
to include rolled out positions? Could I point it to a directory with .SGF
files, which had already set-up positions, with the rolled out results, and
have it include this into the NNs? Would I then need to run the standard
'Train TD' for this to be integrated, or would the Train DB be enough, after
which the question would be to test and see what differences arose.

                                        Albert

> -----Original Message-----
> From: Christian Anthon [mailto:address@hidden
> Sent: Wednesday, August 23, 2006 2:38 PM
> To: Øystein Johansen
> Cc: Albert Silver; bug-gnubg
> Subject: Re: [Bug-gnubg] User training of the Neural Nets
> 
> On 8/23/06, Øystein Johansen <address@hidden> wrote:
> > The simple answer is unfortunately "no". However, if you want to train
> > GNU Backgammon without much programming experience, you should get the
> > gnubg-nn code from the repository. (I remember I once compiled this for
> > Windows, and it worked.) However, training with these tools alone will
> > not make you able to alter the input calculations.
> >
> 
> Okay then what level of programming skills would be needed and what
> should a programmer do? Even though the current nets are quite good,
> it would be unfortunate if you and Joseph became inactive, so that
> nobody could develop the nets further.
> 
> Christian.






reply via email to

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