Author: scott farrell
Date: 06:09:21 12/04/02
Go up one level in this thread
On December 03, 2002 at 15:51:53, Dezhi Zhao wrote:
>On December 03, 2002 at 04:26:50, scott farrell wrote:
>
>>I choose to start the story now ...
>>
>>A few months ago I added code to show "mate in 4" rather than just mate. It uses
>>mate-depthTree in order to document the mate depth.
>>
>>This helped alot in actually mating opponents, as it can choose the shortest
>>route to mate. rather than as it was just doing lots of checks, that all lead to
>>mate, and randomly actually mating the opponent.
>>
>>This worked well, EXCEPT, in deep checking lines, it actually fails slower, as
>>it tries to look for a better mate. So it actually blew nodes out somewhat on
>>some test positions.
>>
>>I still go a few more plies during iterative deepening to try to find shorter
>>mates - due to null move and pruning etc. But these searches seem to be slow
>>also.
>>
>>I recently added a nice piece of code:
>>
>>if (alpha>INFINITY/2){
>> matein= INFINTY-alpha; //or whatever you use to calc mate depth in plies
>> if (depthTree>matein)
>> return alpha;
>>}
>>
>>Do others think this is valid? fail low if you are already too deep, and just
>>fail low.
>>
>>It sure verifies the mates about 1000 times faster.
>
>This does cause some problems. You could notice the problem of producing longer
>than necessary mate sequence if you test more positions.
>Worse than that, sometimes your modification may cost much more nodes in some
>positions. Your trick sounds logical at first sight. However, you must take
>transposition into account, which makes the seemingly futile search productive
>instead sometimes. I realized this when I played with such idea 2 or 3 years
>ago.
mmmm ... good thinking .... transpositions ..... if you disregard my pruning, is
INIFINITY-depth accurate when you take into account transpositions? I think not,
as the same position, may be achieved through a shorter or longer route.
Do you have a solution to this?
Scott
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.