Rebel 16.2 release
+12 Ipmanchess
Damir Desevac
texium
Dann Corbit
mwyoung
Peter Berger
stevemaughan
Mclane
Dio
pohl4711
Chris Whittington
Admin
16 posters
Author Message Admin Admin
Posts : Join date : Location : http://rebel13.nl/windows/rebel-16.html Update REBEL 16.2
1. Bugfix version time control and playing with multiple threads.
2. No engine or network improvements, some elo might come from the time control changes.
3. New UCI options.
• Polyglot book support. Default setting is off controlled by the UCI parameter Bookdepth=0 Changing the value (for instance) to 5 Rebel will limit the book moves till 5 moves. Occasionally this might be handy to avoid long book move lines that end up in positions where there is noting to play for any longer. For full use of the book set Bookdepth=100.
The book is created by Jeroen Noomen, book creator since Rebell 5.0 released 1986. Many voluminous experimental books are available from the Rebel website. [ 1 ] [ 2 ]
• Evalcorrect - This option can be used to change the playing style of the engine. The default value is 202. Changing the value to 256 (the maximum) will increase the engine aggressiveness, a value of 1 (the minimum) will change the playing style to boring. In general values between 128 and 202 are good without much loss in elo.
-------------
Other than that, for insiders, all cutechess testing is done with a timemargin of 500ms per game.
Damir Desevac, Dio, Eelco and Ipmanchess like this post
Like 4
pohl4711
Posts : Join date : Location : I will do a testrun for my SPCC-Ratinglist of Rebel 16.2 with Evalcorrect-parameter set to 256 (maximum), which will increase the engine aggressiveness... I am curious, if this will improve the EAS-scoring of Rebel again (Rebel 14.1 had a very high EAS-Score, but newer releases regressed very badly in EAS-scoring). And, lets see, how much Elo it costs, to set Evalcorrect-parameter to maximum (compared to the results of Rebel 16.1 in my ratinglist)...
https://www.sp-cc.de/eas-ratinglist.htm Results in (around) 5-6 days from now (first, the Berserk 11 testruns (classical and UHO) and the testrun of bugfixed Clover 3.3.1 have to be finished)
Admin Admin
Posts : Join date : Location : pohl4711 wrote: I will do a testrun for my SPCC-Ratinglist of Rebel 16.2 with Evalcorrect-parameter set to 256 (maximum), which will increase the engine aggressiveness... I am curious, if this will improve the EAS-scoring of Rebel again (Rebel 14.1 had a very high EAS-Score, but newer releases regressed very badly in EAS-scoring). And, lets see, how much Elo it costs, to set Evalcorrect-parameter to maximum (compared to the results of Rebel 16.1 in my ratinglist)... The high EAS-score of Rebel 14.1 (in retrospect) had everything to do with an immature net, the time I was still
learning how to create a well and balanced
learned net
Meaning to say, don't expect too much from Evalcorrect=256.
Quote : https://www.sp-cc.de/eas-ratinglist.htm Looking forward [!]
pohl4711
Posts : Join date : Location : Admin wrote: learning how to create a well and balanced learned net OK. But IMHO, Rebel 16.2 is the first nnue-engine, which has a parameter, that changes the aggressiveness of play. Other engines, like KomodoDragon for example, are switching the nnue-eval off, when the playing-style is changed in the parameters...
So, the testrun of Rebel 16.2 with Evalcorrect=256 is definitly very interesting...
Admin Admin
Posts : Join date : Location : I see... On the other hand I am afraid that the max of 256 is well beyond the safe side. For some dramatic effect you need higher values, 300, 400, maybe even 500. That would mean a private Stefan Pohl version
Chris Whittington
Posts : Join date : Location : Admin wrote: I see... On the other hand I am afraid that the max of 256 is well beyond the safe side. For some dramatic effect you need higher values, 300, 400, maybe even 500. That would mean a private Stefan Pohl version Is there actually a hard limit applied to the eval correct parameter? I will check ….
Yes, there is. clamped to 1, 256.
Chris Whittington
Posts : Join date : Location : Chris Whittington wrote: Admin wrote: I see... On the other hand I am afraid that the max of 256 is well beyond the safe side. For some dramatic effect you need higher values, 300, 400, maybe even 500. That would mean a private Stefan Pohl version There is another way to do it, in CSTal dev version, but unfortunately(?) disabled in the Rebel version, this allows for adjustment in same way as eval correct, but using an MG value and an EG value, which allows for adjust over the phase range.
Testers should note however, that the engine response to changing these parameters is surprisingly dull, in that they don't much affect move choice or Elo (in my rough testing anyway).
pohl4711
Posts : Join date : Location : Admin wrote: I see... On the other hand I am afraid that the max of 256 is well beyond the safe side. For some dramatic effect you need higher values, 300, 400, maybe even 500. That would mean a private Stefan Pohl version Why not releasing this? I would love to do some experimental testruns with some higher values (256, 300, 400, 500 or so). Perhaps, we will find a number, letting Rebel playing really spectacular and not too weak?!
Regards - Stefan
Dio
Posts : Join date : Admin likes this post
Like 1
Admin Admin
Posts : Join date : Location :
Dio
Posts : Join date : I was able, to delete the tests in the 40/4-list.. sorry..
Dio
Posts : Join date : I am trying to restore the previous results....
Admin Admin
Posts : Join date : Location : Dio wrote: I am trying to restore the previous results.... Code: Rebel 16.2NN x64 1CPU
Dio
Posts : Join date : Admin likes this post
Like 1
Mclane
Posts : Join date : Age : Location : here are my resuts from the 16 core with permanent brain ON:
Motor Punkte Lc St Re Lc St Et Et St Dr Et Sl Dr St Et Dr St Sl Ko De Dr Ko Ma Wa Wa To Xi Ch Bo Wa Se De Se De Re Re Ch St En Co Pr Be Pr Di Co Za De Fr De Co Co Gr Ma Ch Lo Ge Ts S-B
Admin Admin
Posts : Join date : Location : Thanks Thorsten, can you mail me the pgn ?
Chris Whittington
Posts : Join date : Location : Admin wrote: Thanks Thorsten, can you mail me the pgn ? Those results don’t make a lot of sense
stevemaughan
Posts : Join date : Location : Rebel 16.2 seems to be extremely strong and has an amazingly aggressive and interesting style of play. I've just played a 100 game match against Shredder 13 — both using Shredder's book. Rebel won 74 games, drew 24 and lost only 2, for a +300 ELO difference in strength. That's amazing!
Admin, Chris Whittington, Brendan and Dio like this post
Like 4
Admin Admin
Posts : Join date : Location : Just for the fun of it.
Code: RANK ENGINE GAMES POINTS 1 2 3 4 5 6 7 8 9 10
Graham's current tournament.
Mclane, Dio, Eelco and gerhard like this post
Like 4
Peter Berger
Posts : Join date : Welcome Rebel 16.2
Mclane likes this post
Like 1
Mclane
Posts : Join date : Age : Location : [pgn]
mwyoung
Posts : Join date : Location : Chris Whittington wrote: Admin wrote: Thanks Thorsten, can you mail me the pgn ? Why?
Mclane
Posts : Join date : Age : Location : mwyoung wrote: Chris Whittington wrote: Admin wrote: Thanks Thorsten, can you mail me the pgn ? Maybe he dislikes that some engines have problems with the time controls.
IMO it is not my job judge why some engines, e.g. Stockfish 15 or 15.1 have time control problems no matter if you test in 40/3‘ or 40/10 or even 40/20.
An engine that makes zillions of NPS and goes into the 40-50 ply should be capable to play 40/10‘ games.
If not it’s obviously a bug.
Rebel16.2 is doing a very good job and plays very strong.
Thats what counts.
Ghppn likes this post
Like 1
mwyoung
Posts : Join date : Location : Mclane wrote: mwyoung wrote: Chris Whittington wrote: Admin wrote: Thanks Thorsten, can you mail me the pgn ? I agree.
Ghppn likes this post
Like 1
mwyoung
Posts : Join date : Location : Admin wrote: Just for the fun of it. Code: RANK ENGINE GAMES POINTS 1 2 3 4 5 6 7 8 9 10
Great results for Rebel 16.2
You know what this means!
You are now on the enemies list of the Stockfish fanboys. They do not like it when another chess engine challenges Stockfish. As that is a hard and fast rule for testers. As the programmer of Rebel 16.2. You will need to hire body guards.
Admin and Ghppn like this post
Like 2
Sponsored content
Rebel 16.2 release
Similar topics
Permissions in this forum: You cannot reply to topics in this forum
Home 
Meaning to say, don't expect too much from Evalcorrect=256.
