Author: Dann Corbit
Date: 15:04:42 09/17/04
Go up one level in this thread
On September 17, 2004 at 17:56:21, Uri Blass wrote: >On September 17, 2004 at 15:50:38, Dann Corbit wrote: > >>On September 17, 2004 at 15:06:40, Uri Blass wrote: >> >>>On September 17, 2004 at 14:21:16, Dann Corbit wrote: >>> >>>>On September 17, 2004 at 11:42:38, Robert Hyatt wrote: >>>> >>>>>On September 16, 2004 at 19:48:59, Dann Corbit wrote: >>>>> >>>>>>On September 16, 2004 at 18:18:13, Robert Hyatt wrote: >>>>>> >>>>>>>On September 16, 2004 at 12:52:43, Dann Corbit wrote: >>>>>>> >>>>>>>>On September 16, 2004 at 07:37:01, Ricardo Gibert wrote: >>>>>>>> >>>>>>>>>Your post is a good example of what happens when one jumps to a conclusion >>>>>>>>>without taking the minimum amount of effort needed to understand what is really >>>>>>>>>going on. >>>>>>>> >>>>>>>>I disagree completely, also with everyone else. >>>>>>> >>>>>>>Taking a _longer_ path to win is counter to a tactical test idea. Clearly the >>>>>>>move given is bad, because it just extends the game and reaches the same >>>>>>>position a second time where the _real_ solution has to be played. That is >>>>>>>pointless... >>>>>>> >>>>>>>Otherwise a mate in 3 might turn into a mate in 40 if one side takes every >>>>>>>opportunity to first repeat a second time before making progress... >>>>>> >>>>>>The pv shown by Arasan leads to a win. >>>>>>If it lead to a draw or some other problem I would agree. >>>>> >>>>>You ask a student to add 2 + 2. >>>>> >>>>>He turns in the following: >>>>> >>>>> >>>>>sqrt(100) / sqrt(25) + log10(100) and solves that and turns it in with his >>>>>scratch paper. Do you give him credit? I do not. There are a zillion longer >>>>>ways to do something, such as a tree search in chess. Tactical solutions are >>>>>about the shortest way to win. IE if there is a mate in 8 and a mate in 10, the >>>>>correct answer is the mate in 8. >>>> >>>>There are often strange solutions to test problems (e.g. tossing a queen due to >>>>tablebase simplification). If a chosen move wins 100% of the time, and the >>>>program sees a clear solution, it is a winning move. A winning move cannot be >>>>said to be incorrect. It has exactly the same value as a winning move to a >>>>shorter solution. >>>> >>>>>If there are two equal ways to win, then yes, >>>>>either is correct. But to intentionally repeat a position makes no sense and I >>>>>give it a "zero" as it is pointless... >>>>> >>>>>If you want to count it right, that's ok, but I disagree and I won't. Otherwise >>>>>each WAC position probably has _multiple_ correct solutions... >>>> >>>>If a problem is not proven all the way to win/loss/draw then it is really open. >>>>If there are 10 moves that lead to a definite win, then all ten moves are >>>>solution moves. Including crazy moves like tossing a queen for a tablebase >>>>simplification. >>> >>>I disagree. >>>There are cases when there is only one solution that you can expect chess >>>programs without bugs to find and it is the right solution(tablebase win is not >>>a bug in chess programs so it can be included as a solution but a win by >>>repetition of the root position certainly incdicates a bug). >>> >>>WAC is used to test chess programs for not having bugs in the search and failure >>>should suggest a bug. >>> >>>If you include the stupid move of arasan as a solution then the wac test is >>>losing it's value to detect bugs. >>> >>>By your definition a lot of problems in WAC have more than one solution >>>I found 7 position with more than one solution in the first 20 positions simply >>>by using Fritz at 2 best move in 6 cases or by using my brain in one case to see >>>that white can force repetition and win later. >>> >>>2rr3k/pp3pp1/1nnqbN1p/3pN3/2pP4/2P3Q1/PPB4P/R4RK1 w - - bm Qg6; id "WAC.001"; >>> >>>1)Nh5 also wins and not only Qg6 >>> >>>r1bq2rk/pp3pbp/2p1p1pQ/7P/3P4/2PB1N2/PP3PPR/2KR4 w - - bm Qxh7+; id "WAC.004"; >>> >>>2)Qe3 also wins >>> >>>r4q1k/p2bR1rp/2p2Q1N/5p2/5p2/2P5/PP3PPP/R5K1 w - - bm Rf7; id "WAC.008"; >>> >>>3)Nf7+ Kg8 Nh6+ Kh8 Rf7 also wins >>> >>> >>>r2rb1k1/pp1q1p1p/2n1p1p1/2bp4/5P2/PP1BPR1Q/1BPN2PP/R5K1 w - - bm Qxh7+; id >>>"WAC.014"; >>> >>>4)Ne4 also wins >>> >>>1R6/1brk2p1/4p2p/p1P1Pp2/P7/6P1/1P4P1/2R3K1 w - - bm Rxb7; id "WAC.015" >>> >>>5)b4 also wins >>> >>>1k5r/pppbn1pp/4q1r1/1P3p2/2NPp3/1QP5/P4PPP/R1B1R1K1 w - - bm Ne5; id "WAC.017"; >>> >>>6)I suspect that b6 also wins. >>> >>>r2qkb1r/1ppb1ppp/p7/4p3/P1Q1P3/2P5/5PPP/R1B2KNR b kq - bm Bb5; id "WAC.020"; >>> >>>7)Be6 also wins >> >>If your engine choses one of those alternatives and shows a winning line, or a >>checkmate is found, then those are solutions. > >There are lines that if my engine choose them I know that I have a bug even if >the lines are winning. > >The target of the test suite is to test chess programs so deciding that they are >solutions is not useful because it means that I may miss bugs that the results >suggest. > >finding mate that is not the fastest mate is not always a bug but if the mate is >done by repetition of the root position then it is certainly a bug in my >repetition detection. > >I think that we should consider as solution only a winning move that it is >logical to expect programs without bugs to get. > >> >>If two different moves clearly win, then the moves are equivalent in value. >> >>Unless a move ends in a proven checkmate (e.g. by Chest) then we do not really >>know if it is going to win, and so there is some doubt in it. > >There are cases when there is no doubt about it and if computers give score of >more than +5 and I also see no indication of perpetual check or fortress or >strong attack that the computer underestimates then I can say that I practically >sure about a win but it does not mean that I consider the move as a solution >when there is a better winning move that almost every other program finds. >> >>There are (perhaps) some winning moves that are better than others. But until a >>forced checkmate is demonstrated, we really cannot know which one is better. > >The problem what solution is better is a practical problem and I think that the >right solution is simply the solution that you expect program with no bugs to >find. > >Cases when there are 2 solutions are only cases when you expect part of the >programs with no bugs to find one move and another part to find the second move. > >If programs usaually see that move A win a queen when move B wins a rook then it >is clear that A is the solution because I expect programs without bugs to >choose A(this opinion can be changed if you search deeper and find that A wins >more than a rook but if program choose A because it wins a rook then it has a >bug). I will agree that there is probably some defect here, since the hash table should have shown the shorter solution. I disagree that the objective of a test suite is to find the stated answer. For most test suites, many of the stated answers are just plain wrong, or there do exist answers that are better. I am guessing that a 24 hours per position search by Shredder 8 on a fast machine against the WAC positions that do not end in mate will uncover additional solutions not considered.
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.