Weird results Tunguska 1.1 vs Jonny 4.0

Discussion of chess software programming and technical issues.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
User avatar
Guenther
Posts: 3108
Joined: Wed Oct 01, 2008 4:33 am
Location: Regensburg, Germany
Full name: Guenther Simon
Contact:

Re: Weird results Tunguska 1.1 vs Jonny 4.0

Post by Guenther » Tue May 14, 2019 9:43 am

xr_a_y wrote:
Tue May 14, 2019 9:31 am
I checked some of the game, there are not that bad.
It would be sufficient to have depth lowered by some factor...
(cpu load due to too much concurrent games and other possibilities)

example vs. Minic
10. Bf4? discarded at depth 8 , 13.Bf2?? discarded at depth 7, both in less than 0.5 sec
(and Jonny is exremely fast in nps and reaches quite high depths normally despite being from 2010)

Current foe list count : [101]
http://rwbc-chess.de/chronology.htm

User avatar
xr_a_y
Posts: 778
Joined: Sat Nov 25, 2017 1:28 pm
Location: France

Re: Weird results Tunguska 1.1 vs Jonny 4.0

Post by xr_a_y » Tue May 14, 2019 9:56 am

Yes clearly something wrong but not like win on time or crash.

User avatar
Guenther
Posts: 3108
Joined: Wed Oct 01, 2008 4:33 am
Location: Regensburg, Germany
Full name: Guenther Simon
Contact:

Re: Weird results Tunguska 1.1 vs Jonny 4.0

Post by Guenther » Tue May 14, 2019 10:12 am

xr_a_y wrote:
Tue May 14, 2019 9:56 am
Yes clearly something wrong but not like win on time or crash.
Of course wins on time or crashes would be checked for a reason.

More subtile errors could go unchecked (for a while), although one would expect
a check, if an established program suddenly loses 10 or 20 or more games
in a row vs. an equal opponent.
Moreover I would always run the full pgn (which we don't have now) against a depth stats program.
(e.g. search for the 'Devel' incident
viewtopic.php?f=6&t=65572&p=736948&hili ... em#p736948)
Current foe list count : [101]
http://rwbc-chess.de/chronology.htm

User avatar
xr_a_y
Posts: 778
Joined: Sat Nov 25, 2017 1:28 pm
Location: France

Re: Weird results Tunguska 1.1 vs Jonny 4.0

Post by xr_a_y » Tue May 14, 2019 11:37 am

I wasn't aware, and now even more thankful, that the testing process was so professional.
Thanks again to all testers.

User avatar
Guenther
Posts: 3108
Joined: Wed Oct 01, 2008 4:33 am
Location: Regensburg, Germany
Full name: Guenther Simon
Contact:

Re: Weird results Tunguska 1.1 vs Jonny 4.0

Post by Guenther » Tue May 14, 2019 11:52 am

xr_a_y wrote:
Tue May 14, 2019 11:37 am
I wasn't aware, and now even more thankful, that the testing process was so professional.
Thanks again to all testers.
Well, I am no CCRL tester, I only described what I would do, if it was my test ;-)
Current foe list count : [101]
http://rwbc-chess.de/chronology.htm

Sergio Martinez
Posts: 1788
Joined: Fri Oct 05, 2012 12:35 pm
Location: Spain
Contact:

Re: Weird results Tunguska 1.1 vs Jonny 4.0

Post by Sergio Martinez » Sat May 18, 2019 9:37 pm

Certainly there are something bad here. I cant check now (out of home) but i will investigate it the monday.

I have seen these last games of Jonny were played in my Dual Xeon 5675. I leave some cores free there (Xeon has 12cores and I use 8-9 at same time max) so it cant be the reason.

I use Arena Gui (3.51), at the moment I dont know the reason why Jonny plays this way but i will try to repeat the games and I will tell you.

Thank you guys.
Member of the CCRL Group. Write me if you want I test your engine.

User avatar
xr_a_y
Posts: 778
Joined: Sat Nov 25, 2017 1:28 pm
Location: France

Re: Weird results Tunguska 1.1 vs Jonny 4.0

Post by xr_a_y » Fri May 24, 2019 5:54 am

So you discarded all recent Jonny's games ?

Modern Times
Posts: 2417
Joined: Thu Jun 07, 2012 9:02 pm

Re: Weird results Tunguska 1.1 vs Jonny 4.0

Post by Modern Times » Fri May 24, 2019 7:20 pm

xr_a_y wrote:
Fri May 24, 2019 5:54 am
So you discarded all recent Jonny's games ?
A work in progress. Jonny's default hash size is a tiny 1MB and the theory is that Arena is sometimes not setting the hash size correctly, which means it plays with the default 1MB which cripples it. The Arena bug is discussed here:

viewtopic.php?t=46586
.

Opinions expressed here are my own, and not necessarily those of the CCRL Group.

User avatar
Guenther
Posts: 3108
Joined: Wed Oct 01, 2008 4:33 am
Location: Regensburg, Germany
Full name: Guenther Simon
Contact:

Re: Weird results Tunguska 1.1 vs Jonny 4.0

Post by Guenther » Sat May 25, 2019 8:41 am

Modern Times wrote:
Fri May 24, 2019 7:20 pm
xr_a_y wrote:
Fri May 24, 2019 5:54 am
So you discarded all recent Jonny's games ?
A work in progress. Jonny's default hash size is a tiny 1MB and the theory is that Arena is sometimes not setting the hash size correctly, which means it plays with the default 1MB which cripples it. The Arena bug is discussed here:

viewtopic.php?t=46586
I don't think 1MB hash cripples as much as the results shown in some cases, especially if the real tc was perhaps 40/2.5.
(will try a few test matches with 1MB for Jonny vs. 512MB others)
I guess you still have the commented games somewhere with eval/depths and may be taskmanager logs?

BTW except for the 7 or 8 results already mentioned by me, I also later noticed a very strange behaviour in a very long match
vs. Godel 4.07 in 2018, where the CCRL 40/4 head to head table shows 24 losses in a row for Jonny 4.00 in the last
third of the match.
This match has the strange length of 112 games, which seems unusual?

(not sure though, if you resort the games even in a single match for whatever reason?)
The chance for this between programs of may be a diff of 40-45 rating points is... uhm... slightly around zero.
Current foe list count : [101]
http://rwbc-chess.de/chronology.htm

Modern Times
Posts: 2417
Joined: Thu Jun 07, 2012 9:02 pm

Re: Weird results Tunguska 1.1 vs Jonny 4.0

Post by Modern Times » Sat May 25, 2019 11:23 am

Guenther wrote:
Sat May 25, 2019 8:41 am

BTW except for the 7 or 8 results already mentioned by me, I also later noticed a very strange behaviour in a very long match
vs. Godel 4.07 in 2018, where the CCRL 40/4 head to head table shows 24 losses in a row for Jonny 4.00 in the last
third of the match.
Yes they go back to 2017 and maybe before. There are several opponents affected.
.

Opinions expressed here are my own, and not necessarily those of the CCRL Group.

Post Reply