Author: Peter Berger
Date: 13:56:57 03/22/04
Go up one level in this thread
On March 21, 2004 at 19:14:10, Uri Blass wrote:
>I looked at the PGN and one of the problems is that I am looking too much in
>games instead of working more on the program.
This might well be true. I think your problem as an engine author is also not
lack of awareness and ideas what your program lacks, or what could/should be
improved. When I started to do my little experiments I was also not aware that
there would be that many one-sided results, that could well be interpreted as
bashing which is certainly not my intention at all.
I think I will give movei a break and try some other engine(s) instead.
>These results are clearly bad.
>I waited until movei was lucky to get positive score against Junior5 in the
>nunn2 match at 3:1 time advantage(at least it was the case with 00_8_174 ponder
>off when both sides used 16 mbytes hash and movei got 90 minutes against 30
>minutes of Junior and 178 is the same as 174 except speed if I have no bugs)
>before releasing it but I think now that it was simply lucky and I did not do
>significant improvement in the last year.
I have no clear idea if there was a significant improvement in movei during the
final year. My impression is that there was a very clear one with 0.8.112, but I
am not sure about later versions.
But all this is a double-blind test, where it isn't really known how strong
Movei is, nor how strong its opponents are these days.
Take MChess for example - I think there are quite obviously some pretty nasty
bugs in MCP 8, but how strong is MCP 7.1 on modern hardware? I don't really
know. If I make a guess based on the little data I already produced I am
completely unsure if it is stronger or weaker than Yace, and assume they are
probably very close.
Genius is easy to evaluate of course, as it is well-known that it loses much of
its strength with longer time controls, and Movei is probably about equal to
Genius 2 then ( another guess).
But who would have expected that Shredder gets completely smashed by Movei at 30
0 which is true?
I am trying to get an idea about the old programs mainly (an exercise in
futility maybe ;) ) , and Movei was only used as an example here.
>
>I know that evaluation is one of the weaknesses of movei relative to the top
>programs but my opinion is that I do not need to work first about the evaluation
>but need first to work more about the search and only later to work about
>improving the evaluation.
>
>I believe that it is possible to get 200 elo only by search improvement and
>first I need to rewrite my alphabeta and I still did not decide exactly how to
>do it.
>
Based on posts I read here this is most probably true, if only for practical
reasons. As long as you make major changes to your search your eval will
probably get messed up in the process anyway.
But I also remember that Amir Ban posted that his search is still pretty much
the same as with Junior 5, and all recent improvements are based on eval. And
Junior 8's games are definitely the most interesting to watch of modern
commercials.
And watching MChess, that gets outsearched often ( although its "depth" is
definitely an understatement most of the time) also gives the impression that
its eval is heavily tuned. It's the program that looks most "human" to me
currently.
I hope you didn't take offense with my little match reports.
Peter
PS: Another nice game from today:
[Event "Computer chess game"]
[Site "P418"]
[Date "2004.03.22"]
[Round "4"]
[White "MChess 7.1"]
[Black "movei00_8_178"]
[Result "1-0"]
[TimeControl "7200"]
1. e4 c6 2. d4 d6 3. Nc3 Nf6 4. Nf3 Bg4 5. h3 Bxf3 6. Qxf3 Qb6 7. e5 Nfd7
8. exd6 Qxd4 9. Nb5 Qe5+ 10. Be3 Na6 11. O-O-O Qf6 12. Qg4 Ne5 13. d7+ Kd8
14. Qa4 cxb5 15. Qxb5 Nc7 16. Qxb7 Qc6 17. Qb3 e6 18. Be2 Qxg2 19. Rhg1 Qc6
20. Bxa7 Nxd7 21. Bf3 Nd5 22. Bxd5 exd5 23. Rxd5 Kc8 24. Be3 Rb8 25. Qd3 h5
26. Rd1 Rb7 27. Ra5 Be7 28. Ra6 Qb5 29. Qxb5 Rxb5 30. Rc6+ Kd8 31. Bb6+
Rxb6 32. Rxb6 Bg5+ 33. Kb1 Kc7 34. Rb5 Bf6 35. Rf5 h4 36. a4 Re8 37. a5 Re5
38. Rd5 Rxd5 39. Rxd5 Nf8 40. Rb5 Bd4 41. a6 Bb6 42. c4 Kc6 43. Rf5 f6 44.
b4 Ne6 45. f3 Bf2 46. Rd5 g6 47. Kc2 Ba7 48. Kb3 Bb6 49. Rd1 Nd4+ 50. Ka4
Ba7 51. c5 Nb5 52. Rd8 f5 53. f4 Nc3+ 54. Ka5 Nb5 55. Ra8 Bxc5 56. Rc8+ Kd7
57. Rxc5 Na7 58. Kb6 g5 59. fxg5 Ke6 60. Kxa7 Kf7 61. Rc6 f4 62. Rf6+ Kg7
63. Kb7 f3 64. a7 f2 65. a8=Q f1=Q 66. Qf8+ Kh7 67. g6#
{Movei->White mates} 1-0
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.