Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: IMOH no

Author: Uri Blass

Date: 17:26:12 09/19/03

Go up one level in this thread


On September 19, 2003 at 20:06:02, Sune Fischer wrote:

>On September 19, 2003 at 19:46:38, Uri Blass wrote:
>
>>On September 19, 2003 at 17:42:47, Sune Fischer wrote:
>>
>>>On September 18, 2003 at 10:02:33, Edward Seid wrote:
>>>
>>>>I'm learning how to program by reading Deitel's Visual Basic.NET How to Program.
>>>> I'm eager to try out my new skills on a chess-related project.
>>>>
>>>>The Pawn Game - as presented by GM Lev Alburt in Comprehensive Chess Course, Vol
>>>>1
>>>>
>>>>[D]8/pppppppp/8/8/8/8/PPPPPPPP/8 w - - 0 0
>>>>
>>>>The game is won by:
>>>>1- capturing all of the opponent's pawns
>>>>2- reaching the last rank first
>>>>3- 'stalemating' the opponent, while still having at least one move for yourself
>>>
>>>I tried a simple unoptimzed search, just returning matescore for reaching 7th
>>>rank. It's not going as deep as I expected, of course you would not expect to
>>>get the same kind of hitrate as with pawn tables (no color or depth or bounds
>>>dependency in pawn tables):
>>>
>>>1	-3	5	1		1.a3
>>>1	0	10	9		1.a4
>>>2	0	10	33		1.a4 h5
>>>3	0	12	121		1.a4 g5 2.g4
>>>4	0	13	1056		1.a4 g5 2.g4 a5
>>>5	0	15	2529		1.a4 g5 2.e4 g4 3.h4
>>>6	0	22	14205		1.a4 f5 2.f4 a5 3.d4 h5
>>>7	0	26	33039		1.a4 f5 2.h4 a5 3.c4 c5 4.f4
>>>8	0	51	135649		1.a4 f5 2.f4 d5 3.d4 a5 4.h4 h5
>>>9	0	107	360649		1.a4 f5 2.f4 c5 3.h3 a5 4.h4 h5 5.c4
>>>10	0	339	1303780		1.a4 f5 2.d4 a5 3.c4 g5 4.e3 e6 5.c5 g4
>>>11	0	689	2716466		1.a4 f5 2.c4 g5 3.d4 f4 4.e4 g4 5.b4 h6 6.h4
>>>12	0	2241	8928849		1.a4 d5 2.d4 h5 3.h4 a5 4.f4 f5 5.b3 b6 6.c4 c5
>>>13	0	6084	23078374	1.a4 a5 2.c4 h5 3.f4 f5 4.h4 c5 5.d3 e6 6.e3 d5 7.d4
>>>14	0	16715	64599238	1.a4 f5 2.d4 a5 3.h4 c6 4.e3 b5 5.b3 h5 6.c4 bxa4
>>>15	0	37273	146539781	1.a4 f5 2.e3 g5 3.f4 h6 4.b4 c6 5.d4 b5 6.axb5 cxb5
>>>
>>>This is without nullmove or specially tuned eval or move ordering of any kind.
>>>Optimizing for it will of course give a few more plies, but double depth seems
>>>out of reach.
>>>
>>>-S.
>>
>>I do not think that you are right.
>>The branching factor is going to get smaller when you search deeper
>
>Why should that be the case?

because the number of moves get smaller when pawns are blocked.

>I would not expect such a thing to happen until you start hitting the solving of
>the game limit.
>
>You mostly get many transposition in blocked pawn type positions, in this
>position there are no blocked pawns at all, so it's going to take a while for
>the transpositions to really fire.

No tranposition.
after a4 a5 pawns are blocked and you have less moves.
less moves mean smaller branching factor.

>
>There is also another effect, the TTable gets overfilled at some point and
>efficiency starts to drop, even with a good scheme you get more and more
>outdated entries.

I am talking about solving it without hash tables.

>
>>and
>>evaluating passed pawns as winning if they are more advanced than the opponent
>>pawns may significantly help.
>
>Yes, if you do it right you might even be able to cut branches directly at that
>point (not sure, but there must be some obvious cases which would be theoreticly
>sound). The question is if this represents a significant amount of the nodes.
>Perhaps trying to solve an even simpler game: "win by creating a passed pawn"
>could answer this question. I think that game would be pretty hard too.
>
>>I also guess that you should generate pushing moves 2 steps forward first and
>>moves that lose a pawn only last.
>
>The move ordering is the standard move ordering I use for normal search, so it's
>not that far from optimal (I hope not!:).
>
>Again I don't care much about fiddling to get 3-5 more plies, to go from 15 to
>80 plies I need something revolutionary.

You do not need 80 plies.

I believe that there are even no logical games of 50 plies.

I suspect that even your existing version may solve the game if you give it a
lot of time.

I already see a tendency for smaller branching factor in the last plies based on
your data.

I guess that you may get even effective branching factor smaller than 2 if you
continue to run it for many hours if you have no bugs.

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.