Posts : 160 Join date : 2022-03-01 Location : Berlin
Subject: Re: News about Rebel or Chess System Tal ? Thu Dec 01, 2022 9:03 am
Admin wrote:
Looking good....
Time management of Rebel 16 looking bad...
After 3400 games, Rebel 16 used 178 secs in average in my testings (3min+1sec). Other engines in my ratinglist use (in average) 233 sec per game !! So Rebel 16 uses only (around) 75% of the time, other engines use.
Average game duration of my ratinglist: 7min 46sec = 233 secs per color/engine Average game duration of Rebel 16 gauntlet games: 6min 50sec = 205sec per color/engine, so 28 secs less. But, if the other engines does not play faster (why should they?), Rebel has 56 sec less (28 *2), because Rebel 16 is the only reason for the smaller amount of time...
Of course, 25% more time (with a fix in the time-management), would give Rebel 16 not much Elo (around 5 Elo, perhaps), but it is definitly a waste of gametime/Elo. And should get checked (and fixed) in the forthcoming CSTal release. Using the tool, I linked here, this can be done very easy. If cutechess(-cli) is used, each game has a GameDuration-TAG...
Last edited by pohl4711 on Thu Dec 01, 2022 9:18 am; edited 5 times in total
Damir Desevac, Dio and Ipmanchess like this post
Ipmanchess
Posts : 42 Join date : 2022-06-08
Subject: Re: News about Rebel or Chess System Tal ? Thu Dec 01, 2022 9:12 am
Have same findings.. plays too fast ,each 1000games it finish +/-30min. faster then other engines! TC: 10s+1s and Rebel 16 ends with +30sec left ..so the 10sec. don't go down ,or plays the moves faster then 1s increment it gets.
For sure it do well ..but in this case Rebel could win easy some extra Elo by tuning TM.
Will see how it handle 5min.games that will start now on other system..
matejst, Damir Desevac and Dio like this post
Chris Whittington
Posts : 1254 Join date : 2020-11-17 Location : France
Subject: Re: News about Rebel or Chess System Tal ? Thu Dec 01, 2022 11:00 am
Thanks for those. The time control algorithms are 1997, and I don’t think we ever tested game in N, let alone plus increment. IIRC the move control for game in N is time left divided by 50 (or something). I’ll try decreasing the divisor by the ratio suggested in your post.
Alex-Microsmeta
Posts : 5 Join date : 2022-12-01
Subject: Re: News about Rebel or Chess System Tal ? Thu Dec 01, 2022 2:54 pm
Admin wrote:
The next (and last) Rebel will be the next step in the cooperation between Chris and me. We have been working very hard to replace Toga and port the code base of Chess System Tal instead. The advantages of this move are, 1) clean compiles, 2) a stronger search and last but not least we now work together on creating stronger nets and I can do what I like most (and was my only goal to return to computer chess) getting the most out of NNUE programming.
On the latter, the last Rebel will come with the same version 10 net architecture as Rebel 15.1 but much improved, peppered with special HCE ProDeo evaluation stuff.
In the meantime Chris has created another even more brilliant net architecture, version 20 [!], currently already 30-35 elo stronger than net version 10, with no end in sight.
So this will be the new reality, 1+1 = 3, team work.
I like to present the first beta result.
[code]No. Name Win Draw Loss Unf. Score Games % ---------------------------------------------------------------- 1 BETA-ONE +1525 =4180 -1295 *0 3615.0
No idea how much elo that is.
But net version 20 beats them all.
Thank you Ed and Chris! You are 2 stars in chess engines history and I'm so happy that you are back!
Testing It right now on my TCEC 23 Active chess Gauntlet with all other bigs! http://t.ly/iToU-
Admin and Mclane like this post
Mclane
Posts : 3024 Join date : 2020-11-17 Age : 57 Location : United States of Europe, Germany, Ruhr area
Subject: Re: News about Rebel or Chess System Tal ? Thu Dec 01, 2022 9:27 pm
Ed and chris, our two old fellows, are the best that could happen to us.
Admin and Alex-Microsmeta like this post
Mclane
Posts : 3024 Join date : 2020-11-17 Age : 57 Location : United States of Europe, Germany, Ruhr area
Subject: Re: News about Rebel or Chess System Tal ? Fri Dec 02, 2022 7:52 am
Last edited by Admin on Fri Dec 02, 2022 9:44 am; edited 1 time in total
Mclane, Dio and Ipmanchess like this post
Alex-Microsmeta
Posts : 5 Join date : 2022-12-01
Subject: Re: News about Rebel or Chess System Tal ? Fri Dec 02, 2022 8:43 am
Mclane wrote:
here the first results:
2244 Partien gespielt / Turnier beendet Name des Turniers: Ryzen7 5700 Ort/ Land: LAPTOP-JTQQP8M7, Deutschland Spielstufe: Turnier 40/4 Hardware: AMD Ryzen 7 5700U with Radeon Graphics mit 4.096 MB Speicher
Thorsten is really a BACK TO THE FUTURE ... How many nice souvenirs of 25 years ago, when I was a Ed's Rebel Italian reseller Rebel was so wonderful (also the gui)
Even my first results against Stockfish & derivatives are good
Admin and Mclane like this post
Dio
Posts : 222 Join date : 2021-08-28
Subject: Re: News about Rebel or Chess System Tal ? Fri Dec 02, 2022 1:55 pm
I have played 1400 games with Rebel 16 so far. I use Cutechess (GUI), 8 games are played simultaneously.
In 3 games (against different opponents) it happened that Rebel 16.0 stopped calculating already on search depth 1, mate values and normal evaluations were displayed alternately. I then had to restart the GUI to continue the match.
Have other testers had the same experience ?
pohl4711
Posts : 160 Join date : 2022-03-01 Location : Berlin
Subject: Re: News about Rebel or Chess System Tal ? Fri Dec 02, 2022 2:45 pm
I had no freezes, but something is definitly wrong with Rebel. Look at move 109 by Rebel 16 (black)... Drawn position, but a thinking depth of 1 after 1.2 seconds and a bogus evaluation: Rh6+ {-M124/1 1.2s} (This happens again and again in my testrun) (I use cutechess-cli)
(And Rebel 16a moves still too fast. Now, 7min 5secs per game in my testings (better than Rebel 16 (6min 50sec)) Normal average value in my ratinglist is 7min 45secs
So, I aborted the Rebel testruns. Makes no sense right now. Engine is buggy. And time management is still bad.
Posts : 2609 Join date : 2020-11-17 Location : Netherlands
Subject: Re: News about Rebel or Chess System Tal ? Fri Dec 02, 2022 5:28 pm
Thanks for reporting those. Typical for first time public releases.
Dio likes this post
Chris Whittington
Posts : 1254 Join date : 2020-11-17 Location : France
Subject: Re: News about Rebel or Chess System Tal ? Fri Dec 02, 2022 5:34 pm
Dio wrote:
I have played 1400 games with Rebel 16 so far. I use Cutechess (GUI), 8 games are played simultaneously.
In 3 games (against different opponents) it happened that Rebel 16.0 stopped calculating already on search depth 1, mate values and normal evaluations were displayed alternately. I then had to restart the GUI to continue the match.
Have other testers had the same experience ?
I think it’s me trying to be too clever at reporting mates. Am looking into it.
Dio likes this post
Chris Whittington
Posts : 1254 Join date : 2020-11-17 Location : France
Subject: Re: News about Rebel or Chess System Tal ? Fri Dec 02, 2022 5:41 pm
All reports are -M124, which will make trapping it easier. -124 is outside maximum depth, so theoretically not possible, which gives me a good clue what it is (bounds error in hash table where depth is held in 8 bit, and depth gets used to adjust mate scores returned from the hash).
Shall I break from doing the pondering code to fix this?
Three things -M124 Sudden death still too slow Add ponder
pohl4711 wrote:
I had no freezes, but something is definitly wrong with Rebel. Look at move 109 by Rebel 16 (black)... Drawn position, but a thinking depth of 1 after 1.2 seconds and a bogus evaluation: Rh6+ {-M124/1 1.2s} (This happens again and again in my testrun) (I use cutechess-cli)
(And Rebel 16a moves still too fast. Now, 7min 5secs per game in my testings (better than Rebel 16 (6min 50sec)) Normal average value in my ratinglist is 7min 45secs
So, I aborted the Rebel testruns. Makes no sense right now. Engine is buggy. And time management is still bad.
Subject: Re: News about Rebel or Chess System Tal ? Fri Dec 02, 2022 5:51 pm
under ARENA there are no problems, also with me was alternately displayed the mate value -124.00 and then a balanced evaluation. The engine then hangs on depth 1. The GUI freezes briefly.
After about 3 to 5 minutes the game continues as if nothing had happened. Sometimes both clocks run at the same time.
I don't see Ponder as very important at the moment.
This bug does not seem to have any influence on the game strength. Since the bug occurs rarely, it is not easy to take a screenshot.
Posts : 1254 Join date : 2020-11-17 Location : France
Subject: Re: News about Rebel or Chess System Tal ? Fri Dec 02, 2022 5:55 pm
Dio wrote:
under ARENA there are no problems, also with me was alternately displayed the mate value -124.00 and then a balanced evaluation. The engine then hangs on depth 1. The GUI freezes briefly.
After about 3 to 5 minutes the game continues as if nothing had happened. Sometimes both clocks run at the same time.
I don't see Ponder as very important at the moment.
This bug does not seem to have any influence on the game strength. Since the bug occurs rarely, it is not easy to take a screenshot.
0.19 (average time use) is already a lot better than 16, but needs a small extra push.
Dio likes this post
Dio
Posts : 222 Join date : 2021-08-28
Subject: Re: News about Rebel or Chess System Tal ? Fri Dec 02, 2022 9:00 pm
Time controls such as 40 moves in 4 minutes and repeating date from a time when chess computers were tested (they had only this timing).
Even Stockfish 15.0 has considerable problems with this old time control (Andreas Matthies, Rubichess), i.e. Stockfish very often gets into time trouble and has to answer even after very short thinking time). Another candidate is LCZero...
Sponsored content
Subject: Re: News about Rebel or Chess System Tal ?