Re: [Bug-gnubg] evaluation engine

[Joseph Heled]

I for one would be surprised if that would make an improvement.
Remember the nets are trained for money play, no cube. Perhaps play
can improve if nets are trained for a particular score where the cube
is relevant.

-Joseph

On Sat, Feb 7, 2009 at 7:36 PM, bob koca <address@hidden> wrote:
>
>    Has it been suggested to have a separately trained evaluation engine for
> DMP (1away 1 away score or equivalent),  GS (1 away, 2away Crawford
> or equivalent score) and GG (2 away 1 away Crawford or equivalent score)?
> The current engine could be used as a starting point.
>
> Bob Koca
>
>>
>> Message: 7
>> Date: Fri, 6 Feb 2009 15:01:20 +0100
>> From: ?ystein Johansen (OJOHANS) <address@hidden>
>
>>
>> Frankly, the strength of the evaluation engine, which is (mostly)
>> based on neural networks, has become so strong that it is really hard
>> for us to improve it further. Joseph, who train the current neural
>> networks, did a fantastic job, and fundamental changes to the neural
>> networks must be done to improve the strength of it game.
>>
>> (I don't think anyone in the development team have been working
>> on the neural networks lately?)
>>
>> -Øystein
>>
>>
>
> ________________________________
> Windows Live™: E-mail. Chat. Share. Get more ways to connect. See how it
> works.
> _______________________________________________
> Bug-gnubg mailing list
> address@hidden
> http://lists.gnu.org/mailman/listinfo/bug-gnubg
>
>