Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Evaluation in Rebel

Author: Uri Blass

Date: 03:30:54 01/07/03

Go up one level in this thread


On January 07, 2003 at 06:20:15, Ed Schröder wrote:

>On January 07, 2003 at 06:00:58, Uri Blass wrote:
>
>>On January 07, 2003 at 05:47:37, Ed Schröder wrote:
>>
>>>On January 07, 2003 at 05:05:31, Tony Werten wrote:
>>>
>>>>Hi Ed,
>>>>
>>>>your board tables seem to ignore xray attacks.
>>>
>>>xray attacks are done, it's only not mentioned as I thought it would be
>>>self-understood.
>>>
>>>
>>>>Is that because it doesn't matter too much, or because it's mainly used for move
>>>>ordering ( so it's only less effective but not wrong ) ?
>>>
>>>I use xray also for evaluation, for instance for mobilty, king safety, so xray
>>>must be done and in an accurate way, I will write a little about it in the
>>>mobility section.
>>>
>>>Ed
>
>
>>Hi Ed,
>>
>>I have 2 questions about the evaluation.
>>
>>1)Did you try to do incremental evaluation.
>
>I have tried it in the past, it was maddening, if you are in love with bugs,
>self-torture and headaches then doing incremental evaluation is the right path
>to go.
>
>
>>I think that incremental evaluation may be faster than doing the full evaluation
>>at every node until depth-1.
>
>It is faster of course, but don't hold your breath, you will have to restore
>things too when climbimg back in the tree (decremental evaluation), meaning
>things have to be done twice. On "incremental evaluation", you can of course put
>all things on stack and just copy things back but that is also costly.
>
>
>>The problem is that doing it without bugs is not easy espacially when you have a
>>big evaluation.
>
>Yes, that's why I never finished the job, the second reason: it wasn't that much
>faster.
>
>
>>2)Did you check that your full evaluation is really productive if you assume the
>>same number of nodes.
>>I remember that when I tested personalities of Rebel I found that knowledge=500
>>changes it's mind more often so it needs more nodes to finish the same
>>iteration.
>
>Natuarally [Chess Knowledge = 500] beats [Chess Knowledge = 100] slightly based
>on fixed depth, but it loses with a big margin using a normal time control,
>remember the gain from Lazy Eval is over a factor of 3.
>
>Ed

I know it but the interesting question is based on fixed number of nodes because
my impression was that chess knowledge=500 needs more nodes to finish the same
depth.

Uri



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.