Author: Uri Blass
Date: 11:36:25 11/08/02
Go up one level in this thread
On November 08, 2002 at 13:37:12, Gerd Isenberg wrote:
>On November 08, 2002 at 10:56:03, Uri Blass wrote:
>
>>On November 08, 2002 at 10:13:44, Gerd Isenberg wrote:
>>
>>>On November 08, 2002 at 09:48:13, Uri Blass wrote:
>>>
>>>>On November 08, 2002 at 08:50:46, Gerd Isenberg wrote:
>>>>
>>>>>On November 08, 2002 at 07:01:15, Grzegorz Sidorowicz wrote:
>>>>>
>>>>>>What do you think about formula?
>>>>>>
>>>>>>if (!(alpha>(CHECKMATE-QVALUE)))
>>>>>>{
>>>>>> do_not_extent=true;
>>>>>>}
>>>>>>
>>>>>>Currently I'm testing this formula for all extensions
>>>>>>and for example I have got 5 solutions more in WAC test...
>>>>>>but now my program can't solve some other positions
>>>>>>(for example CMB-10 from LCTII test)
>>>>>>
>>>>>>Grzegorz
>>>>>
>>>>>Hi Grzegorz,
>>>>>
>>>>>I suppose you want to disable extensions, if alpha is already a winning score.
>>>>>Is the not operator correct, or what is your intention?
>>>>>
>>>>>Gerd
>>>>
>>>>I do not understand this idea.
>>>>
>>>>If alpha is already a winning score then it means that beta is also winning
>>>>score so the game is over.
>>>
>>>Yes, so doing further extensions now may be a waste of time.
>>
>>When the game is already over it is not very important if you do more
>>extensions.
>
>
>May be, that was the intention of the initial question.
>But if alpha in some subtrees already has a winning score, doesn't mean the game
>is over.
In normal search when you do not calculate exact score of moves I see no case
when it happens.
Possibly Grzegorz (corrected) statement will safe some nodes in not
>important subtrees, specially for static extensions.
>
>>
>>The question if you do extensions is important when the root position is not
>>clear and in this case alpha is not going to be a winning score unless you find
>>a win for your opponent and in that case you can not investigate the exact score
>>but search the other moves.
>>
>>Using the evaluation instead of alpha seems more logical but if you have not
>>incremental evaluation you may need to spend time for it or use lazy evaluation.
>>>
>>>>
>>>>The only case when it may be relevant is if you search not in a normal way(for
>>>>example search for exact score of the second best move)
>>>>
>>>>It is one of the ideas that I consider to try in the future(but of course I need
>>>>to do it to reduced depth in order not to waste too much time about it).
>>>>
>>>>Wasting a small amount time can be useful to get knowledge(for example you can
>>>>see if there is a forced move and play it faster if you have an exact score of
>>>>the second best move)
>>>
>>>Similar to singular extensions. No experience with this topic.
>>
>>Yes
>>
>>There are similiarities but I do not like the idea of extending singular moves
>>everywhere in the tree and the exact score may be used also for pruning and not
>>also for exatensions.
>>
>
>
>If you aleady detects singularity, why not using this information for some more
>fractional plies in possible critical variations?
I do not detect singularity today but the point is that I do not want to extend
lines like 1.e4 h5 Qxh5 Rxh5 only because of the fact that Rxh5 is singular.
This is what I mean when I say that I do not want to extend singular moves every
where in the tree
I also do not like to detect sngularity after some quiet moves.
>
>
>>An extreme example is that if you find an exact mate score for a position then
>>you never need to search it in the future.
>>In the normal way you will find less exact mate scores and the only case that
>>you can find a mate score is in case that the move you considered as best has a
>>mate score and this is rare.
>>
>
>Don't think so, at least in my program. My eval detects most mates in one or
>even in N moves.
You may be right here and I do not know.
I admit that I did not check it but the way alphabeta works says that if the
move is good enough you do not search for better move.
I looked at alpha beta as something that is supposed to give me usually alpha
when the score is below alpha and beta when the score is above beta but when I
think about it again it seems that I can have better information and have upper
bound that is often above beta or lower bound that is often below alpha.
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.