REBEL-EAS

Subject: REBEL-EAS Mon Aug 21, 2023 7:22 pm

REBEL-EAS is an in between version based on the same, but more, neural net data as REBEL-16.2 but heavily modified during the last 5 months in order to make REBEL to play more aggressive and runs as a module under Chess System Tal 2.0

. REBEL 16.2 gains an EAS score of : 80.258
. REBEL-EAS gets an EAS score of : 160.569

As a cherish on the cake the playing strength of the neural net has increased with 20-25 elo using balanced positions. Using the TCEC positions composed by Jeroen Noomen and GM Matthew Sadler adds another 26 elo points while the draw rate dropped with almost 20%.

....

https://rebel7775.wixsite.com/rebel/blank-2

Posts : 222 Join date : 2021-08-28

first test:


Eigenmann ERET test, 15s/move, 111 positions
Stockfish 16.0: 92/111 06:04
Komodo Dragon 3.2: 90/111 06:35
Stoovlees II b5: 83/111 09:01 
Berserk 11.1: 83/111 09:21
CSTal 2.0: 77/111 10:40
Revenge 3.0: 77/111 10:51
Uralochka 3.39d: 75/111 11:17
Caissa 1.11: 75/111 11:59
Rebel 16.2: 74/111 11:22 *
Rebel EAS: 73/111 11:19 *
Velvet 5.3.0: 74/111 11:53
Alexandria 4.0.0: 72/111 12:25
Booot 7.2: 72/111 12:3ß
RofChade 3.0: 72/111 12:35
Deep Sjeng 3.6 a16: 70/111 12:54
LGEvolution 3.16.0: 70/111 12:54
Black Marlin 8.0: 64/111 14:09
PowerFritz 18: 62/111 16:46
Hiarcs 15.2: 61/111 16:14

Posts : 222 Join date : 2021-08-28

[Event "Fast"]
[Site "isle of Rügen"]
[Date "2023.08.21"]
[Round "1"]
[White "Ethereal 14.00"]
[Black "Rebel EAS"]
[Result "0-1"]
[BlackElo "2200"]
[ECO "A48"]
[Opening "Neo-Königsindisch"]
[Time "20:47:36"]
[Variation "Torre-Angriff, 2..g6 3.Bg5 Bg7 4.Nbd2"]
[WhiteElo "2200"]
[TimeControl "1+0"]
[Termination "adjudication"]
[PlyCount "81"]
[WhiteType "program"]
[BlackType "program"]

1. d4 Nf6 2. Nf3 g6 3. Bg5 Bg7 4. Nbd2 c5 5. e3 d6 6. c3 O-O 7. Bd3 h6 8.
Bh4 Nc6 9. O-O {+0.37/16 0} cxd4 {+0.20/14 0} 10. exd4 {+0.16/19 1} e5
{+0.43/15 0} 11. dxe5 {+0.26/20 0} dxe5 {+0.62/15 0} 12. Bb5 {+0.01/20 0}
Qc7 {+0.48/14 0} 13. Re1 {-0.13/22 0} Nh5 {-0.27/13 0} 14. Nc4 {-0.02/22 0}
Be6 {+0.01/14 0} 15. Qd6 {-0.16/20 0} Rfc8 {-0.10/15 0} 16. Rad1 {-0.12/17
0} a6 {+0.02/15 0} 17. Bxc6 {+0.33/20 0} Bxc4 {+0.38/16 0} 18. Bd7
{+0.45/19 0} Qxd6 {-0.08/16 0} 19. Rxd6 {-0.14/19 0} Rc5 {0.00/17 0} 20. b3
{-0.13/21 0} e4 {-0.04/19 1} 21. Nd2 {-0.08/20 0} Bd3 {-0.32/18 0} 22. Nxe4
{-0.13/20 0} Bxe4 {+0.39/16 0} 23. Rxe4 {-0.07/21 0} g5 {+1.72/17 0} 24.
Bg3 {-0.08/22 0} f5 {+1.51/17 0} 25. Re7 {-1.45/20 3} Kf8 {+2.40/16 0} 26.
Rde6 {-2.40/16 0} f4 {+3.07/17 0} 27. Be8 {-2.63/19 0} Nf6 {+2.82/18 0} 28.
Bg6 {-2.76/18 0} fxg3 {+2.97/15 0} 29. hxg3 {-3.42/18 1} Rxc3 {+3.23/19 0}
30. Rf7+ {-2.35/4 0} Kg8 {+3.25/15 0} 31. Kf1 {-3.96/20 0} Rc1+ {+3.99/12
0} 32. Ke2 {-0.98/4 0} Rc2+ {+4.97/13 0} 33. Kf1 {-4.23/20 0} Rxa2
{+4.96/15 0} 34. f3 {-4.74/18 0} Rd8 {+5.64/16 0} 35. Re1 {-6.79/18 1} Rdd2
{+7.00/15 0} 36. Rc7 {-6.49/4 0} Bf8 {+8.10/14 0} 37. Rec1 {-11.00/19 0}
Ba3 {+10.74/14 0} 38. Re1 {-6.39/6 0} Rxg2 {+14.43/15 0} 39. Bc2 {-17.46/21
0} Rgxc2 {+15.85/17 0} 40. Rxc2 {-11.60/4 0} Rxc2 {+15.97/17 0} 41. Re6
{-17.11/22 0 Arena Entscheidung} 0-1

Posts : 42 Join date : 2022-06-08

Code:: My Tournament 2023

1 Chess-System-Tal 2.00 avx512 +29 +42/=241/-17 54.17% 162.5/300
2 REBEL-EAS -29 +17/=241/-42 45.83% 137.5/300

TC: 10s+1s

Subject: Re: REBEL-EAS Mon Aug 21, 2023 10:53 pm

Ipmanchess wrote:

Code:: My Tournament 2023

1 Chess-System-Tal 2.00 avx512 +29 +42/=241/-17 54.17% 162.5/300
2 REBEL-EAS -29 +17/=241/-42 45.83% 137.5/300

TC: 10s+1s

Seems about right.

Posts : 160 Join date : 2022-03-01 Location : Berlin

Admin wrote:: REBEL-EAS is an in between version based on the same, but more, neural net data as REBEL-16.2 but heavily modified during the last 5 months in order to make REBEL to play more aggressive and runs as a module under Chess System Tal 2.0

. REBEL 16.2 gains an EAS score of : 80.258
. REBEL-EAS gets an EAS score of : 160.569

I started my testruns (SPCC-Ratinglist and UHO-Top10-Ratinglist). Will take 4 days (10000 games UHO-Top10 Testrun and 16000 games SPCC-Ratinglist testrun). Then we will see, what the EAS-scoring of Rebel-EAS really is - in my EAS-Ratinglist (calculated out of my SPCC-Ratinglist gamebase)...

Posts : 9 Join date : 2021-07-02

Admin wrote:: REBEL-EAS is an in between version based on the same, but more, neural net data as REBEL-16.2 but heavily modified during the last 5 months in order to make REBEL to play more aggressive and runs as a module under Chess System Tal 2.0

. REBEL 16.2 gains an EAS score of : 80.258
. REBEL-EAS gets an EAS score of : 160.569

As a cherish on the cake the playing strength of the neural net has increased with 20-25 elo using balanced positions. Using the TCEC positions composed by Jeroen Noomen and GM Matthew Sadler adds another 26 elo points while the draw rate dropped with almost 20%.

....

https://rebel7775.wixsite.com/rebel/blank-2

Congrats, Ed!
Are broken book issue fixed? Thanks

Subject: Re: REBEL-EAS Tue Aug 22, 2023 9:22 am

pohl4711 wrote:

Admin wrote:: REBEL-EAS is an in between version based on the same, but more, neural net data as REBEL-16.2 but heavily modified during the last 5 months in order to make REBEL to play more aggressive and runs as a module under Chess System Tal 2.0

. REBEL 16.2 gains an EAS score of : 80.258
. REBEL-EAS gets an EAS score of : 160.569

I started my testruns (SPCC-Ratinglist and UHO-Top10-Ratinglist). Will take 4 days (10000 games UHO-Top10 Testrun and 16000 games SPCC-Ratinglist testrun). Then we will see, what the EAS-scoring of Rebel-EAS really is - in my EAS-Ratinglist (calculated out of my SPCC-Ratinglist gamebase)...

Cool....

I don't hold my breath about the really, EAS scores fluctuate on 1) the engine mix, different opponents gives different EAS, 2) used opening sets, REBEL-EAS does best on balanced positions as I found out and 3) of course a valid elo pool.

For my testing I use 10 fixed opponents within a range of -30|+30 elo using balanced positions. During a training run of ~400 epochs I play about 15-20 matches of 2000 games and calculate the EAS which (for REBEL-EAS) fluctuate between 100.000 and 200.000. Lumping them together (at least 20.000 games) gives a reliable EAS score, but..... 1) for that particular elo pool and 2) that particular opening set.

And so I was able to make a comparison between REBEL-16 and REBEL-EAS but only for that particular elo pool and that particular (balanced) opening set.

I tried other sets, randomly picked one of your UHO opening sets, tried my gambit positions and the EAS score dropped significantly. And what to say about the TCEC positions, REBEL-EAS gained 26 elo [!] more than all other sets but the EAS was not so great, but in all cases always better than REBEL-16.2

Adding all things together I managed to create a more aggressive engine and that's hard enough with NNUE and your EAS tool was very helpful, actually the best what is available, but to say and I quote "Then we will see, what the EAS-scoring of Rebel-EAS really is" as a sort of absolute truth are a couple of bridges too far as I stipulated in the above, best summarized as, it depends.

Subject: Re: REBEL-EAS Tue Aug 22, 2023 9:48 am

TigreNegro wrote:

Admin wrote:: REBEL-EAS is an in between version based on the same, but more, neural net data as REBEL-16.2 but heavily modified during the last 5 months in order to make REBEL to play more aggressive and runs as a module under Chess System Tal 2.0

. REBEL 16.2 gains an EAS score of : 80.258
. REBEL-EAS gets an EAS score of : 160.569

As a cherish on the cake the playing strength of the neural net has increased with 20-25 elo using balanced positions. Using the TCEC positions composed by Jeroen Noomen and GM Matthew Sadler adds another 26 elo points while the draw rate dropped with almost 20%.

....

https://rebel7775.wixsite.com/rebel/blank-2

Congrats, Ed!
Are broken book issue fixed? Thanks

If you activate "bookdepth" then yes.

Posts : 222 Join date : 2021-08-28

a first small tournament with fast TC, 40/4-test are also running..

Fast

Rank	Engine	Score	%	St	Ko	Be	Cl	CS	Et	Re	Re	Re	Ca	De	Ur	Ro	Wa	Po	S-B
01	Stockfish 16	540.5/700	77.2	· ·· ·· ·	11-0-39	19-1-30	21-0-29	22-1-27	25-1-24	27-0-23	24-0-26	30-0-20	27-0-23	33-0-17	33-0-17	36-0-14	37-0-13	39-0-11	177837,50
02	Komodo Dragon 3.2	469.5/700	67.0	0-11-39	· ·· ·· ·	7-5-38	8-8-34	16-1-33	17-1-32	17-0-33	17-0-33	15-0-35	18-3-29	28-0-22	28-2-20	32-0-18	31-0-19	39-3-8	152877,30
03	Berserk 11.1	443.5/700	63.3	1-19-30	5-7-38	· ·· ·· ·	17-4-29	11-3-36	10-4-36	18-3-29	9-4-37	14-0-36	20-0-30	19-1-30	26-0-24	28-1-21	32-1-17	25-1-24	144981,30
04	Clover 6.0	394.5/700	56.3	0-21-29	8-8-34	4-17-29	· ·· ·· ·	8-8-34	3-8-39	10-4-36	5-1-44	11-5-34	6-2-42	18-1-31	19-1-30	18-0-32	28-0-22	27-0-23	128510,30
05	CSTal 2.00	380.5/700	54.3	1-22-27	1-16-33	3-11-36	8-8-34	· ·· ·· ·	9-8-33	4-3-43	9-1-40	5-4-41	13-4-33	11-4-35	16-2-32	23-2-25	21-3-26	26-1-23	124353,30
06	Ethereal 14.00	371.5/700	53.0	1-25-24	1-17-32	4-10-36	8-3-39	8-9-33	· ·· ·· ·	10-5-35	8-7-35	6-2-42	9-4-37	13-6-31	11-5-34	13-2-35	20-1-29	27-0-23	121821,80
07	Revenge 3.0	363.5/700	51.9	0-27-23	0-17-33	3-18-29	4-10-36	3-4-43	5-10-35	· ·· ·· ·	3-9-38	4-6-40	10-4-36	10-0-40	19-1-30	23-1-26	22-1-27	30-1-19	116833,00
08	Rebel EAS	353.5/700	50.5	0-24-26	0-17-33	4-9-37	1-5-44	1-9-40	7-8-35	9-3-38	· ·· ·· ·	2-2-46	4-3-43	12-2-36	11-4-35	11-7-32	21-2-27	22-3-25	116528,00
09	Rebel 16.2	347.0/700	49.5	0-30-20	0-15-35	0-14-36	5-11-34	4-5-41	2-6-42	6-4-40	2-2-46	· ·· ·· ·	6-3-41	7-1-42	12-1-37	14-4-32	19-0-31	17-4-29	113684,80
10	Caissa 1.11	333.5/700	47.6	0-27-23	3-18-29	0-20-30	2-6-42	4-13-33	4-9-37	4-10-36	3-4-43	3-6-41	· ·· ·· ·	7-3-40	10-3-37	12-4-34	19-1-30	21-1-28	108986,50
11	Deep Sjeng 3.6 a16	297.5/700	42.5	0-33-17	0-28-22	1-19-30	1-18-31	4-11-35	6-13-31	0-10-40	2-12-36	1-7-42	3-7-40	· ·· ·· ·	7-11-32	19-3-28	13-5-32	18-3-29	96865,25
12	Uralochka 3.39d	284.5/700	40.6	0-33-17	2-28-20	0-26-24	1-19-30	2-16-32	5-11-34	1-19-30	4-11-35	1-12-37	3-10-37	11-7-32	· ·· ·· ·	9-7-34	12-4-34	23-2-25	92213,50
13	RofChade 3.0	255.0/700	36.4	0-36-14	0-32-18	1-28-21	0-18-32	2-23-25	2-13-35	1-23-26	7-11-32	4-14-32	4-12-34	3-19-28	7-9-34	· ·· ·· ·	13-5-32	12-3-35	83378,00
14	Wasp 6.50	221.0/700	31.5	0-37-13	0-31-19	1-32-17	0-28-22	3-21-26	1-20-29	1-22-27	2-21-27	0-19-31	1-19-30	5-13-32	4-12-34	5-13-32	· ·· ·· ·	13-6-31	72957,25
15	PowerFritz 18	194.5/700	27.7	0-39-11	3-39-8	1-25-24	0-27-23	1-26-23	0-27-23	1-30-19	3-22-25	4-17-29	1-21-28	3-18-29	2-23-25	3-12-35	6-13-31	· ·· ·· ·	65618,50

5250 games played / Tournament finished

Tournament start: 2023.02.22, 17:59:20
Latest update: 2023.08.23, 13:46:03
Site/ Country: isle of Rügen, Germany
Level: Blitz 0:01/0,5
Hardware: Intel(R) Core(TM) i3-7100 CPU @ 3.90GHz mit 7,9 GB Speicher
Operating system: Windows 10 Enterprise Professional (Build 9200) 64 bit
Table created with: Arena 3.5.1

Subject: Re: REBEL-EAS Wed Aug 23, 2023 7:38 pm

Clover is going surprising fast.

Posts : 222 Join date : 2021-08-28

Clover has tremendous potential. With a time control of 4 min + 2 sec, however, it is still weaker than CSTal 2.00. In the current TCEC tournament, Clover has qualified for the main tournament.
The author is 18 years old and comes from Romania.

https://www.chessprogramming.org/Luca_Metehau

Subject: Re: REBEL-EAS Wed Aug 23, 2023 8:16 pm

At 18 he should be after girls.

Posts : 222 Join date : 2021-08-28

running..

https://cegt.forumieren.com/t1972-testing-rebel-eas-x64-1cpu

Subject: Re: REBEL-EAS Thu Aug 24, 2023 10:24 am

So far +20, for once my elo gain prediction could be right Wink

Posts : 42 Join date : 2022-06-08

Rebel-EAS added to new list: https://ipmanchess.yolasite.com/r9-7945hx.php

Ipman.

Posts : 160 Join date : 2022-03-01 Location : Berlin

Ratinglist-testruns of Rebel EAS finished.

https://www.sp-cc.de

https://www.sp-cc.de/uho_ratinglist.htm

Also take a look at the EAS-Ratinglist, the world's first engine-ratinglist not measuring strength of engines but engines's style of play:
https://www.sp-cc.de/eas-ratinglist.htm

(Perhaps you have to clear your browsercache (press STRG+SHIFT+DEL) or reload the website))

We see a clear progress in the EAS-Ratinglist: Number of sacs is quite good (25.64%) und much better than Rebel 16.1 (17.22%). Number of bad draws is clearly smaller, too (progress,too) for Rebel EAS. But the overall length of all won games is worse (84) compared to Rebel 16.1 (80) and the number of short wins is a little bit worse (no big deal, but 11.98% short wins is overall not a good value).
So a lot of room for improvements for Rebel in the EAS-List: A new Rebel, which is winning games faster, will lower the overall length of all won games and will raise the number of short wins, too...both would mean a gain of EAS-points...

Code::    bad avg.win
Rank EAS-Score sacs shorts draws moves Engine/player
-------------------------------------------------------------------
  27 98215 25.64% 11.98% 20.41% 84 Rebel EAS avx2

  36 82699 17.22% 12.96% 24.10% 80 Rebel 16.1

By the way: The EAS-Ratinglist built out of the UHO-Top10 Ratinglist games show exactly the same as the "normal" EAS-Ratinglist:
https://www.sp-cc.de/files/uho_statistics_eas_ratinglist.txt

Code::    bad avg.win
Rank EAS-Score sacs shorts draws moves Engine/player
-------------------------------------------------------------------
   1 178580 21.45% 23.84% 09.95% 71 Stockfish 16 230630
   2 171649 20.28% 22.71% 09.72% 71 Stockfish 230807 avx2
   3 120433 16.42% 23.52% 16.51% 71 KomodoDragon 3.2 avx2
   4 118350 13.87% 24.61% 16.91% 73 Revenge 3.0 avx2
   5 111863 18.91% 22.83% 19.73% 71 RubiChess 230410 avx2
   6 75890 18.03% 10.14% 18.81% 85 Rebel EAS avx2
   7 75379 12.14% 11.39% 18.42% 83 CSTal 2.0 avx2
   8 72965 15.69% 07.02% 17.16% 88 Ethereal 14.00 nnue
   9 55126 07.16% 12.86% 23.10% 79 Clover 6 avx2
  10 46298 08.24% 10.29% 23.32% 84 Berserk 11 avx2
  11 43194 06.00% 06.29% 22.08% 88 Koivisto 9.2 avx2
-------------------------------------------------------------------
*** Average length of all won games: 77 moves

(Number of sacs and bad draws is good for Rebel EAS, overall length of won games and number of short wins are bad. What we learn is, that the EAS-Tool produces very similar results (in the single statistics, compared to other engines (do not just look at the EAS-score!!!)) in different databases, if the number of played games is big enough).

Subject: Re: REBEL-EAS Sat Aug 26, 2023 10:06 am

Surprised by the elo gain, must be the longer time control.

Thanks for playing 16,000 games, that helps too.

Subject: Re: REBEL-EAS Sat Aug 26, 2023 10:18 am

Regarding EAS, I keep on saying that, in my test elo pool against 10 other engines the pool varies -30/+30 elo and then REBEL-EAS is clearly on top of the rest.

But nevermind, the next version (if ever) the focus will be on playing strength, the EAS version was just a fun exercise.

Posts : 42 Join date : 2022-06-08

Rebel-EAS added to a newer list ,also with Lco : https://ipmanchess.yolasite.com/i7-13700h.php

Ipman.

Posts : 222 Join date : 2021-08-28

+20 Elo after 800 games, +20 Elo after 2200 games.

https://cegt.forumieren.com/t1972-testing-rebel-eas-x64-1cpu#3647

-21 Elo compared with CSTal 2.00

Posts : 42 Join date : 2022-06-08

Rebel-EAS now also in 3th list: https://ipmanchess.yolasite.com/i9-7980xe.php

Here it played 32000games ,TC: 10s+1s

Ipman.

Posts : 131 Join date : 2020-11-20

Admin wrote:

TigreNegro wrote:

Admin wrote:: REBEL-EAS is an in between version based on the same, but more, neural net data as REBEL-16.2 but heavily modified during the last 5 months in order to make REBEL to play more aggressive and runs as a module under Chess System Tal 2.0

. REBEL 16.2 gains an EAS score of : 80.258
. REBEL-EAS gets an EAS score of : 160.569

As a cherish on the cake the playing strength of the neural net has increased with 20-25 elo using balanced positions. Using the TCEC positions composed by Jeroen Noomen and GM Matthew Sadler adds another 26 elo points while the draw rate dropped with almost 20%.

....

https://rebel7775.wixsite.com/rebel/blank-2

Congrats, Ed!
Are broken book issue fixed? Thanks

If you activate "bookdepth" then yes.

This doesn't work for me.

I am using the Fritz17 GUI, does it work for others?

Subject: Re: REBEL-EAS Fri Dec 08, 2023 10:52 am

Install REBEL-EAS.EXE in the same folder as Chess System Tal 2.0

Book are stored in the polyglot-books folder.

Posts : 131 Join date : 2020-11-20

Admin wrote:: Install REBEL-EAS.EXE in the same folder as Chess System Tal 2.0

Book are stored in the polyglot-books folder.

This doesn't work for me.

As Rebel EAS is working engine-wise I assume I have set up the engine correctly in principle.

Also the book is working with Rebel 16.2, so I assume the book is set up correctly, too.

I am out of ideas.

Happy Christmas time!
Peter

» SPCC: Testruns of Rebel EAS 2 and Rebel Extreme finished
» Rebel in my FACE -Tournaments
» Rebel, CS Tal
» Rebel EAS and CS Tal Bug
» Rebel 14.1