Author: Robert Hyatt
Date: 06:58:51 10/06/99
Go up one level in this thread
On October 06, 1999 at 06:51:31, Jeremiah Penery wrote: >On October 06, 1999 at 00:58:15, Robert Hyatt wrote: > >>On October 05, 1999 at 15:15:01, John Stanback wrote: >> >>>On October 05, 1999 at 01:14:03, Harald Faber wrote: >>> >>>>I remember I have heard that Junior has problems with "winning" pawns on a2/a7 >>>>and h2/h7. I couldn't imagine. Now I have seen it with my own eyes: >>>> >>>>[Event "Chessnet"] >>>>[Site "?"] >>>>[Date "????.??.??"] >>>>[Round "?"] >>>>[White "King 2.55"] >>>>[Black "Junior 5"] >>>>[Result "*"] >>>>[ECO "D49"] >>>>[PlyCount "43"] >>>> >>>>1. d4 Nf6 2. c4 e6 3. Nf3 d5 4. Nc3 c6 5. e3 Nbd7 6. Bd3 dxc4 7. Bxc4 b5 8. Bd3 >>>>a6 9. e4 c5 10. e5 cxd4 11. Nxb5 Nxe5 12. Nxe5 axb5 13. Qf3 Bb4+ 14. Ke2 Rb8 >>>>15. Nc6 Bb7 16. Bf4 Qd7 17. Bxb5 O-O 18. Bxb8 Rxb8 19. Rhc1 Bd6 20. Kf1 Bxc6 >>>>21. Bxc6 Qa7 22. Rc2 * >>>> >>>>Now Junior5 takes the Ph2 and only is down with -1.00. After 23.g3 h5 (-1.32) >>>>24.Kg2 Ng4 (-1.40) 25.Rd1 Rf8 (-1.55) >>>> >>>>Black has some problems getting the bishop back into the game... >>>> >>>>So Uri, what does your latest Junior version (5.9?) play, also take the pawn? >>>> >>>>And I wonder which other programs take this pawn and which avoid this trap? >>> >>>Zarkov takes about 10 milliseconds to switch from Bxh2 to Rc8 :). >>> >>>John >> >> >>Crafty _never_ considers Bxh2: >> >> nss depth time score variation (1) >> 1 0.00 -2.19 22. ... Qc7 >> 1-> 0.00 -2.19 22. ... Qc7 >> 2 0.00 -2.32 22. ... Qc7 23. Qd3 >> 2 0.01 -2.31 22. ... Qa6+ 23. Kg1 d3 >> 2-> 0.01 -2.31 22. ... Qa6+ 23. Kg1 d3 >> 3 0.02 -2.48 22. ... Qa6+ 23. Kg1 Qc8 24. Qd3 >> 3 0.03 -2.31 22. ... Qc7 23. Qd3 Bb4 >> 3 0.03 -2.30 22. ... Qe7 23. Qd3 Be5 >> 3-> 0.03 -2.30 22. ... Qe7 23. Qd3 Be5 >> 4 0.04 -2.39 22. ... Qe7 23. Qd3 Be5 24. Kg1 >> 4 0.06 -2.35 22. ... Qc7 23. Qd3 Rb6 24. Bd5 >> 4-> 0.08 -2.35 22. ... Qc7 23. Qd3 Rb6 24. Bd5 >> 5 0.12 -2.33 22. ... Qc7 23. Qd3 Be5 24. Bd5 Qd6 >> >> >>It has an eval term to stop that nonsense. Many programs don't, still. > >How long did you let it search? This is what I get from 16.18: > > 1 0.00 -2.16 22. ... Qc7 > 1-> 0.03 -2.16 22. ... Qc7 > 2 0.04 -2.48 22. ... Qc7 23. a4 > 2 0.04 -2.31 22. ... Qa6+ 23. Kg1 d3 > 2-> 0.09 -2.31 22. ... Qa6+ 23. Kg1 d3 > 3 0.10 -2.55 22. ... Qa6+ 23. Kg1 d3 24. Rd2 > 3 0.11 -2.28 22. ... Qc7 23. Qd3 Bb4 > 3 0.12 -2.27 22. ... Qe7 23. Qd3 Be5 > 3-> 0.15 -2.27 22. ... Qe7 23. Qd3 Be5 > 4 0.17 -2.59 22. ... Qe7 23. a4 e5 24. b3 > 4 0.20 -2.43 22. ... Qa6+ 23. Kg1 d3 24. Rd2 Rc8 > 4-> 0.23 -2.43 22. ... Qa6+ 23. Kg1 d3 24. Rd2 Rc8 > 5 0.30 -2.21 22. ... Qa6+ 23. Kg1 d3 24. Rd2 Be5 > 25. Rxd3 Rxb2 > 5 0.52 -2.16 22. ... Bxh2 23. g3 Qc7 24. Kg2 Bxg3 > 25. Qxg3 > 5-> 0.53 -2.16 22. ... Bxh2 23. g3 Qc7 24. Kg2 Bxg3 > 25. Qxg3 > 6 0.65 -2.22 22. ... Bxh2 23. g3 Qa6+ 24. Kg2 d3 > 25. Rd2 Rc8 > 6 0.79 -2.21 22. ... Qa6+ 23. Kg1 d3 24. Rd2 Be5 > 25. Rxd3 Rxb2 > 6-> 0.93 -2.21 22. ... Qa6+ 23. Kg1 d3 24. Rd2 Be5 > 25. Rxd3 Rxb2 > 7 1.67 -2.33 22. ... Qa6+ 23. Kg1 Be5 24. Rd2 Rb6 > 25. Be4 Qc4 26. a4 Nxe4 27. Qxe4 > 7 2.31 -2.08 22. ... Bxh2 23. g3 Rd8 24. Qd3 Bxg3 > 25. fxg3 Qe7 > 7-> 2.44 -2.08 22. ... Bxh2 23. g3 Rd8 24. Qd3 Bxg3 > 25. fxg3 Qe7 > 8 3.54 -1.91 22. ... Bxh2 23. g3 Qa6+ 24. Kg2 d3 > 25. Rd2 Bxg3 26. Kxg3 Qc4 > 8-> 4.26 -1.91 22. ... Bxh2 23. g3 Qa6+ 24. Kg2 d3 > 25. Rd2 Bxg3 26. Kxg3 Qc4 > 9 6.68 -2.05 22. ... Bxh2 23. g3 Qa6+ 24. Kg2 d3 > 25. Rd2 Bxg3 26. Kxg3 Qc4 27. a4 > 9-> 16.67 -2.05 22. ... Bxh2 23. g3 Qa6+ 24. Kg2 d3 > 25. Rd2 Bxg3 26. Kxg3 Qc4 27. a4 > 10 20.17 -1.93 22. ... Bxh2 23. g3 Qa6+ 24. Kg2 d3 > 25. Rd2 Bxg3 26. Kxg3 Qc4 27. a4 Rc8 > 10-> 41.77 -1.93 22. ... Bxh2 23. g3 Qa6+ 24. Kg2 d3 > 25. Rd2 Bxg3 26. Kxg3 Qc4 27. a4 Rc8 > >:( > >Jeremiah I let it run in the background for about 30 minutes. It is possible that some of the new eval changes mis-tuned that eval term. The new majority/candidate code has _really_ broken a lot of things and has taken more time than I expected to get it back into semi-reality... The problem here is that Bxh2 gets +1.0, but my trapped bishop code adds in -1.5. It is possible that if you adjust king safety (assuming the king is not castled or is on the kingside) you can swamp that term by king safety exposure. The eval is very highly tuned... and any changes have unexpected side-effects all the time. I will try this on 16.18 to see if I notice anything wrong. Are you using stock evaluation weights???
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.