need help in testing

[Uri Blass]

Geschrieben von:/Posted by: Uri Blass at 01 April 2004 23:16:03:

I am not sure if to ask Leo to use a previous version of movei in his tournament inspite of the good result against Ufim.
I tested movei00_8_178d against movei00_8_174(ponder off and 40 minutes/4 moves and it lost 30-20(20-10 with 10 draws)
The surprising thing is that it was leading 4-0 with 4 draws after the first 8 games.
178d is almost the same as movei00_8_178c that I did not send to Leo so I cannot use it and the only difference is that I changed slightly the value of the pawns(reduced most of the pieces square table by 0.03) because I remember few games when movei sacrificed a piece for pawns without justification in the middle game and I thought that it is not going to change the strength significantly so it cannot explain the 30-20.
If people can help me in testing 00_8_174 and 00_8_178 and 00_8_178d against different opponents to decide if to go back to previous version it may be productive

A possible change from reducing the value of pawns by 0.03 may make movei more careful in pruning because it use the evaluation to prune but I do not think that it should change a lot.
Note that previous version that I did not send to Leo is movei00_8_178c and that version beated 174 but I used faster time control of 1/40

Here is a game when movei sacrificed a piece without justification
(I checked and 00_8_178 also fail high for Nxf2 at depth 14 and I consider this move to be wrong because black has the advantage with Nxe1 and movei won only because Hagrid blundered later)
This game encouraged me to see reducing the value of pawn as a step in the right direction(this is not the only example and I remember another case when it happened but it does not happen very often)
[Event "WBEC6 2nd Division"]
[Site "ATHLON-MP2200"]
[Date "2004.01.11"]
[Round "15.3"]
[White "Hagrid 0.7.56"]
[Black "Movei 0.08.131"]
[Result "0-1"]
[PlyCount "132"]
[EventDate "2004.??.??"]
[TimeControl "40/2400:0"]
1. d4 f5 2. g3 e6 3. Nf3 d5 4. Bg2 Nc6 5. Bf4 Bd6 6. Bxd6 Qxd6 7. Nc3 Nf6 8.
O-O Ne4 9. Nb5 Qd7 10. Nh4 a6 11. Na3 Qe7 12. Bxe4 fxe4 13. c4 O-O 14. e3 Bd7
15. Ng2 Nb4 16. Qb3 c6 17. Nf4 g5 18. Ng2 b5 19. c5 Nd3 20. Ne1 Nxf2 21. Rxf2
Rxf2 22. Kxf2 Qf7+ 23. Ke2 e5 24. Kd1 Qh5+ 25. Kc1 Qxh2 26. dxe5 Qxg3 27. Qc3
b4 28. Qxb4 Qxe3+ 29. Kc2 Qf2+ 30. Kb3 a5 31. Qc3 Be6 32. Ka4 d4 33. Qc2 Qf4
34. Nc4 d3 35. Qc3 e3 36. Ng2 Qxc4+ 37. Qxc4 Bxc4 38. Nxe3 Bb5+ 39. Kb3 Re8 40.
Ng4 Kg7 41. a4 Ba6 42. Rh1 Bc8 43. Nf2 Rxe5 44. Nxd3 Rd5 45. Nc1 Be6 46. Ka3
Rxc5 47. Re1 Kf6 48. Nd3 Rd5 49. Re3 g4 50. b3 Rg5 51. Nf4 Bd5 52. Re1 Rf5 53.
Ne2 Rf3 54. Kb2 Rxb3+ 55. Kc2 Rb4 56. Nc3 h5 57. Kd3 g3 58. Rf1+ Ke5 59. Re1+
Kf4 60. Re7 g2 61. Rg7 Kf3 62. Rg6 Be6 63. Rf6+ Rf4 64. Rxf4+ Kxf4 65. Ne2+ Kf3
66. Nd4+ Kf2 {White resigns} 0-1

[Tom Likens]

Geschrieben von:/Posted by: Tom Likens at 01 April 2004 23:43:18:
Als Antwort auf:/In reply to: need help in testing geschrieben von:/posted by: Uri Blass at 01 April 2004 23:16:03:

I am not sure if to ask Leo to use a previous version of movei in his >tournament inspite of the good result against Ufim.

Hey Uri,
Unfortunately, I'm swamped and can't help out with testing. But I can
offer a bit of unsolicited advice on the subject. Be *very* careful about
sending Leo a last minute "improved" replacement for your engine. I did
this previously and managed to get relegated for my efforts. I also
suspect Tony did something similar for Xinix (and for his efforts I suspect
he will also get relegated).
It's better, IMHO, to send Leo a version that has been throughly tested,
even if it *may* be slightly weaker than a new and improved untested version.
Leo's tournament is high profile and runs for a long time, so if you do
succumb to this temptation, the pain will last for quite awhile.
good luck with your decision,
--tom

[Uri Blass]

Geschrieben von:/Posted by: Uri Blass at 01 April 2004 23:59:49:
Als Antwort auf:/In reply to: Re: need help in testing geschrieben von:/posted by: Tom Likens at 01 April 2004 23:43:18:

I am not sure if to ask Leo to use a previous version of movei in his >tournament inspite of the good result against Ufim.

Hey Uri,
Unfortunately, I'm swamped and can't help out with testing. But I can
offer a bit of unsolicited advice on the subject. Be *very* careful about
sending Leo a last minute "improved" replacement for your engine. I did
this previously and managed to get relegated for my efforts. I also
suspect Tony did something similar for Xinix (and for his efforts I suspect
he will also get relegated).
It's better, IMHO, to send Leo a version that has been throughly tested,
even if it *may* be slightly weaker than a new and improved untested version.
Leo's tournament is high profile and runs for a long time, so if you do
succumb to this temptation, the pain will last for quite awhile.
good luck with your decision,
--tom

I only did some small changes in endgames in the evaluation from 00_8_178(except reducing slightly the value of pawns.
I used 00_8_194 and 00_8_195 in leo's test tournament.
I have a newer version than 00_8_178d that played in the test tournament and I decided not to trust it after some bad results in my tests with 00_8_195 against 00_8_174 and loss on time that I could not reproduce.
Note that when I think about it the result of reducing the evaluation of pawns may be slightly less pruning because I use pruning based on evaluation.
I remember that Quark2.35 had bad results in games against Quark2.05 and good result against other programs so I also doubt if to trust results against previous version and it is better to compare results based on the nunn matches
with other programs.
Uri