Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: has any program reached master level on chinese chess ?

Author: Dezhi Zhao

Date: 11:15:21 06/25/04

Go up one level in this thread


On June 25, 2004 at 13:59:51, Uri Blass wrote:

>On June 25, 2004 at 13:24:34, Dezhi Zhao wrote:
>
>>On June 25, 2004 at 12:07:03, Uri Blass wrote:
>>
>>>On June 25, 2004 at 09:43:43, Dezhi Zhao wrote:
>>>
>>>>On June 24, 2004 at 20:18:34, Keith Evans wrote:
>>>>
>>>>>On June 24, 2004 at 15:07:41, Dezhi Zhao wrote:
>>>>>
>>>>>>On June 24, 2004 at 00:24:39, Keith Evans wrote:
>>>>>>
>>>>>>>On June 23, 2004 at 13:48:03, Vincent Diepeveen wrote:
>>>>>>>
>>>>>>>>On June 22, 2004 at 07:46:02, TEERAPONG TOVIRAT wrote:
>>>>>>>>
>>>>>>>>>
>>>>>>>>>Hi,
>>>>>>>>>
>>>>>>>>>I've never seen any downloadable Chinese chess program that can match
>>>>>>>>>a strong human player. IMHO, it's much more difficult to create a master
>>>>>>>>>level program than in chess. In chess, when you lose the first pawn,it's
>>>>>>>>>likely that you'll lose the game eventually. But, in Chinese chess, you may
>>>>>>>>>find yourself in trouble after you have 1-2 pawns up in the opening.
>>>>>>>>>And you have to handle many specific endgame positions differently.
>>>>>>>>
>>>>>>>>>Regards,
>>>>>>>>>Teerapong
>>>>>>>>
>>>>>>>>There is no commercial motivation for writing chinese chess programs that's the
>>>>>>>>sole reason why the more popular of the 2 games is dominant in computer games.
>>>>>>>>
>>>>>>>>I know the rules of both games and can assure you that it is for an outsider
>>>>>>>>much harder to write a chessprogram beating the strongest chessprograms, than it
>>>>>>>>is to write a chinese chess program beating the strongest chinese chess
>>>>>>>>programs.
>>>>>>>>
>>>>>>>>Note that both require a big effort, but chinese chess is at a much lower level
>>>>>>>>thanks to commercial driven developments in chess.
>>>>>>>
>>>>>>>Don't you think that rules for xiangqi are more complicated that for chess? (See
>>>>>>>Chapter 4 Section 4 of http://www.clubxiangqi.com/rules/asiarule.htm) I don't
>>>>>>>think that there is any free Xiangqi program which understands these rules. If
>>>>>>>you could distill these rules into some trivial C code, then you could help to
>>>>>>>improve the computer referees at the online servers, and you could also help to
>>>>>>>improve the state of non-commercial Xiangqi software.
>>>>>>>
>>>>>>
>>>>>>Agreed. The rules are so complicated that a restricted search seems to be
>>>>>>necessary for an implementation.  Chinese rules are even more complicated than
>>>>>>Asian rules.
>>>>>>
>>>>>>>It is my belief that Xiangqi masters could exploit programs which don't
>>>>>>>completely understand these rules, but I don't have any firm evidence of this.
>>>>>>>It's difficult to find much written by masters which has been translated into
>>>>>>>English. It's not clear to me that the rules are well defined in
>>>>>>>computer-computer competitions, in fact there are some that believe that the
>>>>>>>rules should be simplified for computers. I assume that masters insist on some
>>>>>>>official rules when playing computers, but I don't know this for a fact. (Either
>>>>>>>AXF or CXA rules?)
>>>>>>
>>>>>>The effect is more than that you have described. If your program does not
>>>>>>understand rules, the search can not produce a correct move in too many cases.
>>>>>>I don't see any trend CXA wants to simplify the rules. Instead they tend to make
>>>>>>it more complex in each revison. Believe it or not, some Xiangqi masters even
>>>>>>admmitted to me that they do not understand the arcane rules well:)
>>>>>
>>>>>Do you have any examples of really decisive wins by masters over computer
>>>>>programs? The shorter the better.
>>>>An example come to my mind at this time is a game between my program and a
>>>>Hongkong master at ICCS years ago. The program had a wining position, ahead with
>>>>materials. However it only knew CXA rules and the game was supposed to be under
>>>>Asian rules. The master drew the game by a one-check and one-mate-threat
>>>>repetition which is not allowed under Chinese rules. Should the program know
>>>>Asian rules, it could have won the game.
>>>
>>>I see no interest in games when the hard part is to know the rules.
>>
>>still far from such extent:)
>>
>>>The hard part should be to choose the correct move and not to find the list of
>>>legal moves.
>>>
>>>I can also define rules that will make it very hard for humans to play the game.
>>>
>>>For example define a game that is identical to chess except the list of legal
>>>moves that is different.
>>>
>>>Players need to play every time one of the moves that win the most material
>>>based on 7 ply brute force search(if there is mate in 4 they have to play the
>>>move that lead to the shortest mate and if they can win a pawn by 7 ply
>>>conmbination but not more than it they have to play the move that wins a pawn
>>>based on 7 ply brute force search).
>>>
>>>Do not confuse win a pawn based on x ply search with real winning of material
>>>and if we play the same game with 1 ply instead 7 plies then every side has to
>>>play the biggest capture in 1 ply except cases that the biggest capture
>>>generates stalemate and in this case they have to play it if all the alternative
>>>leads to position with material disadvantage.
>>>
>>>Computers can do 7 ply brute force search and find the list of the legal moves
>>>but it will be very hard for humans only to find if a move is legal.
>>>
>>>Uri
>>
>>I think the major cause of complicated Xiangqi rules is to prevent easy draws.
>>Perpetual checks, as for a simple example, is an easy-to-do in Xiangqi even for
>>the losing side because a king is confined to only 9 squares. Therefore,
>>perpetual checks are illegal.
>
>I do not think that rule that forbid perpetual check is hard to implement in
>software.
>

Yes. Not fun to implement at all. But it is still doable.

>Simply consider every check that force repetition as illegal move.
>

the one that leads to three fold repetition is illegal or to be verdicted lose
by a TD.

>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.