Mclane, Peter Berger, matejst, Damir Desevac, Dio, Eelco, Ipmanchess and Ghppn like this post
The author of this message was banned from the forum - See the message
Krisnatoon
Posts : 7 Join date : 2022-04-30
Subject: Re: REBEL 16.3 Mon Mar 11, 2024 5:01 pm
Geweldig, heel veel dank!
Admin likes this post
Dann Corbit
Posts : 189 Join date : 2020-11-26
Subject: Re: REBEL 16.3 Mon Mar 11, 2024 8:53 pm
Thank you kindly. I got my copy. People get these engines and play online to collect a rating and never even look at the games. Ed and Chris create engines that spit out fun games. What more could we ask for? Well, truthfully we shall never be satisfied. Like Oliver Twist, I stick out my bowl and say, "Please Sir, I want some more!"
Admin likes this post
Raymond
Posts : 2 Join date : 2023-01-03
Subject: Re: REBEL 16.3 Tue Mar 12, 2024 12:47 am
Helemaal top!! Dankjewel! Ik ben al fan sinds rebel 6!
Admin likes this post
Admin Admin
Posts : 2608 Join date : 2020-11-17 Location : Netherlands
Subject: Re: REBEL 16.3 Tue Mar 12, 2024 7:29 am
Raymond wrote:
Helemaal top!! Dankjewel! Ik ben al fan sinds rebel 6!
On top my head Rebel 6 was 1994, that's 30 years ago
Admin Admin
Posts : 2608 Join date : 2020-11-17 Location : Netherlands
Subject: Re: REBEL 16.3 Tue Mar 12, 2024 8:16 am
Dann Corbit wrote:
Thank you kindly. I got my copy. People get these engines and play online to collect a rating and never even look at the games. Ed and Chris create engines that spit out fun games. What more could we ask for? Well, truthfully we shall never be satisfied. Like Oliver Twist, I stick out my bowl and say, "Please Sir, I want some more!"
Keep on wishing more
I have a question for you Dann and maybe you can help. I am using SF10 with the bench function to analyze the 4-6 men positions with the Syzygy bases. At depth=1 SF10 is lightening fast doing 1 million positions. But then in the next run it suddenly may crash after 27.000 positions or so. Clearing the Syzygy Path there is no crash. So it seems that the Syzygy in SF10 has some kind of bug.
Since SF12 I can not compile the program any longer, SF team doesn't support Visual Studio. One solution would be a SF 16.1 compile with the following change :
Or maybe there are other (fast) solutions / tools?
The idea is to update all unclear scores of 4-6 men positions in my data or even one step further create an endgame neural net.
Dann Corbit
Posts : 189 Join date : 2020-11-26
Subject: Re: REBEL 16.3 Tue Mar 12, 2024 8:28 am
Admin wrote:
Dann Corbit wrote:
Thank you kindly. I got my copy. People get these engines and play online to collect a rating and never even look at the games. Ed and Chris create engines that spit out fun games. What more could we ask for? Well, truthfully we shall never be satisfied. Like Oliver Twist, I stick out my bowl and say, "Please Sir, I want some more!"
Keep on wishing more
I have a question for you Dann and maybe you can help. I am using SF10 with the bench function to analyze the 4-6 men positions with the Syzygy bases. At depth=1 SF10 is lightening fast doing 1 million positions. But then in the next run it suddenly may crash after 27.000 positions or so. Clearing the Syzygy Path there is no crash. So it seems that the Syzygy in SF10 has some kind of bug.
Since SF12 I can not compile the program any longer, SF team doesn't support Visual Studio. One solution would be a SF 16.1 compile with the following change :
Not sure why my message got truncated. Here is a test that automatically loads and uses the tablebase files without being told where to look C:\Users\dannc\chess\Stockfish-master2\src>stockfish-x64-avx2.exe Stockfish dev-20240312-nogit by the Stockfish developers (see AUTHORS file) info string Found 145 tablebases bench 32767 30 60 h:\sparse.epd info string Found 145 tablebases
=========================== Total time (ms) : 1512 Nodes searched : 75685664 Nodes/second : 50056656
Dann Corbit
Posts : 189 Join date : 2020-11-26
Subject: Re: REBEL 16.3 Tue Mar 12, 2024 8:33 pm
Warning to others: This special build for Ed assumes that the tablebase files are found here: G:\syzygy\3-4-5 G:\syzygy\dtz G:\syzygy\wdl if you are not Ed and you want to use it, you will need to use UCI parameters to create appropriate defaults
Admin Admin
Posts : 2608 Join date : 2020-11-17 Location : Netherlands
SF doesn't crash on the first error and eval is good nevertheless but does crash on the second error and the analysis process is aborted.
After downloading the 2 files from Lichess and replacing the bad ones SF finishes the analysis process. At least now I know how to fix things in case there are more corrupted Syzygy files.
My UHO-Top15 Ratinglist is the world's first engine-ratinglist, using UHO-openings, and the world's first ratinglist offering additionally Gamepair-statistics.
(Perhaps you have to clear your browsercache (press STRG+SHIFT+DEL) or reload the website))
Testrun of Rebel 16.3 finished: +29 Elo to Rebel EAS A small progress in the EAS Ratinglist too (Rebel 16.3 plays less sacrifices than Rebel EAS, but the other stats (short wins, bad draws, overall length of all won games) are an improvement, so the overall EAS-Score is a small progress). Here the EAS-calculation out of the 120000 games of the UHO-Top15 Ratinglist (before and after replacing Rebel EAS with Rebel 16.3), so exact the same opponents, same number of games, same openings... (in the EAS-list calculated out of the gamebase of the full ratinglist, Rebel EAS has played more games versus more opponents, so the EAS-score of Rebel EAS is little bit lower there)
Code:
bad avg.win Rank EAS-Score sacs shorts draws moves Engine/player ------------------------------------------------------------------- 6 95445 14.72% 12.63% 18.42% 80 Rebel 16.3 avx2 6 93130 19.72% 11.34% 19.19% 83 Rebel EAS avx2
Admin, matejst, Dio and Ipmanchess like this post
Admin Admin
Posts : 2608 Join date : 2020-11-17 Location : Netherlands
I should be working on EAS again, ah, I have already started with it.
pohl4711 likes this post
Eelco
Posts : 233 Join date : 2021-10-08
Subject: Re: REBEL 16.3 Wed Apr 03, 2024 9:45 pm
Peter Berger wrote:
Unfortunately this turned out to be not true.
With a very long timecontrol, Rebel on my computer, CSTal I think as well, starts outputting what move it is currently thinking about (not with a PV, just a move from the movelist). My guess is this is so the user can check the engine is still thinking and also not to lose contact with the GUI? This in Shredder interface but if you are using Chessbase my guess would be you do not see all that because Chessbase only outputs the very last line from an engine. At least that was my experience with the free versions (Chessbase Reader). So it is not impossible that the engine loses contact as it were, just goes thinking in the void. But this is just one possibility? I tried one very long run a while ago in Shredder 11 and Rebel 16.3 did not crash, for a very long time. I lost the first page of analysis because of all these single line outputs that filled up the first page This came on the second page:
1r3rk1/3bpppp/1qn2P2/p2pP1P1/3P4/2PB1N2/2p3K1/qNBQ1R2 w - -
Engine: Rebel-16.3 (512 MB) gemaakt door Ed Schroder and Chris Whittington (C) 2
beste zet: Ld3xh7 tijd: 4777:46.171 min nodes: 163.960.340.142
Dann Corbit
Posts : 189 Join date : 2020-11-26
Subject: Re: REBEL 16.3 Wed Apr 03, 2024 9:58 pm
Very often chess engines and chess GUIs lose contact with each other due to buffered I/O. I generally put this in chess programs at the start of main():
That turns off buffering for C input and output and for C++ input and output. It won't fix every problem, because most chess GUIs also use buffered I/O. The irony is that there is no performance benefit, so turning off buffering causes no slowdown.
Peter Berger
Posts : 131 Join date : 2020-11-20
Subject: Re: REBEL 16.3 Sat Apr 06, 2024 4:08 pm
Eelco wrote:
I tried one very long run a while ago in Shredder 11 and Rebel 16.3 did not crash, for a very long time. I lost the first page of analysis because of all these single line outputs that filled up the first page
Thanks for your feedback, Eelco. I am currently investigating as Rebel 16.3 runs stable enough here to be useable for me in general. The phenomenon is a bit hard to track down as there are several positions where Rebel runs absolutely fine even for some days. The crashes seem to happen somewhat more often if the position is very complicated with lots of pieces on the board. I am currently trying whether lower hash sizes make a difference in stability. As always my feedback will take more time, as I do a lot of holidays from computerchess.
Peter Berger
Posts : 131 Join date : 2020-11-20
Subject: Re: REBEL 16.3 Sat Apr 06, 2024 6:59 pm
On my computer the following position is exceptionally reliable at crashing Rebel 16.3 in infinite analysis after like one hour at most more or less regardless of settings:
[puzzle] [SetUp "1"] [FEN "r2q3r/4bk1p/3p1nbB/1B2p3/pp2P3/6P1/PPPQ4/R1N1K2R w KQ - 0 1"] [/puzzle]
FEN without graphics: r2q3r/4bk1p/3p1nbB/1B2p3/pp2P3/6P1/PPPQ4/R1N1K2R w KQ - 0 1
If no one can reproduce this, I am out of testing ideas for now.
Kind regards Peter
Last edited by Peter Berger on Sat Apr 06, 2024 7:01 pm; edited 1 time in total
Dann Corbit
Posts : 189 Join date : 2020-11-26
Subject: Re: REBEL 16.3 Sat Apr 06, 2024 7:01 pm
Peter Berger wrote:
Eelco wrote:
I tried one very long run a while ago in Shredder 11 and Rebel 16.3 did not crash, for a very long time. I lost the first page of analysis because of all these single line outputs that filled up the first page
Thanks for your feedback, Eelco. I am currently investigating as Rebel 16.3 runs stable enough here to be useable for me in general. The phenomenon is a bit hard to track down as there are several positions where Rebel runs absolutely fine even for some days. The crashes seem to happen somewhat more often if the position is very complicated with lots of pieces on the board. I am currently trying whether lower hash sizes make a difference in stability. As always my feedback will take more time, as I do a lot of holidays from computerchess.
I wonder if an array bound such as the move list or the pv could be overrun. Some deep searches are simple and it is easy to get 200 ply and for others 60 plies takes all day. Here are some sample asserts from Komodo which were exercised in debug mode before writing to an array. In each case, the macro parameter is the array index assert( lmc < 200 ); // legal move count assert( ply < 256 ); assert( to < 64 ); assert( fr < 64 ); assert( me < 2 ); assert( thr < 20 ); assert( ty < 6 ); Add an assert to any array insert and compile in debug mode. Then using an EPD that can reproduce the problem, let it run all night. Running it in the debugger, it will stop right on the problem line.
Home 
This came on the second page: