Thanks - I'll look at this. I haven't had time to do a really deep analysis. If I force Bh6, I get this output from Shredder (15 min search):
no 16 13410 -391 14353935 Bf8 Nf6
no 16 17120 -398 19001935 Bf8 Nf6 Qe7 Rg3 Qxf6 Bg5 Qxg5 Qxg5 Rxb2
f6 g6 Ra3 Rc2 Rc1 Rxc1+ Qxc1 Rb6 Kf1 Kg8
Ke2 e4 f3
no 17 26880 -400 29623204 Bf8 Nf6 Qe7 Rg3 Qxf6 Bg5 Qxg5 Qxg5 f6 Qh5
Rxb2 Ra3 Kg8 Rxa6 Rc2 Qe8 Rxc4 Rxd6 Nf7
Qxb8 Nxd6
no 18 45860 -415 50300048 Bf8 Nf6 Qe7 Rg3 Qxf6 Bg5 Qxg5 Qxg5 Rxb2
f6 g6 Ra3 Rb1 Kf1 Rxe1+ Kxe1 Rb6 f4 e4
no 19 87340 -394 95736622 Bf8 Nf6 Qe7 Rg3 Qxf6 Bg5 Qxg5 Qxg5 f6 Qh5
Rxb2
no 20 157840 -418 169667408 Bf8 Nf6 Qe7 Rg3 Qxf6 Bg5 Be7 Bxf6
no 21 412280 -443 429832533 Bf8 Nf6
no 21 472560 -493 488725965 Bf8 Nf6
while if I force g6, it gets to ply 18 with a score of -305. But maybe Shredder will eventually score Rg3 even or better vs. g6. I'll run it longer and see.
arasan8.epd
Moderator: Ras
-
jdart
- Posts: 4411
- Joined: Fri Mar 10, 2006 5:23 am
- Location: http://www.arasanchess.org
-
jdart
- Posts: 4411
- Joined: Fri Mar 10, 2006 5:23 am
- Location: http://www.arasanchess.org
Re: arasan8.epd Alternative solution for 4.17
Yes, this is an alternate solution, it seems.
Shredder at 1 hr /move, Athlon 64X2 4400+:
After g6:
yes 15 47800 -406 48358809 g6 Bg5 Bxg5 Qxg5
yes 16 69580 -407 69409873 g6 Bg5 Nb7 Qh4 Bxg5 Nxg5 f6 Ne6 g5 Qh6
Qf7 h4 Rg8 hxg5 fxg5 f6 Nd8 Nxg5
yes 17 96770 -432 96758822 g6 Bg5
yes 17 133450 -482 134505113 g6 Bg5
yes 17 423050 -608 430736887 g6 Bg5 Nc6 Qh4 Bxg5
yes 18 576800 -633 574913460 g6 Bg5
yes 18 822280 -682 813039615 g6 Bg5 Bxg5 Qxg5
yes 19 1321620 -707 1283970411 g6 Bg5
yes 19 1881280 -755 1815018091 g6 Bg5 Qe8 Bf6+
yes 20 2503840 -756 2147483647 g6 Bg5 Bxg5 Qxg5 Qxf5 Qh6
After Rg3:
no 15 7640 -356 8248108 Bf8 Nf6 Qe7 Rg3 Qxf6 Bg5 Qxg5 Qxg5 f6 Qh5
Rxb2 Ra3 R2b6 Kf1 Kg8 Qe8 Nf7 Qe6 Rb4
no 16 13750 -373 14895949 Bf8 Nf6 Qe7 Rg3 Qxf6 Bg5 Qxg5 Qxg5 f6 Qh5
Rxb2 Ra3 R2b6 Kf1 Rb4
no 17 21230 -398 22795418 Bf8 Nf6
no 17 29830 -420 33140552 Bf8 Nf6 Qe7 Rg3 Qxf6 Bg5 Be7 Bxf6
no 18 55760 -420 61551332 Bf8 Nf6 Qe7 Rg3 Qxf6 Bg5 Qxg5 Qxg5 f6 Qh5
Rxb2 Ra3 R2b4 Re4 R4b6 Rh4 h6 Qg6 Rb1+
Kh2
no 19 115330 -445 125457540 Bf8 Nf6
no 19 133970 -495 144490323 Bf8 Nf6
no 19 930450 -635 941551551 Bf8 Rg3 f6 Nxf6 Qf7 Bg5 Be7 Nxh7
yes 19 1022640 -634 1031570030 gxh6
yes 19 1061380 -599 1071556450 gxh6
yes 19 1126640 -599 1134886108 gxh6 Rg3
yes 20 1390840 -614 1395259203 gxh6 Rg3 Bg5 Nxg5 f6 Nxh7 Qxh7 Qh4 R6b7
Rg6 Qg7 Re3
yes 21 1724910 -629 1717142650 gxh6 Rg3 Bg5 Nxg5 f6 Nxh7 Qxh7 Qh4 R6b7
Rg6 Qxg6 fxg6 Kg7 Re3 Rb3 Rxb3 Rxb3 Qg4
Rb8 Qd7+
yes 22 2519940 -647 2147483647 gxh6 Rg3 Bg5 Nxg5 f6 Nxh7 Qxh7 Qh4 R6b7
Rg6 Qg7 Re3
Shredder at 1 hr /move, Athlon 64X2 4400+:
After g6:
yes 15 47800 -406 48358809 g6 Bg5 Bxg5 Qxg5
yes 16 69580 -407 69409873 g6 Bg5 Nb7 Qh4 Bxg5 Nxg5 f6 Ne6 g5 Qh6
Qf7 h4 Rg8 hxg5 fxg5 f6 Nd8 Nxg5
yes 17 96770 -432 96758822 g6 Bg5
yes 17 133450 -482 134505113 g6 Bg5
yes 17 423050 -608 430736887 g6 Bg5 Nc6 Qh4 Bxg5
yes 18 576800 -633 574913460 g6 Bg5
yes 18 822280 -682 813039615 g6 Bg5 Bxg5 Qxg5
yes 19 1321620 -707 1283970411 g6 Bg5
yes 19 1881280 -755 1815018091 g6 Bg5 Qe8 Bf6+
yes 20 2503840 -756 2147483647 g6 Bg5 Bxg5 Qxg5 Qxf5 Qh6
After Rg3:
no 15 7640 -356 8248108 Bf8 Nf6 Qe7 Rg3 Qxf6 Bg5 Qxg5 Qxg5 f6 Qh5
Rxb2 Ra3 R2b6 Kf1 Kg8 Qe8 Nf7 Qe6 Rb4
no 16 13750 -373 14895949 Bf8 Nf6 Qe7 Rg3 Qxf6 Bg5 Qxg5 Qxg5 f6 Qh5
Rxb2 Ra3 R2b6 Kf1 Rb4
no 17 21230 -398 22795418 Bf8 Nf6
no 17 29830 -420 33140552 Bf8 Nf6 Qe7 Rg3 Qxf6 Bg5 Be7 Bxf6
no 18 55760 -420 61551332 Bf8 Nf6 Qe7 Rg3 Qxf6 Bg5 Qxg5 Qxg5 f6 Qh5
Rxb2 Ra3 R2b4 Re4 R4b6 Rh4 h6 Qg6 Rb1+
Kh2
no 19 115330 -445 125457540 Bf8 Nf6
no 19 133970 -495 144490323 Bf8 Nf6
no 19 930450 -635 941551551 Bf8 Rg3 f6 Nxf6 Qf7 Bg5 Be7 Nxh7
yes 19 1022640 -634 1031570030 gxh6
yes 19 1061380 -599 1071556450 gxh6
yes 19 1126640 -599 1134886108 gxh6 Rg3
yes 20 1390840 -614 1395259203 gxh6 Rg3 Bg5 Nxg5 f6 Nxh7 Qxh7 Qh4 R6b7
Rg6 Qg7 Re3
yes 21 1724910 -629 1717142650 gxh6 Rg3 Bg5 Nxg5 f6 Nxh7 Qxh7 Qh4 R6b7
Rg6 Qxg6 fxg6 Kg7 Re3 Rb3 Rxb3 Rxb3 Qg4
Rb8 Qd7+
yes 22 2519940 -647 2147483647 gxh6 Rg3 Bg5 Nxg5 f6 Nxh7 Qxh7 Qh4 R6b7
Rg6 Qg7 Re3
-
Eelco de Groot
- Posts: 4681
- Joined: Sun Mar 12, 2006 2:40 am
- Full name: Eelco de Groot
Re: arasan8.epd Alternative solution for 4.17
Thanks for testing this Jon, it was interesting for me that this version of Toga Checkov played another move than all the other Checkovs. Most play Bh6, but in two best mode after a while Rg3 gets a better score.
This has not much to do with the Arasan test, but may be of some interest, I compared the sources of Toga II 3.2.1 SE with the ancestor Toga II 1.3.1 version and there are really very little changes. Basically what was added is a new intermediate level of Futility pruning, the margin of which is still fixed, there is not an UCI option to set it at a different level. Because the other two margins are variable in the UCI options I think maybe this Intermediate Futility Margin could be UCI-optionally changed too.
The other major change is that in the endgame the material values for Knights and for Bishops are now slightly lower than the opening values. I was surprised that so few changes could make a significant gain in the 40/4 tests. There are not yet a lot of games but still after 700 games the TPR for Toga II 3.2.1 SE is 40 elo better that is quite a jump considering the few changed lines.
In the Checkov settings I have somewhat larger Futility margins already, I changed the intermediate margin level accordingly in the source and managed to compile it with MSVC! No errors only a lot of warnings but I have not studied those yet. It is not as fast as the original by Charles Formula or maybe the original SE compile was made by Denis, but the difference in speed is also not that big and this compile is very small, only 92 Kb, probably in the optimized compile there are a lot of unrolled loops or something? But in short there is now an official fork of the Toga Checkov!
It only has one number changed so I'm not thinking that it will make a lot of difference, but it is nice to have made my own version. No official release, that would be a bit silly with so few changes apart from the Checkov setting changes.
The Beta did well in Alekhine - Sterk, in fact with a slightly faster compile it could have actually found the solution here at twenty seconds per move!
[d]r4rk1/p4ppp/qp2p3/b5B1/n1R5/5N2/PP2QPPP/1R4K1 w - -
Engine: Toga Checkov Beta 1f (256 MB, Athlon 2009 MHz)
by Thomas Gaksch and Fabien Letouzey
7/19 0:00 +0.92 1.Nd4 Nc5 2.b4 e5 3.Nf5 f6 4.bxc5 fxg5
5.Qxe5 (41.394)
8/22 0:00 +0.92 1.Nd4 Nc5 2.b4 e5 3.Nf5 f6 4.bxc5 fxg5
5.Qxe5 (82.302)
8/24 0:00 +1.31 1.Qf1 Nc5 2.b4 Bxb4 3.Rbxb4 Qxa2
4.Be7 Rfc8 5.Bxc5 Rxc5 6.Rxc5 bxc5 (136.806)
9/24 0:00 +1.45 1.Qf1 Nc5 2.Be3 Rfc8 3.b4 Bxb4
4.Rcxb4 Qxf1+ 5.Kxf1 Nd3 6.Rd4 Rc3 (201.263)
10/28 0:00 +1.31 1.Qf1 Nc5 2.b4 Bxb4 3.Rbxb4 Qxa2
4.Be7 Rfc8 5.h4 f6 6.Bxc5 Rxc5
7.Rxc5 bxc5 8.Rb7 (421.984)
11/36 0:02 +1.31 1.Qf1 Nc5 2.b4 Bxb4 3.Rbxb4 Qxa2
4.Be7 Rfc8 5.h4 f6 6.h5 e5 7.Bxc5 Rxc5
8.Rxc5 bxc5 9.Rb7 (1.217.888) 535
11/36 0:02 +1.71++ 1.Nd4 Nc5 2.b4 e5 3.Nf5 Ne6 4.Bh6 Rae8
5.bxa5 Ra8 6.axb6 gxh6 7.bxa7 Rxa7
8.Nxh6+ Kg7 (1.452.425) 535
12/36 0:04 +1.64 1.Nd4 Nc5 2.b4 f6 3.Be3 Bxb4 4.Rbxb4 Rac8
5.Nb5 Rfd8 6.Rc2 e5 7.Qc4+ Kh8 8.Qg4 (2.242.878) 530
13/41 0:09 +1.49 1.Nd4 Nc5 2.b4 Ne4 3.Be7 Rfe8 4.bxa5 Nd2
5.Qxd2 Qxc4 6.Rc1 Qd5 7.Rc7 Qxa5
8.Qxa5 bxa5 (5.108.360) 535
13/48 0:22 +2.04++ 1.Bf6 Rfc8 2.Qe5 h6 3.Qg3 g6 4.Qh4 Qxc4
5.Qxh6 Bd2 6.Nxd2 Qc1+ 7.Rxc1 Rxc1+
8.Nf1 (12.109.626) 539
14/58 0:58 +2.63++ 1.Bf6 Rfc8 2.Qe5 Rc5 3.Rxc5 gxf6
4.Qg3+ Kh8 5.Rc7 Qd3 6.Rbc1 Rg8
7.Ng5 Qxg3 8.Nxf7+ Kg7 9.hxg3 Kg6
10.Nd6 Nxb2 11.Rxa7 (32.115.638) 546
15/58 1:23 +2.68 1.Bf6 Rfc8 2.Qe5 Rc5 3.Rxc5 gxf6
4.Qg3+ Kh8 5.Rc7 Qd3 6.Rbc1 Qg6
7.Qxg6 hxg6 8.Rxf7 Nxb2 9.Rxf6 Nd3
10.Rc7 (46.532.887) 554
16/58 3:51 +3.05 1.Bf6 Rfc8 2.Qe5 Rxc4 3.Qg5 Rg4
4.Qxg4 g6 5.Qxa4 b5 6.Qb3 Bb6 7.Rd1 Rc8
8.h4 Bc5 9.Be5 Qc6 10.Ng5 (132.290.612) 572
17/81 10:59 +3.79++ 1.Bf6 Rfc8 2.Qe5 Rxc4 3.Qg5 Rg4
4.Qxg4 g6 5.Qxa4 b5 6.Qa3 b4 7.Qa4 Qd3
8.Rc1 Qf5 9.Be5 Bb6 10.Qxb4 Rd8 (384.043.717) 582
best move: Bg5-f6 time: 13:55.250 min n/s: 583.979 CPU 100.0% n/s(1CPU): 583.979 nodes: 487.650.000
I don't think I've ever seen this position solved so quickly
Not on my hardware!
This version finds Bh6 first again in Arasan 4.17, not Rg3. Because of the large value implemented for King Safety, 125 as standard against default 100, the strength in comp-comp games will probably not be very good, especially if it would be pitted against all those Rybka versions in the rating list it would lose. But I don't think it is very bad either.
I'm currently playing some games against the old Belka 1.8.11, I'm curious if Toga Checkov Beta can win that match, these two programs should not differ much in strength, that is, if Belka 1.8.11 is not weaker than Strelka 1.8 the ratings and head to head results should be fairly close. (I think I will soon buy one of the Rybka versions, if only because I don't really want to test against Belka anymore without actually owning Rybka)
Eelco
This has not much to do with the Arasan test, but may be of some interest, I compared the sources of Toga II 3.2.1 SE with the ancestor Toga II 1.3.1 version and there are really very little changes. Basically what was added is a new intermediate level of Futility pruning, the margin of which is still fixed, there is not an UCI option to set it at a different level. Because the other two margins are variable in the UCI options I think maybe this Intermediate Futility Margin could be UCI-optionally changed too.
The other major change is that in the endgame the material values for Knights and for Bishops are now slightly lower than the opening values. I was surprised that so few changes could make a significant gain in the 40/4 tests. There are not yet a lot of games but still after 700 games the TPR for Toga II 3.2.1 SE is 40 elo better that is quite a jump considering the few changed lines.
In the Checkov settings I have somewhat larger Futility margins already, I changed the intermediate margin level accordingly in the source and managed to compile it with MSVC! No errors only a lot of warnings but I have not studied those yet. It is not as fast as the original by Charles Formula or maybe the original SE compile was made by Denis, but the difference in speed is also not that big and this compile is very small, only 92 Kb, probably in the optimized compile there are a lot of unrolled loops or something? But in short there is now an official fork of the Toga Checkov!
It only has one number changed so I'm not thinking that it will make a lot of difference, but it is nice to have made my own version. No official release, that would be a bit silly with so few changes apart from the Checkov setting changes.
The Beta did well in Alekhine - Sterk, in fact with a slightly faster compile it could have actually found the solution here at twenty seconds per move!
[d]r4rk1/p4ppp/qp2p3/b5B1/n1R5/5N2/PP2QPPP/1R4K1 w - -
Engine: Toga Checkov Beta 1f (256 MB, Athlon 2009 MHz)
by Thomas Gaksch and Fabien Letouzey
7/19 0:00 +0.92 1.Nd4 Nc5 2.b4 e5 3.Nf5 f6 4.bxc5 fxg5
5.Qxe5 (41.394)
8/22 0:00 +0.92 1.Nd4 Nc5 2.b4 e5 3.Nf5 f6 4.bxc5 fxg5
5.Qxe5 (82.302)
8/24 0:00 +1.31 1.Qf1 Nc5 2.b4 Bxb4 3.Rbxb4 Qxa2
4.Be7 Rfc8 5.Bxc5 Rxc5 6.Rxc5 bxc5 (136.806)
9/24 0:00 +1.45 1.Qf1 Nc5 2.Be3 Rfc8 3.b4 Bxb4
4.Rcxb4 Qxf1+ 5.Kxf1 Nd3 6.Rd4 Rc3 (201.263)
10/28 0:00 +1.31 1.Qf1 Nc5 2.b4 Bxb4 3.Rbxb4 Qxa2
4.Be7 Rfc8 5.h4 f6 6.Bxc5 Rxc5
7.Rxc5 bxc5 8.Rb7 (421.984)
11/36 0:02 +1.31 1.Qf1 Nc5 2.b4 Bxb4 3.Rbxb4 Qxa2
4.Be7 Rfc8 5.h4 f6 6.h5 e5 7.Bxc5 Rxc5
8.Rxc5 bxc5 9.Rb7 (1.217.888) 535
11/36 0:02 +1.71++ 1.Nd4 Nc5 2.b4 e5 3.Nf5 Ne6 4.Bh6 Rae8
5.bxa5 Ra8 6.axb6 gxh6 7.bxa7 Rxa7
8.Nxh6+ Kg7 (1.452.425) 535
12/36 0:04 +1.64 1.Nd4 Nc5 2.b4 f6 3.Be3 Bxb4 4.Rbxb4 Rac8
5.Nb5 Rfd8 6.Rc2 e5 7.Qc4+ Kh8 8.Qg4 (2.242.878) 530
13/41 0:09 +1.49 1.Nd4 Nc5 2.b4 Ne4 3.Be7 Rfe8 4.bxa5 Nd2
5.Qxd2 Qxc4 6.Rc1 Qd5 7.Rc7 Qxa5
8.Qxa5 bxa5 (5.108.360) 535
13/48 0:22 +2.04++ 1.Bf6 Rfc8 2.Qe5 h6 3.Qg3 g6 4.Qh4 Qxc4
5.Qxh6 Bd2 6.Nxd2 Qc1+ 7.Rxc1 Rxc1+
8.Nf1 (12.109.626) 539
14/58 0:58 +2.63++ 1.Bf6 Rfc8 2.Qe5 Rc5 3.Rxc5 gxf6
4.Qg3+ Kh8 5.Rc7 Qd3 6.Rbc1 Rg8
7.Ng5 Qxg3 8.Nxf7+ Kg7 9.hxg3 Kg6
10.Nd6 Nxb2 11.Rxa7 (32.115.638) 546
15/58 1:23 +2.68 1.Bf6 Rfc8 2.Qe5 Rc5 3.Rxc5 gxf6
4.Qg3+ Kh8 5.Rc7 Qd3 6.Rbc1 Qg6
7.Qxg6 hxg6 8.Rxf7 Nxb2 9.Rxf6 Nd3
10.Rc7 (46.532.887) 554
16/58 3:51 +3.05 1.Bf6 Rfc8 2.Qe5 Rxc4 3.Qg5 Rg4
4.Qxg4 g6 5.Qxa4 b5 6.Qb3 Bb6 7.Rd1 Rc8
8.h4 Bc5 9.Be5 Qc6 10.Ng5 (132.290.612) 572
17/81 10:59 +3.79++ 1.Bf6 Rfc8 2.Qe5 Rxc4 3.Qg5 Rg4
4.Qxg4 g6 5.Qxa4 b5 6.Qa3 b4 7.Qa4 Qd3
8.Rc1 Qf5 9.Be5 Bb6 10.Qxb4 Rd8 (384.043.717) 582
best move: Bg5-f6 time: 13:55.250 min n/s: 583.979 CPU 100.0% n/s(1CPU): 583.979 nodes: 487.650.000
I don't think I've ever seen this position solved so quickly
Not on my hardware!This version finds Bh6 first again in Arasan 4.17, not Rg3. Because of the large value implemented for King Safety, 125 as standard against default 100, the strength in comp-comp games will probably not be very good, especially if it would be pitted against all those Rybka versions in the rating list it would lose. But I don't think it is very bad either.
I'm currently playing some games against the old Belka 1.8.11, I'm curious if Toga Checkov Beta can win that match, these two programs should not differ much in strength, that is, if Belka 1.8.11 is not weaker than Strelka 1.8 the ratings and head to head results should be fairly close. (I think I will soon buy one of the Rybka versions, if only because I don't really want to test against Belka anymore without actually owning Rybka)
Eelco