Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: question to Tord.

Author: Tord Romstad

Date: 09:02:31 04/12/04

Go up one level in this thread


On April 12, 2004 at 05:51:08, martin fierz wrote:

>On April 11, 2004 at 09:54:37, Tord Romstad wrote:
>
>>On April 11, 2004 at 08:14:38, martin fierz wrote:
>>
>>>On April 11, 2004 at 08:02:45, Tord Romstad wrote:
>>>>When a reduced move fails low, I search it again with full depth.
>>>
>>>hi tord,
>>>
>>>don't you mean "when a reduced move fails high"?
>>
>>Yes, of course.  Thanks for the correction, Martin!
>>
>>Tord
>
>hi tord,
>
>i never tried any of this selective stuff (yet). as your recipe sounds rather
>simple to implement, i will perhaps try it one day.

It is true that it is relatively simple to implement, but it is not absolutely
trivial.  I don't
think it would work very well to consider the history values of moves and
nothing else.
You have to take care in order to avoid tactical mistakes.

And of course, like for all tuning tricks, you must be prepared to do a lot of
tuning before
everything works well.

>do you have any estimate of
>rating gain that you get from this kind of selective search?

It's hard to estimate.  Because my whole search and eval is built with the
purpose of
supporting this kind of tricks, it is not a very fair test to simply disable all
selectivity
and see how much the strength drops.

At least I am fairly sure it is clearly better than a plain recursive nullmove
search with
no other pruning.  Gothmog is competitive with quite a few engines which are
much faster
(nps-wise) and have better evaluation functions, so I must be doing something
right
in my search.

Tord



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.