Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: The Scalable Search Test

Author: blass uri

Date: 03:05:47 06/20/00

Go up one level in this thread


On June 20, 2000 at 04:11:32, Ed Schröder wrote:

>On June 20, 2000 at 02:12:32, blass uri wrote:
>
>>On June 19, 2000 at 19:02:51, Ed Schröder wrote:
>>
>>>On June 19, 2000 at 13:22:31, blass uri wrote:
>>>
>>>>On June 19, 2000 at 11:47:12, Ed Schröder wrote:
>>>>
>>>>>On June 19, 2000 at 04:45:33, blass uri wrote:
>>>>>
>>>>>>On June 19, 2000 at 03:03:49, Ed Schröder wrote:
>>>>>>
>>>>>>>On June 18, 2000 at 19:46:03, Dave Gomboc wrote:
>>>>>>>
>>>>>>>>On June 18, 2000 at 18:21:01, Ed Schröder wrote:
>>>>>>>>
>>>>>>>>>On June 18, 2000 at 16:34:16, blass uri wrote:
>>>>>>>>>
>>>>>>>>>>On June 18, 2000 at 12:00:08, Ed Schröder wrote:
>>>>>>>>>>
>>>>>>>>>>>On June 18, 2000 at 10:36:40, Gian-Carlo Pascutto wrote:
>>>>>>>>>>>
>>>>>>>>>>>>On June 18, 2000 at 10:17:08, Ed Schröder wrote:
>>>>>>>>>>>>
>>>>>>>>>>>>>According Rebel it is a mate in 10.
>>>>>>>>>>>>>
>>>>>>>>>>>>>Ed
>>>>>>>>>>>>>
>>>>>>>>>>>>>00:00  07.00  0.81  1.Rxh5 gxh5 2.Rxh5 Kf8 3.Rxh7 Ke8 4.Rh8+ Kd7 5.Qf5+ Kc7
>>>>>>>>>>>>>00:01  08.00  2.37  1.Rxh5 gxh5 2.Rxh5 Kf8 3.Rxh7 Ke8 4.Qf5 Kd8 5.Qxf7 Kc8
>>>>>>>>>>>>>6.Qxe7
>>>>>>>>>>>>>00:03  09.00  2.47  1.Rxh5 gxh5 2.Rxh5 Kf8 3.Rxh7 Ke8 4.Qf5 Kd8+ 5.Qxf7 Kc7
>>>>>>>>>>>>>6.Qxe7+ Kc8
>>>>>>>>>>>>>00:43  10.00  11.74  1.Rxh5 gxh5 2.Rxh5 Kf8 3.Rxh7 e6 4.Bf6 Ke8 5.dxe6 Qxc4+
>>>>>>>>>>>>>6.Qxc4
>>>>>>>>>>>>>02:35  11.00  Mate in 10 moves 1.Rxh5 Kf8 Rxh7 2.Ke8 Rxf7 3.Qxc4 Qxc4 4.g5 Rhh7
>>>>>>>>>>>>>5.b5 Rxe7+ 6.Kd8
>>>>>>>>>>>>
>>>>>>>>>>>>I find it interesting to notice that even Rebel has some trouble with this,
>>>>>>>>>>>>taking more than 10 times as much time to resolve the 10th ply.
>>>>>>>>>>>
>>>>>>>>>>>That's pretty normal whith such big score differences (2.47 -> 11.74) as
>>>>>>>>>>>search is confrontated with big score fluctuations which causes move
>>>>>>>>>>>ordering to collapse. I wouldn't bother too much about it, it's quite
>>>>>>>>>>>normal.
>>>>>>>>>>
>>>>>>>>>>Not for all programs.
>>>>>>>>>
>>>>>>>>>Of course.
>>>>>>>>>
>>>>>>>>>>Chessmaster6000(ss=10) can see a forced mate in only 4 seconds and does not have
>>>>>>>>>>a big branching factor
>>>>>>>>>
>>>>>>>>>Sure. Maybe not in this position.
>>>>>>>>>
>>>>>>>>>>This is normal for chessmaster to see mates clearly faster than other programs
>>>>>>>>>>and I think that this is because the other programs do some mistakes in their
>>>>>>>>>>search rules.
>>>>>>>>>
>>>>>>>>>Nope. Johan has special (mate) stuff in his search no program has.
>>>>>>>>>
>>>>>>>>>Ed
>>>>>>>>
>>>>>>>>Just curious... why don't you add it to Rebel too? :-)
>>>>>>>
>>>>>>>a) because I don't know "exactly" what it is;
>>>>>>>b) too much emphasis on mates loses too much elo points in games.
>>>>>>>
>>>>>>>Ed
>>>>>>
>>>>>>I think reason a is the right reason because I do not think that chessmaster
>>>>>>loses too much elo points in games(chessmaster6000 is one of the top programs
>>>>>>and is only one point behind Fritz5.32 in the ssdf list inspite of having no
>>>>>>learning function and having an older engine)
>>>>>>
>>>>>>I believe that having special mate stuff is productive for chess programs if you
>>>>>>do it in the right way.
>>>>>
>>>>>The right way? Now tell me what that is... You might ask yourself the
>>>>>question if such a thing exist in computer chess.
>>>>>
>>>>>Finding deep mates requires extensions. Extensions mean a higher branch
>>>>>factor. A higher branch factor means a lower search depth. A lower search
>>>>>depth means a lower elo.
>>>>
>>>>The right way is to do it almost without a lower search depth.
>>>>In order to do it you need to do the extensions only in the right places so when
>>>>there is no mate danger you will not have lower search depth.
>>>
>>>The right way to do it ussue again. Ever tried to do it yourself?
>>>If not try it yourself and let's talk again 4-5 years later.
>>
>>I did not say that it is easy to do it.
>>>
>>>
>>>>You can also decide to have a time limit for searching with the extensions(for
>>>>example 1/10 of the time per move).
>>>
>>>Tried it?
>>
>>No
>>I did not say it is easy to do it but only that I believe that it is possible.
>>I guess that there are more important things for you to do in the near future.
>>
>>>
>>>It's no good. In fact it is a non working idea. When you for instance
>>>find something interesting in the first part with many extensions you
>>>will lose it as soon as you go back to normal search. Now that looks
>>>real silly for a chess program that after (say) ply 5 finds a nice
>>>tactical shot and then fails low on the 6th ply and then finds the
>>>combination back on ply 9 or 10.
>>
>>This was not what I meant.
>>I did not think to forget about what you found.
>>Of course if you find something intersting(a fail high or a fail low) there is
>>no reason to forget it and you can choose between only avoid doing extensions in
>>nodes that you did not search and if there is a problem with doing it to
>>continue with the extensions.
>
>And that is exactly the point. Of course you will find thousand and 10
>thousands of interesting cases you are forced to "remember" while having
>no single effect on the best move. All these 10 thousands of useless
>extensions are needed to extend again every iteration resulting in time
>loss only.
>
>When you invest (say) 10% of the allowed time in a special tactical
>search and it brings you nothing (which is true for almost every try)
>then you don't end up with a 10% loss but with a percentage that is
>much higher because of all the extensions you have to "remember". If
>you are lucky you will end up with a total loss of 15-20% in a quiet
>position. If you have bad luck search performance will drop with a
>factor of 2-3.
>
>Ed

I suggested to forget about the extensions if you do not find something
interesting and in this case you are only 10% slower.

I suggested not to forget about the extensions when you find something tactical
(fail high or a fail low).

Uri
>
>
>>If you do not find something interesting you can forget about the extensions.
>
>>I think that it may be better in a correpondence game to use chessmaster for
>>1/10 of the time when I suspect that there may be a mate danger and if
>>chessmaster does not see something interesting to use another program for 9/10
>>of the time.
>>
>>I believe that it is better than to use only the other program.
>>If I am right then the other program could be better if it knew to do the same
>>
>>Uri



This page took 0.01 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.