Author: Robert Hyatt
Date: 10:43:57 01/24/01
Go up one level in this thread
On January 24, 2001 at 11:59:20, Vincent Diepeveen wrote: > >Oh well, Diep knows this of course and any program will after >6 ply or so realize that it gets a passer, >but still it of course prefers Qe6 initially here. Not necessarily. Black has to try a4 b3 b2 b1, but white tosses in d5 d6 d7 d8 so 6 plies might not be enough to handle this. > >Idem for me, but because i'm a human who's always scared making a mistake >i would go calculate. So as a human i solve it because of search. I solve it not using a search in the context of computer chess searches. I try this: (1) black can certainly promote in 4 tempi. 4 or 5 moves depending on what white does when I play b3. (2) white can promote in 4 tempi as well, again allowing for a possible added move by both sides when white plays d6. It is black's move, so black promotes first. It is a check, so white doesn't get the chance to promote and black wins. I don't have to search the entire tree of alternatives for both sides. I break it up into two sub-problems which _greatly_ simplifies the thing. Crafty takes a lot of time to search all the various pawn pushes for both sides, when in reality, it could solve either sub-problem quite easily and quickly... This somewhat reminds me of the wild7 variant on ICC, where there are two distinct sub-problems to solve, although both have to be solved correctly to win with white. > >In 99 out of 100 similar positions white will still win. Nothing as >slow as a majority that runs. A passer mass like this is going to >win from it nearly always. I wouldn't disagree for the most part. But there are cases where that 'majority' is sitting there primed to play b3 (assume a4 has already been played). If your eval says "hey, white has a pawn that can't be stopped and it will promote in 5 moves, so white wins," then it will be wrong as black has a pawn that will promote in 4 moves with check even... As a human I can count quite easily, and in this case I don't have to do a global search, just resolve the two problems independently... > >Basically what you should see here is the tactical line with the >check fast. I don't think this one is going to be solved "fast" by any program. It might be that a singular-extension program would see it pretty quickly, where "quickly" is a vague term... > >Cool stuff for my new passer extensions... > >Note a human is quite effectively calculating this. I just count like >next: passer needs x steps to promote with a check. white needs y steps >to promote. Then i already have a notion that this race is going to >be very close and in this case that black will promote first, >without realizing of course whether you need to go for the >perpetual check then, or whether black can win it. > >On the other hand if the white king is on g2 it's a clear win for white >after you see next line: > >qe6 qxe6 dxe a4 e7 kf7 e8Q kxe8 h4 etc. > >So most likely this position is a bit TOO complex to >stand still at just eval here. I agree that I don't think it can be statically evaluated from the initial position. But I don't think it needs a 12 ply search either. A very few plies to resolve some key issues, then the static race is on... > >Of course i'm very happy everyone is starting to realize slowly how >important evaluation is, but this is of course an ever going on >problem. If any program wants to progress it needs to learn more >about chess. Compared to a human a program has nowadays very little >weak chains, but all weak chains are chess knowledge dependant very >clearly. > >Both openingspreparement with some of us and evaluation for all of us. > >>The thing that looks hard is that white has all those passers, and any decent >>program should know that passers get stronger as pieces come off the board. >>But they also have to know about candidate passers. As that is the deciding >>thing in this position. As a human I took about 5 seconds to realize that >>trading queens was NFG. Because of the black a/b pawns. > >Depends. Some humans never exchange, i as a chessplayer always >exchange. I would go for Qe6 in blitz probably. I would not even consider the exchange until I am sure that black's b-pawn isn't a problem. It is way over on the other side of the board, where I can't stop it, and if it promotes it is going to check me. That is enough to say "wait a minute, can I afford to trade queens?" which really means "can I promote _before_ my opponent (either with check, or two moves before him if it isn't a check so that I can stop the b-pawn with the queen?" > >>There will come a time when my program, (at least) knows this just as well, >>but spots it in microseconds... Otherwise, this is just another hole to >>exploit. > >This combination is simple a certain depth. You need to see the promotion >with check, that's all there is. If you add passer extensions you already >find this at like 9 or 10 ply or something. > >No big deal this position. It is a big deal if you reach this position 10 plies deep in the tree and decide white is winning. > >It shows however that the human 'safe' way of playing is much preferred >and that using primitif evals like most progs have are asking for >trouble. > >This trick only is in this position because white didn't play on safe >some moves ago! > >So in the end it sure is an eval problem! >
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.