Bestia 0.90 vs Ufim 3.02 (1) 120'/40 (1/2) Commented gam

[Kurt Utzinger]

Geschrieben von: / Posted by: Kurt Utzinger at 04 March 2003 23:22:28:

Below the first commented game of this match. Details of playing conditions included in the pgn-file:
Kurt & Rolf Chess
Bestia 0.90 vs Ufim 3.02 [1]
Position after 16...Qxe4 and before 17.Qxe4?
 abcdefgh  8      8  7        7  6        6  5        5  4         4  3         3  2      2  1     1  abcdefgh ChessDiag V1.01 (12-OCT-2002)r1b1k2r/2p4p/p5p1/1p1p4/4q3/2P5/PP2Q1P1/RNB2RK1 w kq - 0 17
[Event "Bestia090_Ufin302 P3 650/32 120'/40"]
[Site "Switzerland"]
[Date "2003.03.03"]
[Round "?"]
[White "Bestia_090"]
[Black "Ufim 3.02"]
[Result "1/2-1/2"]
[ECO "C60"]
[PlyCount "122"]
1. e4 {Comments: Kurt Utzinger, Switzerland, engine/engine match under
Winboard 4.2.5, ponder=off, both engines with owns books, played on P3 650/32
MB hash at time control 120'/40} 1... e5 2. Nf3 Nc6 3. Bb5 g6 4. O-O Bg7 5. c3
a6 {Black last book move} 6. Ba4 {White last book move} 6... Nf6 {
The most common moves here are 6...d6 or 6...b5} 7. d4 Nxe4 8. dxe5 Nxe5 {
(?) A wrong decision. Ufim 3.02 has probably not looked deep enough in this
position and was therefore not aware of the danger on the half-open e-file.} 9.
Qe2 {The alternative 9.Re2 seems to be even somewhat stronger.} 9... f5 {
Maybe the best chance. After 9...Nxf3 10.gxf3 black loses a piece.} 10. Nxe5
Bxe5 11. f3 b5 12. Bc2 d5 13. fxe4 fxe4 {
A bit more resistance was possible by 13...dxe4} 14. Bxe4 Bxh2+ 15. Kxh2 Qh4+
16. Kg1 Qxe4 {After the 14th move, Ufim 3.02 has done his best, but this is
not sufficient to save the game.} 17. Qxe4+ {(?!) Since black's king position
is rather weak, it was much more promising to keep the queens on the board and
to go for a deadly attack with 17.Qf2. Bestia 0.90 has surely calculated that
it can also win black's centre pawn and may for this reason have preferred the
exchange of queens. It was hard to already see here that black will entry on
the first rank with his rook when white's pieces like the bishop c1 and the
rook a1 are rather paralyzed.} 17... dxe4 18. Nd2 Bb7 19. Nxe4 {(?) As the awn
cannot run away, white should rather have done something for his development
by 19.Nb3 or 19.a4} 19... Bxe4 20. Re1 O-O-O 21. Rxe4 Rd1+ 22. Kh2 Rf8 23. Rd4
{Stronger 23.Bh6} 23... Re1 24. Rd2 {And again was 24.Bh6 the better solution.}
24... Rff1 {At once, the situation is no longer so clear and white is damned
for the defence.} 25. Rc2 g5 26. c4 b4 {
Just wright, black should try to keep the position closed.} 27. c5 c6 28. a3 b3
{And again the correct move. White's advantage has gone to the wind.} 29. Rc3
Rh1+ 30. Kg3 h5 31. Rxb3 {If white wants to win, this is the only try.} 31...
Rxc1 32. Rxc1 Rxc1 33. Rb6 Rxc5 34. Rxa6 Kb7 {And now, we have an equal rook
ending on the board. Both programs play the following phase rather well and
the game ends with a draw.} 35. Ra4 Rb5 36. b4 c5 37. bxc5 Rxc5 38. Rb4+ Ka6
39. Rb8 Ka5 40. Rh8 Rc3+ 41. Kf2 h4 42. Rg8 Rg3 43. Rf8 Kb5 44. Rf3 Rg4 {
(!) Of course not 44...Rxf3 45.Kxf3 +-} 45. Rf5+ Ka6 46. Rf6+ Ka7 47. Rc6 Ra4
48. Rc3 g4 49. Ke1 Kb6 50. Ke2 Kb5 51. Rb3+ Kc6 52. Rb4 Rxa3 53. Rxg4 h3 54.
gxh3 Rxh3 55. Rg5 Kd6 56. Rg6+ Ke5 57. Rg7 Rb3 58. Rg8 Ra3 59. Rf8 Ke4 60. Re8+
Kd4 61. Rd8+ Ke5 1/2-1/2

[Thomas Mayer]

Geschrieben von: / Posted by: Thomas Mayer at 05 March 2003 00:57:11:
Als Antwort auf: / In reply to: Bestia 0.90 vs Ufim 3.02 (1) 120'/40 (1/2) Commented game geschrieben von: / posted by: Kurt Utzinger at 04 March 2003 23:22:28:

Hi Kurt,
interesting game (and - as usual - great comments) with it's up and downs for both sides... the position you showed below is maybe a good testposition for king safety and attack abilities of engines... I let it run through several (wb)engines:
Bestia 0.90 vs Ufim 3.02 [1]
Position after 16...Qxe4 and before 17.Qxe4?
 abcdefgh  8      8  7        7  6        6  5        5  4         4  3         3  2      2  1     1  abcdefgh ChessDiag V1.01 (12-OCT-2002)r1b1k2r/2p4p/p5p1/1p1p4/4q3/2P5/PP2Q1P1/RNB2RK1 w kq - 0 17
Some results:
Quark v1.95 switches to Qf2 after around 25 seconds - the scoure jump from +2.08 to +2.36 indicates that it might stay on that move the next few plys...
Deep Sjeng chooses instantly Qf2 with a score of +3.66 growing over 4 after some seconds...
Gandalf 4.32h needs around 50 seconds to switch to Qf2 with a score of +2.92 after opening the fail high (around 2 minutes)
LambChop v10.88 is initially on Qf2 with a growing score to over +3 after 20 seconds...
Phalanx 22 is also initially on Qf2 but changes to Qxe4 after 22 seconds... (+2.67) - After a minute it is back on Qf2 with +3.07
Crafty 18.13 (no idea why I have this one installed... seems to like Qd2? at most at the beginning (+3.07) - after 22 seconds it switches to Qf2 with +3.58
Most engines find the move if not initially then around ply 11... maybe the two contestant Bestia and Ufim did not get deep enough ?
Greets, Thomas

[Andreas Herrmann]

Geschrieben von: / Posted by: Andreas Herrmann at 05 March 2003 01:46:28:
Als Antwort auf: / In reply to: Re: Bestia 0.90 vs Ufim 3.02 (1) 120'/40 (1/2) Commented game geschrieben von: / posted by: Thomas Mayer at 05 March 2003 00:57:11:

Hi Kurt,
interesting game (and - as usual - great comments) with it's up and downs for both sides... the position you showed below is maybe a good testposition for king safety and attack abilities of engines... I let it run through several (wb)engines:
Bestia 0.90 vs Ufim 3.02 [1]
Position after 16...Qxe4 and before 17.Qxe4?
 abcdefgh  8      8  7        7  6        6  5        5  4         4  3         3  2      2  1     1  abcdefgh ChessDiag V1.01 (12-OCT-2002)r1b1k2r/2p4p/p5p1/1p1p4/4q3/2P5/PP2Q1P1/RNB2RK1 w kq - 0 17
Some results:
Quark v1.95 switches to Qf2 after around 25 seconds - the scoure jump from +2.08 to +2.36 indicates that it might stay on that move the next few plys...
Deep Sjeng chooses instantly Qf2 with a score of +3.66 growing over 4 after some seconds...
Gandalf 4.32h needs around 50 seconds to switch to Qf2 with a score of +2.92 after opening the fail high (around 2 minutes)
LambChop v10.88 is initially on Qf2 with a growing score to over +3 after 20 seconds...
Phalanx 22 is also initially on Qf2 but changes to Qxe4 after 22 seconds... (+2.67) - After a minute it is back on Qf2 with +3.07
Crafty 18.13 (no idea why I have this one installed... seems to like Qd2? at most at the beginning (+3.07) - after 22 seconds it switches to Qf2 with +3.58
Most engines find the move if not initially then around ply 11... maybe the two contestant Bestia and Ufim did not get deep enough ?
Greets, Thomas

Holmes 0.8.20h on a slow AMD K6/2-450 in 0,93 seconds:

analyze
hash tables cleared
stat. eval: 1,58 (mat. balance: 1,00 pos. val: 0,58)
1 -776 0 16 17.c4 {0:00}
1 151 0 30 17.Qd1 {0:00}
1 153 0 43 17.Qf3 {0:00}
1 171 0 87 17.Be3 {0:00}
2 171 1 158 17.Be3 {? 16kNs 0:00}
2 85 1 225 17.Be3 Rf8 {22kNs 0:00}
2 150 1 394 17.Qxe4+ dxe4 18.Bg5 {39kNs 0:00}
3 111 2 893 17.Qxe4+ dxe4 18.Nd2 Bf5 {45kNs 0:00}
3 123 3 1327 17.Be3 Rf8 18.Rf3 {44kNs 0:00}
4 123 7 5445 17.Be3 Rf8 18.Rf3 {? 78kNs 0:00}
4 47 8 6040 17.Be3 Bg4 18.Qd2 {76kNs 0:00}
4 131 10 8437 17.Qxe4+ dxe4 {84kNs 0:00}
5 120 28 21408 17.Qxe4+ dxe4 18.Bh6 Bf5 19.Bg7 Rg8 {76kNs 0:00}
5 130 59 59330 17.Qd2 Be6 18.Qg5 Rf8 19.Rf4 {101kNs 0:00}
5 259 93 93806 17.Qf2 Be6 18.Re1 Qg4 19.Bg5 {101kNs 0:00}
6 271 142 150842 17.Qf2 Be6 18.Re1 Qg4 19.Qf6 Kd7 {106kNs 0:01}
7 302 273 295508 17.Qf2 Kd7 18.Bf4 Rf8 19.Nd2 Qc2 20.Qh4 {108kNs 0:02}
8 256 601 669365 17.Qf2 Kd7 18.Qc5 c6 19.Rf7+ Ke8 20.Rf1 Kd7 {111kNs 0:06}
9 284 1920 2144472 17.Qf2 Kd7 18.Re1 Rf8 19.Qg3 Qc4 20.Bh6 Qc5+ 21.Kh2 Re8 {112kNs 0:19}
10 255 9425 10552975 17.Qf2 Be6 18.Re1 Qg4 19.Qf6 Kd7 20.Qg7+ Kc6 21.Qe5 Rae8 22.Be3 {112kNs 1:34}

have a nice day
Andreas

[Thomas Mayer]

Geschrieben von: / Posted by: Thomas Mayer at 05 March 2003 01:56:10:
Als Antwort auf: / In reply to: Re: Bestia 0.90 vs Ufim 3.02 (1) 120'/40 (1/2) Commented game geschrieben von: / posted by: Andreas Herrmann at 05 March 2003 01:46:28:

Hi Andreas,

Holmes 0.8.20h on a slow AMD K6/2-450 in 0,93 seconds:

well, I was quite sure about that - with your high scores for controls around opponent king... Did you solve the problem that occurs against Quark in Paderborn concerning that ?
Greets, Thomas
P.S.: Besides, any chance to get current Holmes for me ? As a non beta-tester ? (But for sure I might play some testgames and if something strange occurs like the Qxf4 in Paderborn (was it f4 ?) I will inform you...

[Andreas Herrmann]

Geschrieben von: / Posted by: Andreas Herrmann at 05 March 2003 02:21:15:
Als Antwort auf: / In reply to: Re: Bestia 0.90 vs Ufim 3.02 (1) 120'/40 (1/2) Commented game geschrieben von: / posted by: Thomas Mayer at 05 March 2003 01:56:10:

Hi Andreas,

Holmes 0.8.20h on a slow AMD K6/2-450 in 0,93 seconds:

well, I was quite sure about that - with your high scores for controls around opponent king... Did you solve the problem that occurs against Quark in Paderborn concerning that ?
Greets, Thomas
P.S.: Besides, any chance to get current Holmes for me ? As a non beta-tester ? (But for sure I might play some testgames and if something strange occurs like the Qxf4 in Paderborn (was it f4 ?) I will inform you...

Yes, i have changed here a lot. But i'm not sure if it's better than before.
Some changes do better in test suites and some do not.
Last night a current version looses against 0.8.13g (the version that has playd in Leos tournament) 28,0:38,0. So the current version seems much weaker than the Paderborn version 0.8.19b. Perhaps one or more heavy new bugs in the current version.
So at the moment i havn't a stabil version. There are a lot changes to test and further a lot to fix, that i have recognized in Paderborn.

I havn't tested the position of our game in Paderborn so far, but i will do tomorrow or so.
I will send you later a current version without book, because the book size is about 19 MB. But you can create a own book from a PGN file.
Insert "help" or "help createbook" in textmode to see how to create a book.
best wishes
Andreas
PS: I hope we see us again in Leiden.

[Thomas Mayer]

Geschrieben von: / Posted by: Thomas Mayer at 05 March 2003 08:55:23:
Als Antwort auf: / In reply to: Re: Bestia 0.90 vs Ufim 3.02 (1) 120'/40 (1/2) Commented game geschrieben von: / posted by: Andreas Herrmann at 05 March 2003 02:21:15:

Hi Andreas,

Yes, i have changed here a lot. But i'm not sure if it's better than before.
Some changes do better in test suites and some do not.
Last night a current version looses against 0.8.13g (the version that has
playd in Leos tournament) 28,0:38,0. So the current version seems much
weaker than the Paderborn version 0.8.19b. Perhaps one or more heavy new
bugs in the current version.
PS: I hope we see us again in Leiden.

business as usual... funny enough, I have changed just some stuff with the nullmove handling in Quark and at least in blitz it seems about 50 ELO stronger... Now I nead another 6 such funny ideas and Quark is ready for the SSDF...
That is the amazing thing about - it seems to work everywhere - lctii is better, wac is better, ecm-gcp is better, all endgame testsuites where I had my biggest doubts that it works are better... really funny... only if it helps in longer games is pretty unsure... And that at the moment where I have start the rewrite - so the old source keeps my attention...
at the moment I use the idea of Rudolf he told me at Paderborn to test - I let it just play overnight 200 blitz games against a variety of opponents (takes around 10 hours) - if that shows significant improvements I say "Aha !" but before that I let it run through several testsuites (wac to look if something is broken, ecm-gcp to test the tactical ability, lctii with 3 mins, QuickTest and Pet_v1 are my favorites...)
That is still planned - also the computer thingy will be fine with me if you need one... I am waiting for the dates of the tourney...
Greets, Thomas

[Uri Blass]

Geschrieben von: / Posted by: Uri Blass at 05 March 2003 10:04:59:
Als Antwort auf: / In reply to: Re: Bestia 0.90 vs Ufim 3.02 (1) 120'/40 (1/2) Commented game geschrieben von: / posted by: Thomas Mayer at 05 March 2003 08:55:23:

Hi Andreas,

Yes, i have changed here a lot. But i'm not sure if it's better than before.
Some changes do better in test suites and some do not.
Last night a current version looses against 0.8.13g (the version that has
playd in Leos tournament) 28,0:38,0. So the current version seems much
weaker than the Paderborn version 0.8.19b. Perhaps one or more heavy new
bugs in the current version.

business as usual... funny enough, I have changed just some stuff with the nullmove handling in Quark and at least in blitz it seems about 50 ELO stronger... Now I nead another 6 such funny ideas and Quark is ready for the SSDF...
That is the amazing thing about - it seems to work everywhere - lctii is better, wac is better, ecm-gcp is better, all endgame testsuites where I had my biggest doubts that it works are better... really funny... only if it helps in longer games is pretty unsure... And that at the moment where I have start the rewrite - so the old source keeps my attention...
at the moment I use the idea of Rudolf he told me at Paderborn to test - I let it just play overnight 200 blitz games against a variety of opponents (takes around 10 hours) - if that shows significant improvements I say "Aha !"
but before that I let it run through several testsuites (wac to look if >something is broken, ecm-gcp to test the tactical ability, lctii with 3 mins, >QuickTest and Pet_v1 are my favorites...)

I believe that most good ideas for test suites are also good for game unless
you try to do on purpose something that is only good for test suites.
I do not know how to do it.
I only know to play automatically a match against one opponent.
There are beta testers who give me some results of movei against different opponents but more computer time will help because I use my computer for correspondence games so if somebody is interested in testing latest movei in the test suites that you tested quark(except the GCP test suite) then it may be productive for me.
It may be interesting to know your results.
I used 5 minutes per position in the GCP test suite to see if there are problems at long time control(latest movei got 156 solutions right on AMD1000Mhz).
Do you use a special utility for test suites like epd2wb or do you use a program that you wrote to read epd files?
Uri

[Thomas Mayer]

Geschrieben von: / Posted by: Thomas Mayer at 05 March 2003 10:36:25:
Als Antwort auf: / In reply to: Re: Bestia 0.90 vs Ufim 3.02 (1) 120'/40 (1/2) Commented game geschrieben von: / posted by: Uri Blass at 05 March 2003 10:04:59:

Hi Uri,

at the moment I use the idea of Rudolf he told me at Paderborn to test - I
let it just play overnight 200 blitz games against a variety of opponents
(takes around 10 hours) - if that shows significant improvements I
say "Aha !"
but before that I let it run through several testsuites (wac to look if
something is broken, ecm-gcp to test the tactical ability, lctii with 3
mins,
QuickTest and Pet_v1 are my favorites...)

I do not know how to do it.
I only know to play automatically a match against one opponent.
There are beta testers who give me some results of movei against different
opponents but more computer time will help because I use my computer for
correspondence games so if somebody is interested in testing latest movei in
the test suites that you tested quark(except the GCP test suite) then it may
be productive for me.
It may be interesting to know your results.
I used 5 minutes per position in the GCP test suite to see if there are
problems at long time control(latest movei got 156 solutions right on
AMD1000Mhz).
Do you use a special utility for test suites like epd2wb or do you use a
program that you wrote to read epd files?

Well, to run a gauntlet is quite easy, you can use a batch file like Dieter is doing (he describes it here: http://f11.parsimony.net/forum16635/messages/43966.htm) or you might use one of the winboard tournament managers around like Jori's WBTM or Galis WBTM...
no prob... all on Athlon MP1800+:
WAC 5 sec: 293/300
ECM-GCP 10 sec: 104/183
ECM-GCP 60 sec: 143/183 (Quark is still not a fantastic tactician but it is growing)
PET 30 sec: 26/50 (still bad)
LCTII 3 min: 23/35 (especially speeded up there)
QuickTest 60 sec: 14/24 (with big speedup)
and some more... for fun sometimes WCSAC in 1 sec, here it solves 799/1001
puuh... 5 mins... that will take a hole day... maybe I will try that once overnight...
I use my own version of epd2wb which has some different features... and I use a batch file to run multiple tests in a row... Quark itself has a builtin testsuite command - but it only understands bestmoves and not avoid moves - and has several different problems - after I saw epd2wb I stopped to implement that in my engine...
Greets, Thomas

[Uri Blass]

Geschrieben von: / Posted by: Uri Blass at 05 March 2003 11:06:18:
Als Antwort auf: / In reply to: Re: Bestia 0.90 vs Ufim 3.02 (1) 120'/40 (1/2) Commented game geschrieben von: / posted by: Thomas Mayer at 05 March 2003 10:36:25:

Hi Uri,

at the moment I use the idea of Rudolf he told me at Paderborn to test - I
let it just play overnight 200 blitz games against a variety of opponents
(takes around 10 hours) - if that shows significant improvements I
say "Aha !"
but before that I let it run through several testsuites (wac to look if
something is broken, ecm-gcp to test the tactical ability, lctii with 3
mins,
QuickTest and Pet_v1 are my favorites...)

I do not know how to do it.
I only know to play automatically a match against one opponent.
There are beta testers who give me some results of movei against different
opponents but more computer time will help because I use my computer for
correspondence games so if somebody is interested in testing latest movei in
the test suites that you tested quark(except the GCP test suite) then it may
be productive for me.
It may be interesting to know your results.
I used 5 minutes per position in the GCP test suite to see if there are
problems at long time control(latest movei got 156 solutions right on
AMD1000Mhz).
Do you use a special utility for test suites like epd2wb or do you use a
program that you wrote to read epd files?

Well, to run a gauntlet is quite easy, you can use a batch file like Dieter is doing (he describes it here: http://f11.parsimony.net/forum16635/messages/43966.htm) or you might use one of the winboard tournament managers around like Jori's WBTM or Galis WBTM...
no prob... all on Athlon MP1800+:
WAC 5 sec: 293/300
ECM-GCP 10 sec: 104/183
ECM-GCP 60 sec: 143/183 (Quark is still not a fantastic tactician but it is growing)
LCTII 3 min: 23/35 (especially speeded up there)
QuickTest 60 sec: 14/24 (with big speedup)
and some more... for fun sometimes WCSAC in 1 sec, here it solves 799/1001
puuh... 5 mins... that will take a hole day... maybe I will try that once overnight...
I use my own version of epd2wb which has some different features... and I use a batch file to run multiple tests in a row... Quark itself has a builtin testsuite command - but it only understands bestmoves and not avoid moves -

Yes, it is still not close to knightdreamer's results on 733 Mhz but I guess the same is for most of the 1st division engines.

>PET 30 sec: 26/50 (still bad)
You can use 3 minutes to do it in less than 10 hours.
I used only 1000 Mhz and your athlon1800 is probably significantly faster.
This is also a problem with epd2wb(I know that there is another application but I did not try it).
Note that it is possible to translate every am to bm(possibly with more than one move) in order to avoid problems.
Uri

[Thomas Mayer]

Geschrieben von: / Posted by: Thomas Mayer at 05 March 2003 11:27:30:
Als Antwort auf: / In reply to: Re: Bestia 0.90 vs Ufim 3.02 (1) 120'/40 (1/2) Commented game geschrieben von: / posted by: Uri Blass at 05 March 2003 11:06:18:

Hi Uri,

I use my own version of epd2wb which has some different features... and I
use a batch file to run multiple tests in a row... Quark itself has a
builtin testsuite command - but it only understands bestmoves and not avoid
moves -

This is also a problem with epd2wb(I know that there is another application
but I did not try it). Note that it is possible to translate every am to bm
(possibly with more than one move) in order to avoid problems.

it's not a problem with my version of epd2wb - it handles the avoid moves just fine and is compatible or adjustable to many more engines then the one of Bruce... at least on my PCs -> Gabor had some troubles with some engines which was never reproduceable here...
Greets, Thomas

[Andreas Herrmann]

Geschrieben von: / Posted by: Andreas Herrmann at 05 March 2003 13:21:24:
Als Antwort auf: / In reply to: Re: Bestia 0.90 vs Ufim 3.02 (1) 120'/40 (1/2) Commented game geschrieben von: / posted by: Thomas Mayer at 05 March 2003 08:55:23:

Hi Andreas,

Yes, i have changed here a lot. But i'm not sure if it's better than before.
Some changes do better in test suites and some do not.
Last night a current version looses against 0.8.13g (the version that has
playd in Leos tournament) 28,0:38,0. So the current version seems much
weaker than the Paderborn version 0.8.19b. Perhaps one or more heavy new
bugs in the current version.
PS: I hope we see us again in Leiden.

business as usual... funny enough, I have changed just some stuff with the nullmove handling in Quark and at least in blitz it seems about 50 ELO stronger... Now I nead another 6 such funny ideas and Quark is ready for the SSDF...
That is the amazing thing about - it seems to work everywhere - lctii is better, wac is better, ecm-gcp is better, all endgame testsuites where I had my biggest doubts that it works are better... really funny... only if it helps in longer games is pretty unsure... And that at the moment where I have start the rewrite - so the old source keeps my attention...
at the moment I use the idea of Rudolf he told me at Paderborn to test - I let it just play overnight 200 blitz games against a variety of opponents (takes around 10 hours) - if that shows significant improvements I say "Aha !" but before that I let it run through several testsuites (wac to look if something is broken, ecm-gcp to test the tactical ability, lctii with 3 mins, QuickTest and Pet_v1 are my favorites...)
That is still planned - also the computer thingy will be fine with me if you need one... I am waiting for the dates of the tourney...
Greets, Thomas

Hi Thomas,
realy impressive. Congratulation 50 ELO at your level is a lot.
Andreas

[Uri Blass]

Geschrieben von: / Posted by: Uri Blass at 05 March 2003 15:46:13:
Als Antwort auf: / In reply to: Re: Bestia 0.90 vs Ufim 3.02 (1) 120'/40 (1/2) Commented game geschrieben von: / posted by: Thomas Mayer at 05 March 2003 11:27:30:

Hi Uri,

I use my own version of epd2wb which has some different features... and I
use a batch file to run multiple tests in a row... Quark itself has a
builtin testsuite command - but it only understands bestmoves and not avoid
moves -

This is also a problem with epd2wb(I know that there is another application
but I did not try it). Note that it is possible to translate every am to bm
(possibly with more than one move) in order to avoid problems.

it's not a problem with my version of epd2wb - it handles the avoid moves just fine and is compatible or adjustable to many more engines then the one of Bruce... at least on my PCs -> Gabor had some troubles with some engines which was never reproduceable here...
Greets, Thomas

Did you change the source code of Bruce to do it better?
I did some comparison between movei results and Quark results in
the GCP test suite and your results
You had 104 solution in 10 seconds when Movei need 11-12 seconds to get the same number of solution but it is a win for Movei when I consider the hardware difference.
You has 143 solutions in 60 seconds when Movei needed 113-114 seconds to get the same number of solutions.
It is not clear who is the winner based on these results.
I guess that the data suggests that quark may be the stronger tactical
engine at long time control but the data is not enough to
give evidence for it(movei also could do slightly better in case of shorter time control because there are cases when it choose the right move only to change it's mind later).
Here are the exact results of movei in a table that was generated by epd2wb
Results:

[Thomas Mayer]

Geschrieben von: / Posted by: Thomas Mayer at 05 March 2003 16:53:32:
Als Antwort auf: / In reply to: Re: Bestia 0.90 vs Ufim 3.02 (1) 120'/40 (1/2) Commented game geschrieben von: / posted by: Uri Blass at 05 March 2003 15:46:13:

Hi Uri,

Did you change the source code of Bruce to do it better?
You had 104 solution in 10 seconds when Movei need 11-12 seconds to get the
same number of solution but it is a win for Movei when I consider the
hardware difference.
You has 143 solutions in 60 seconds when Movei needed 113-114 seconds to get
the same number of solutions.

yes - if you are interested, I can send it to you...
the output looks also a little bit different... so you get e.g. the last eval and last line for each position (which is good to find errors in the testsuite - when your engine does not find the correct solution but reports a mate anyway, there might be something wrong in the testsuite...) also the last result table looks different... an example:


part of the position overview:
159 no [--:--:--,--] [10] +0.18 f1e1 h7h5 c7d6 f8e7 d6e7 e8e7 b5d6
c8d8 g3g5 d5d4
160 yes 12 [00:00:15,79] [14] +2.38 d2b2 a4b2 c4c3 b7b6 c5c4 b6b4 a7a5
b4b5 c3c2 g1f2 c2c1q f2e2 c1b1 a2a4 c4c3
161 yes 7 [00:00:00,34] [14] +3.53 c5b6 d5b5 b1b5 a7b6 c2c4 e6e7 b5b6
e7a7 c4c5 g7g6 c5c6 g8g7 f2f4 f7f6
162 yes 5 [00:00:00,06] [12] +3.64 f4f5 g6g5 e3e8 d7e8 c3g7 g5g7 f5f6
g7f8 d5c6 e8f7 c6b7 f8h6 b2c3 h6h5 c3d4 h5e5
d4f2
163 yes 11 [00:00:01,61] [16] +4.01 d5b3 a2b3 c5c4 d3d4 c4c3 b3a4 b4b3
g5h5 b3b2 d4b4 c3c2 b4b2 c2c1q b2f2 c1a1 f2f4
a1h8 h5g5 c6c5
you see that you also get the depth at which the program finds the solution (number in front of the time)
part of the results-table:
13 1 110 85
14 1 111 44
16 4 115 34 46 142 160
18 1 116 61
19 2 118 8 36
20 1 119 17
21 1 120 83
22 1 121 177
23 1 122 126
24 1 123 118
25 2 125 137 183
26 1 126 92
you can see that the positions are mentioned after the number of solutions... seconds without a solution are not written anymore... (which makes a long test much more readable)
as I said, Movei is very good in tactics... with Quark it is different...
when you compare the +-rating of Athlons and PIII or other Athlons you should look at the real clock speed, the MP1800+ is 1533 MHz, so you might look at 90 seconds - might be that Quark gets faster to depth then Movei without loosing to much in fast time controls - but helps in longer terms...
Greets, Thomas

[Uri Blass]

Geschrieben von: / Posted by: Uri Blass at 05 March 2003 17:11:52:
Als Antwort auf: / In reply to: Re: Bestia 0.90 vs Ufim 3.02 (1) 120'/40 (1/2) Commented game geschrieben von: / posted by: Thomas Mayer at 05 March 2003 16:53:32:

Hi Uri,

Did you change the source code of Bruce to do it better?
You had 104 solution in 10 seconds when Movei need 11-12 seconds to get the
same number of solution but it is a win for Movei when I consider the
hardware difference.
You has 143 solutions in 60 seconds when Movei needed 113-114 seconds to get
the same number of solutions.

yes - if you are interested, I can send it to you...
the output looks also a little bit different... so you get e.g. the last eval and last line for each position (which is good to find errors in the testsuite - when your engine does not find the correct solution but reports a mate anyway, there might be something wrong in the testsuite...) also the last result table looks different... an example:


part of the position overview:
159 no [--:--:--,--] [10] +0.18 f1e1 h7h5 c7d6 f8e7 d6e7 e8e7 b5d6
c8d8 g3g5 d5d4
160 yes 12 [00:00:15,79] [14] +2.38 d2b2 a4b2 c4c3 b7b6 c5c4 b6b4 a7a5
b4b5 c3c2 g1f2 c2c1q f2e2 c1b1 a2a4 c4c3
161 yes 7 [00:00:00,34] [14] +3.53 c5b6 d5b5 b1b5 a7b6 c2c4 e6e7 b5b6
e7a7 c4c5 g7g6 c5c6 g8g7 f2f4 f7f6
162 yes 5 [00:00:00,06] [12] +3.64 f4f5 g6g5 e3e8 d7e8 c3g7 g5g7 f5f6
g7f8 d5c6 e8f7 c6b7 f8h6 b2c3 h6h5 c3d4 h5e5
d4f2
163 yes 11 [00:00:01,61] [16] +4.01 d5b3 a2b3 c5c4 d3d4 c4c3 b3a4 b4b3
g5h5 b3b2 d4b4 c3c2 b4b2 c2c1q b2f2 c1a1 f2f4
a1h8 h5g5 c6c5
you see that you also get the depth at which the program finds the solution (number in front of the time)
part of the results-table:
13 1 110 85
14 1 111 44
16 4 115 34 46 142 160
18 1 116 61
19 2 118 8 36
20 1 119 17
21 1 120 83
22 1 121 177
23 1 122 126
24 1 123 118
25 2 125 137 183
26 1 126 92
you can see that the positions are mentioned after the number of solutions... seconds without a solution are not written anymore... (which makes a long test much more readable)
as I said, Movei is very good in tactics... with Quark it is different...
when you compare the +-rating of Athlons and PIII or other Athlons you should look at the real clock speed, the MP1800+ is 1533 MHz, so you might look at 90 seconds - might be that Quark gets faster to depth then Movei without loosing to much in fast time controls - but helps in longer terms...
Greets, Thomas

Yes
Thanks
I am interested in it.
I am not so sure.
It seems that at fast time control movei is slightly better.

It is possible but I am not sure.

Note that Movei prints also wrong fail high and practically it never plays moves that fail high without solving the fail high(in most cases it changes nothing because movei tries to solve the fail high).
number of solution may be dependent on luck(see the following analysis of latest movei)
Id: ECM.1066
Fen: rn2qbr1/2p4k/p2p1nb1/1p1Pp2p/2P4P/2NBBPN1/PP1Q4/2KR2R1 w - - 0 1
Bm: Nxh5
no 1 0 +263 58 d3g6 e8g6
no 2 0 +114 560 d3g6 g8g6 d2e2
no 2 0 +115 602 c4b5
no 2 10 +144 700 c4b5
no 2 10 +198 905 c4b5 a6b5 d3b5
no 2 10 +198 1810 c4b5 a6b5 d3b5
no 3 20 +198 3001 c4b5 a6b5 d3b5
no 3 40 +198 6036 c4b5 a6b5 d3b5
no 4 50 +174 7599 c4b5 a6b5 d3b5 b8d7
no 4 110 +174 17444 c4b5 a6b5 d3b5 b8d7
no 5 140 +185 22218 c4b5 a6b5 d3b5 b8d7 a2a3
no 5 260 +185 44090 c4b5 a6b5 d3b5 b8d7 a2a3
no 6 330 +195 56014 c4b5 a6b5 d3b5 b8d7 a2a3 h7h8
no 6 510 +195 88282 c4b5 a6b5 d3b5 b8d7 a2a3 h7h8
no 7 770 +194 133927 c4b5 g6d3 d2d3 h7h8 g3e4 f6e4 g1g8 h8g8
c3e4 f8g7
no 7 1390 +194 250441 c4b5 g6d3 d2d3 h7h8 g3e4 f6e4 g1g8 h8g8
c3e4 f8g7
no 8 3760 +197 644321 c4b5 a6b5 d3b5 b8d7 g3e4 g6e4 f3e4 h7h8
yes 8 4980 +198 883437 g3h5
yes 8 6400 +227 1168108 g3h5
yes 8 6700 +231 1218928 g3h5 f6h5 g1g6 g8g6 d1g1 e5e4 d3e4 h5g3
g1g3
yes 8 6950 +231 1267059 g3h5 f6h5 g1g6 g8g6 d1g1 e5e4 d3e4 h5g3
g1g3
yes 9 8230 +231 1521307 g3h5 f6h5 g1g6 g8g6 d1g1 e5e4 d3e4 h5g3
g1g3
yes 9 9680 +231 1790758 g3h5 f6h5 g1g6 g8g6 d1g1 e5e4 d3e4 h5g3
g1g3
yes 10 11000 +191 2051533 g3h5 f6h5 g1g6 g8g6 d1g1 e5e4 d3e4 h5f4
e3f4 e8e4 f3e4 g6g1 c3d1 b5c4
no 10 17470 +192 3188147 c4b5
no 10 27310 +204 4913752 c4b5 g6d3 d2d3 h7h8 c1b1 f8e7 g3f5 g8g1
d1g1 a6b5 f5e7 e8e7 d3b5
no 10 34050 +204 6174614 c4b5 g6d3 d2d3 h7h8 c1b1 f8e7 g3f5 g8g1
d1g1 a6b5 f5e7 e8e7 d3b5
no 11 59760 +198 10775985 c4b5 a6b5 g3h5 f6h5 g1g6 g8g6 d1g1 e5e4
d3e4 h5g3 g1g3
yes 11 78200 +199 14370281 g3h5
no 11 78600 +198 14449108 c4b5 a6b5 g3h5 f6h5 g1g6 g8g6 d1g1 e5e4
d3e4 h5g3 g1g3
no 11 99500 +198 18500982 c4b5 a6b5 g3h5 f6h5 g1g6 g8g6 d1g1 e5e4
d3e4 h5g3 g1g3
no 12 209290 +206 38228929 c4b5 h7h8 g3e4 f6e4 f3e4 a6a5 c1b1 b8d7
e3g5 f8g7 d2f2 a5a4
yes 12 264890 +207 49121561 g3h5
Result: Success
Found in: 264890 ms (00:04:24.890)
Uri