Author: Uri Blass
Date: 07:13:37 07/20/04
Go up one level in this thread
On July 19, 2004 at 07:31:35, Omid David Tabibi wrote: >It is well known that the ECM test suite has a large number of erros and >inaccuracies. GCP has compiled a test suite which doesn't include the more >problematic positions. But I wanted to take a look at all 879 positions of ECM >and correct them. > >On my AMD64 3200+, I ran Junior 8, Deep Fritz 8, Shredder 8, and Falcon for 10 >seconds per position. Results were: > >Shredder 8: 696 >Deep Fritz 8: 696 >Falcon: 680 >Junior 8: 637 > >Next, I analyzing all the positions were all the engines failed to solve (about >100 positions). The resived test suite can be downloaded from: > >http://www.cs.biu.ac.il/~davoudo/pubs/ecmx.epd > >Each of these 100 positions was of one of the following types: > >1. Correct move, but requires very deep search. Naturally I didn't change >anything. > >2. Correct move, but there are additional winning moves. I added the other >winning moves. Sometimes the alternative winning moves were much better, but I >didn't remove the original move, as long as it was not a blunder. > >3. Blunder or bad moves. Some suggested best moves are simply blunders. It seems >that these moves have been introduced as a result of typos (e.g. Bd4 instead of >Bc4). In some other cases the best move is wrong because the defending side has >a nice (and usually deep) defense, apparently not taken into consideration when >compiling the test suite. > >4. I didn't find any win for the suggested best move, or am not sure whether it >is indeed the best move. There are 26 such positions which appear below. It >would be nice if you could analyze them longer and see whether the suggested >best move is indeed the best, or alternatives should be added: > >r3r1k1/pp3ppp/1qp5/2bbNQ2/5B2/2P5/PP3PPP/3RR1K1 w - - bm Rxd5; id "ECM.349"; >1k2rb1r/pbp1q1pp/2pp4/2PnpP2/Q7/1P3NP1/PB5P/2KR1B1R w - - bm Rxd5; id "ECM.371"; >2bq1b2/1p3k2/1r1p1p1p/pNpP1PpP/P1P3P1/6B1/1P5Q/2K1R3 w - - bm Re6; id "ECM.396"; >rq1r2k1/3nbp1p/p2p2p1/1p1Pp1P1/3B1P2/3B3Q/PPP4P/2KR2R1 w - - bm f5; id >"ECM.497"; >r3k1r1/pb1qpp1p/1pnp1n1B/2p2R2/3PP3/2P3PN/P1P3BP/R2Q3K b q - bm Nxe4; id >"ECM.561"; >8/p5k1/2p2p2/4p1p1/1p2P3/1P4PP/1PP5/5K2 b - - bm g4; id "ECM.601"; Movei found the move and based on another post it may be one of the moves that leads to draw. >8/8/4b1p1/2Bp3p/5P1P/1pK1Pk2/8/8 b - - bm g5; id "ECM.605"; >4r2k/3n1Qpp/1pRP1p2/p4P2/1p1P2P1/6rP/P3q1B1/6RK w - - bm Qe6; id "ECM.624"; >r1b2rk1/2q1bppp/p1n1p3/1p1np1P1/5P2/1NNBBQ2/PPP4P/R4RK1 w - - bm Bxh7+; id >"ECM.665"; >2rqr1k1/1p1bbppp/p3p3/n7/3P2Q1/2PB1N2/P4PPP/R1B1R1K1 w - - bm Bxh7+; id >"ECM.673"; >r1b2r1k/ppppq1pp/2n1n3/6N1/2B2P2/4B3/PPP3PP/R2Q1RK1 w - - bm Nxh7; id "ECM.674"; >rn1q1rk1/pb1p1ppp/4p3/2pnP3/1bp5/2N2N2/PP3PPP/RBBQK2R w KQ - bm Bxh7+; id >"ECM.675"; >r1b2rk1/1pq2ppp/p1p1p3/2n1P3/2N2P2/3B4/PPP3PP/R2Q1RK1 w - - bm Bxh7+; id >"ECM.677"; >2k3r1/ppq2p1p/n1Pb1p2/8/6r1/4BN1b/PPQ2PPP/R3RBK1 b - - bm Bxh2+; id "ECM.698"; >r5k1/pn1q1rpp/2pp4/5R1N/bP6/4BQ2/P4PPP/2R3K1 w - - bm Nxg7; id "ECM.714"; >4kbr1/pp1bpp1p/3p1p2/3N4/1PBQP3/q7/P1r2PPP/R4RK1 b - - bm Rxg2+; id "ECM.725"; >r3r1k1/pbq1ppbp/1pp2np1/4N3/3P4/2P5/PPB1QPPP/R1B1R1K1 w - - bm Nxf7; id >"ECM.762"; >2rqr1k1/pb2bp1p/1pn1pnpB/4N3/3P4/P1N3R1/1P3PPP/RB1Q2K1 w - - bm Rxg6+; id >"ECM.810"; >1q1r3k/3P1pp1/ppBR1n1p/4Q2P/P4P2/8/5PK1/8 w - - bm Rxf6; id "ECM.825"; >rr4k1/5p1p/n2p1npQ/q1pP4/6P1/2N2P1N/PP1R3P/2K4R b - - bm Rxb2; id "ECM.853"; >5rk1/5rp1/p2qb3/R2Np2Q/3bP2B/7P/P5PK/1R6 b - - bm Rf1; id "ECM.856"; >3r1rk1/p3q1pp/2Rb4/3P4/8/1P2Qp2/PB3PPP/3R2K1 b - - bm Bxh2+; id "ECM.858"; >8/2q1np1k/6p1/1PPn3p/8/P2Q3P/6P1/2R4K b - - bm Nf5; id "ECM.864"; >7R/1p1r4/2b2p2/2Pp1qk1/3P1bp1/3N4/2Q2PB1/6K1 w - - bm Rg8+; id "ECM.866"; It is one of the problems that Movei needs slightly more than 10 seconds to find the suggested move. Is there a better move than Rg8? Movei's score at depth 14 is 0.04 pawns against white. I will give it to continue to search because it may change it's mind. Uri
This page took 0.01 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.