Author: Uri Blass
Date: 08:19:43 03/10/02
Go up one level in this thread
On March 10, 2002 at 10:47:09, Vincent Diepeveen wrote: >On March 09, 2002 at 17:38:57, Uri Blass wrote: > >>On March 09, 2002 at 15:41:45, Vincent Diepeveen wrote: >> >>>On March 08, 2002 at 10:12:07, Uri Blass wrote: >>> >>>>On March 08, 2002 at 09:22:58, Ed Schröder wrote: >>>> >>>>>[d]4r3/p1p1pPp1/P1P1P1P1/5K2/3p2P1/7p/3P1ppr/3R1nkq w - - id M19; >>>>> >>>>>I came across this beauty, mate in 19 moves. Rebel has no clue. >>>>> >>>>>Ed >>>> >>>>I could solve it by my head in some minutes(I used the program only as a >>>>chessboard to go backward and forward) >>>> >>>>I believe that this position should be easy for computers if people only write >>>>the right programs(something that I did not do). >>>> >>>>singular extension may help but they should be implemented correctly in order >>>>not to do the program significantly slower in other positions. >>> >>>i get the impression you don't know exactly what singular extensions are. >>> >>>a singular extension is if 1 move is better by a margin S than all other >>>moves in that position. >>> >>>However here we talk about giving the opponent 1 legal move. >> >>I think that logic says that single reply extension has to be included in >>singular extensions but the opposite is not true so I agree that single reply >>extensions are enough. > >single reply is only when a check has just a single reply. we talk about >a single legal move and i doubt anyone knows at generation time that >Ne3 and Ng3 are illegal. My program knows at generation time the list of the legal moves and the only extension that is used today by my program is the check or single reply extension that means thatmy program extend one ply for every position when the king is in check or the side to move has only one legal move. > >>The idea of singular extensions is to extend when there is an obvious move. >>A case when one move is clearly better than the other moves is a case of an >>obvious move. > >Not really. Hsu/Campbell have described it as next: > >a move m is singular if it is a margin S better than all >other moves. > >Obviously this isn't the case here, because all legal moves from >white give a draw score, also the moves that you define as >'obvious' :) > >>A case when there is a single reply is also a case of an obvious move. > >>The only problem is that the white moves are not obvious moves and if you extend > >'obvious' is a human word, never used in any algorithmic description AFAIK :) > >>every move that force an obvious move you can do the program significantly >>slower. > >For white i see loads of moves each move. The move that gives black >just 1 legal move is not in my program seen as 'obvious' :) > >>I believe that I have ideas how to solve it without doing the program slower >>in other cases but I did not try them and I do not use them in my program today. > >Black can also nullmove, had you thought about that? My program does not use null move and I doubt if it is going to use it in the future. I prefer to see if I can add productive extensions and only later to test null move. Junior does not use null move so it is possible that my program also is not going to use it. > >So even if you only generate legal moves, which makes this problem like 20 >ply on paper, then still the problem is that black can nullmove and a move >later its stalemate. > >I see no way to find this with normal extensions. I agree that the extension that are used today by most chess programs including mine are not enough to solve this problem. chessmaster could solve it in an hour so it is not true for all the programs. 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.