About BigLion 2.23n (and AEGT)

[Matthias Gemuh]

Geschrieben von:/Posted by: Matthias Gemuh at 28 July 2004 20:59:13:





There have been reports and results that prove that BigLion 2.23m and Taktix
2.23m are broken. I have uploaded a sort of bugfix.
The problem was probably in the new NegaScout search.
Version 2.23i used PVS search. Version 2.23n goes back to PVS search, but
differs from version 2.23i in that:
1) extra search performed on fail low at root.
2) fix attempt for hang if opponent sends move before resigning.
3) less risky time management concerning obvious moves.
Current, unreleased version 2.30 has slighty different data structures, and
much more attack information so going back to tweak a version that is based on
2.23i is understandably risky. I am basically still testing version 2.23n.
It may be wise for AEGT testers to use version 2.23i, because no opponent of
BigLion is likely to resign, so no hangs expected. I am happy that BigLion is
clearly the weakest in the Bishop class because the games played by version
will not distort final outcome. Even stable version 2.23i would have lost
the games.
Now to other testers: if BigLion 2.23n still hangs after opponent resigns,
please inform me. Log files of lost games in general would help me.
Thanks and apologies,
Matthias.



BigLion + Taktix

[Heinz van Kempen]

Geschrieben von:/Posted by: Heinz van Kempen at 28 July 2004 22:54:25:
Als Antwort auf:/In reply to: About BigLion 2.23n (and AEGT) geschrieben von:/posted by: Matthias Gemuh at 28 July 2004 20:59:13:

There have been reports and results that prove that BigLion 2.23m and Taktix
2.23m are broken. I have uploaded a sort of bugfix.
The problem was probably in the new NegaScout search.
Version 2.23i used PVS search. Version 2.23n goes back to PVS search, but
differs from version 2.23i in that:
1) extra search performed on fail low at root.
2) fix attempt for hang if opponent sends move before resigning.
3) less risky time management concerning obvious moves.
Current, unreleased version 2.30 has slighty different data structures, and
much more attack information so going back to tweak a version that is based on
2.23i is understandably risky. I am basically still testing version 2.23n.
It may be wise for AEGT testers to use version 2.23i, because no opponent of
BigLion is likely to resign, so no hangs expected. I am happy that BigLion is
clearly the weakest in the Bishop class because the games played by version
will not distort final outcome. Even stable version 2.23i would have lost
the games.
Now to other testers: if BigLion 2.23n still hangs after opponent resigns,
please inform me. Log files of lost games in general would help me.
Thanks and apologies,
Matthias.

[Gabriel LEPERLIER]

Geschrieben von:/Posted by: Gabriel LEPERLIER at 29. July 2004 10:49:
Als Antwort auf:/In reply to: About BigLion 2.23n (and AEGT) geschrieben von:/posted by: Matthias Gemuh at 28 July 2004 20:59:13:

There have been reports and results that prove that BigLion 2.23m and Taktix
2.23m are broken. I have uploaded a sort of bugfix.
The problem was probably in the new NegaScout search.
Version 2.23i used PVS search. Version 2.23n goes back to PVS search, but
differs from version 2.23i in that:
1) extra search performed on fail low at root.
2) fix attempt for hang if opponent sends move before resigning.
3) less risky time management concerning obvious moves.
Current, unreleased version 2.30 has slighty different data structures, and
much more attack information so going back to tweak a version that is based on
2.23i is understandably risky. I am basically still testing version 2.23n.
It may be wise for AEGT testers to use version 2.23i, because no opponent of
BigLion is likely to resign, so no hangs expected. I am happy that BigLion is
clearly the weakest in the Bishop class because the games played by version
will not distort final outcome. Even stable version 2.23i would have lost
the games.
Now to other testers: if BigLion 2.23n still hangs after opponent resigns,
please inform me. Log files of lost games in general would help me.
Thanks and apologies,
Matthias.

Hi Matthias,
I just sent you a mail with Taktix and Big Lion 2.23m logs. After a difficult beginning, they now play better in my tournament round 8 of 18.
Best regards,
Gabriel

UEL Minime - 45'KO - 27 Juillet 2004
ClassementModuleAuteurPaysScoreBiEnWJDeTaAsCeAlFiMaS-B01BigLion 2.23m Matthias GemuhGermany6,0/9·010110111 22,00 02EnginMax 5.11cEngin ÜstünGermany5,0/81·100101 1 21,00 03WJChess 1.62 Jean-François GAZETFrance5,0/900·=01111= 17,00 04DelphiMax 2.9 Martin BauerGermany4,5/811=·0 10== 19,50 05Taktix 2.23m Matthias GemuhCameroon4,5/80111·01 0= 19,25 06Asterisk_04b Peter HorvathHungary4,5/8000 1·=111 14,25 07Celes 0.77c Johan HuttingNetherland3,5/811000=·01 15,75 08Alex 1.60.09e Normand M. BlaisUSA3,0/80001 01·01 10,50 09FireFly 1.1.1 Andrew FanHongkong2,5/80 0=1001·0 9,75 10Marvin 1.1.0 Martin DanielssonSweden2,5/800===0 01· 9,50
41 de 90 parties jouées
Commencement du tournoi: 2004.07.27, 10:37:36
Dernière mise à jour: 2004.07.29, 10:16:59
Site/ Pays: Le Fou Numerique, France
Niveau: Blitz 45/0
Matériel: Intel 598 MHz
Système d'exploitation: Microsoft Windows 2000 Build 2195 Service Pack 4
Grille créée avec: Arena 1.07