Author: scott farrell
Date: 00:25:48 01/13/04
Go up one level in this thread
On January 13, 2004 at 03:02:52, Uri Blass wrote: >On January 13, 2004 at 02:15:36, scott farrell wrote: > >>On January 13, 2004 at 00:00:14, Uri Blass wrote: >> >>>On January 12, 2004 at 20:05:11, scott farrell wrote: >>> >>>>On January 12, 2004 at 17:34:52, Uri Blass wrote: >>>> >>>>>I do not see when it happens in crafty and note that it is not only a question >>>>>of speed when it is not clear what is faster and if I understand correctly >>>>>crafty may be wrong when a sequence of captures that is finished in trading >>>>>kings happens and it may happen in the endgames. >>>>> >>>>>[D]8/5r2/5k2/3R1p2/5K2/8/8/8 w - - 0 1 >>>>> >>>>>Rxf5+ is a good capture but based on SEE that let trading kings it can be >>>>>pruned. >>>>> >>>>>Uri >>>> >>>> >>>>thanks uri/sune, added some stuff about a king, >>>> >>>>my acutal check was if king is capturing, and -stm still has pieces in the >>>>capture list, return -king for that part. I also added the dont xray the king, >>>>but I didnt see any effect - but clearly it is smart. >>>> >>>>My goal is to have 2 SEE, one fast, and one smart. >>>> >>>>output from chompster: >>>> >>>>the key is: >>>>maxthreat/SEE+/SEE ala crafty/NA >>>> >>>>as you can see it scores the Kxf5 as +1, and kxf5 as -50, interestingly with >>>>checks in the qsearch either way found kxf5+ with the same 33 nodes. I think >>>>crafty might need 2 plies instead of one, with an incorrect SEE and no checks in >>>>the qsearch. >>>> >>>>notice RE5 says 0 on both SEE and SEE+, I think SEE (ala crafty) just minimax's >>>>the king swap and accidently gets the right score. >>>> >>>>8/5r2/5k2/3R1p2/5K2/8/8/8/ w - - 0 0 >>>>normal moves: >>>>RD8 (D5D8) , maxThreat=0/0/0/0 >>>>RD7 (D5D7) , maxThreat=0/-5/-5/0 >>>>RD6 (D5D6) , maxThreat=9/0/0/0 >>>>RA5 (D5A5) , maxThreat=0/0/0/0 >>>>RB5 (D5B5) , maxThreat=0/0/0/0 >>>>RC5 (D5C5) , maxThreat=0/0/0/0 >>>>RE5 (D5E5) , maxThreat=0/0/0/0 >>>>RxF5 (D5F5) , maxThreat=9/1/-4/0 >>>>RD4 (D5D4) , maxThreat=0/0/0/0 >>>>RD3 (D5D3) , maxThreat=0/0/0/0 >>>>RD2 (D5D2) , maxThreat=0/0/0/0 >>>>RD1 (D5D1) , maxThreat=0/0/0/0 >>>>KE5 (F4E5) , maxThreat=9/-50/0/0 >>>>KxF5 (F4F5) , maxThreat=9/-50/-4/0 >>>>KG5 (F4G5) , maxThreat=9/-50/-50/0 >>>>KE4 (F4E4) , maxThreat=1/-50/-50/0 >>>>KG4 (F4G4) , maxThreat=1/-50/-50/0 >>>>KE3 (F4E3) , maxThreat=0/0/0/0 >>>>KF3 (F4F3) , maxThreat=0/0/0/0 >>>>KG3 (F4G3) , maxThreat=0/0/0/0 >>> >>>I do not understand it. >>> >>>Why maxthreat of Re5 is 0 when maxthreat of the illegal Ke4 is 1? >>>both threat the pawn at f5 and if we talk about new threat then they do not have >>>a new threat. >>yeh, maxThreat is only the maximum piece that can be attacked after the move is >>made, Its not SEE at all, I guess Re5 should return 1, attacking the pawn. I >>think maxThreat is not used much these days, but still in the debug. I'll check >>Re5 anyway. >> >>> >>>I also do not understand why threat against the king is 9 when SEE of exposing >>>the king to check is -50. >>> >>Maxthreat doesnt differentiate between kings and queens, its only to detect >>moves that are threatening, as opposed to actually capturing anything, if some >>threats a king its nearly the same as threat to a queen, you better do something >>about it - or probably lose the game. >> >>I guess you saw the nice 1.0 SEE+ gain for Rxf5 - you have to like that like. > >Yes > >I am debugging my SEE. >I already found errors in previous versions and I hope that latest version have >no errors. > >In the position that you posted >movei find also 1 for Rxf5 -5 for Rd7 and 0 for other moves. > looks like we are on the same page, let me know if you find any more tricky positions. >I think that asking the program to print the moves with SEE different than 0 can >be productive for debugging. > >I will probably calculate manually SEE for all legal moves of 100 different >positions to verify that movei does not get them wrong. > >Uri I was thinking of comparing it to a material only qsearch, that searches all moves, win or loss, and then troll through by hand the ones that are different. But I havent done it yet. Scott
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.