Cyclone 2.0 released

[kranium]

Cyclone 2.0 is now available

2.0 is a major departure from 1.0, which had an emphasis on features and configurablility (with 60 configurable UCI parameters).

The development of version 2.0 has focused primarily on 2 things: implementing (fixing) Multi_PV, and increasing strength.

Multi_PV now works (up to 10 PVs can be displayed simultaneously).

Cyclone's strength has improved significantly through code optimization and the implementation of new search features.

Executables are available for 1, 2, 4, 8, and 16 cpus.

It is avalable for download here: http://www.cyclonechess.com/

Best Regards-
Norm

[Golem]

Thank you Norman for fixing the Multi-PV code, this is really a great functionnality when analyzing games.

PS : Do you have an ELO estimation about the improvment of your engine (compared with toga 1.4beta5c) ?

[kranium]

This is a very tempting question, from which i will (with some difficulty!) refrain from answering. There's nothing worse than setting peoples expectation high, ... then of course shit happens!

Yes, of course...i do have very accurate idea of cyclone 2.0's rating, after all I ran many many thousands of games against the most established engines that exist.

But protocol (and propriety) dictate silence here...i must allow the process to run it's course and let the expert testers that are all around us, (especially the CCRL, CEGT, SSFD, etc.) give it a fair, thorough, and impartial testing. (hopefully, the testing procedures i have used will not! be proven innacurate).

Speaking of testing procedures, i have a lot share here...I especially want to thank Marc Lacrosse, who has developed the MLmfl ultrafast testing method I describe below. I highly recommend it.

(If you're getting bored already, perhaps you should stop reading here because what follows is a bit mundane), but perhaps it is of great interest to fellow independant lone developers who don't have the kind of financial resources to hire staff, IMs, etc. like some of the top engine(s?)...

Marc has developed an exceptional testing method which enables any small individual developer or tester to obtain high quality results with a minimum of time. here's how it works:

Question:
if an engine wins at 15 secs per game, will the same result manifest itself at longer time controls? perhaps...maybe! ...probably? why not?
Choose 8 well tested engines for a reference database. I recommend that the engines chosen for the database should have been tested by the CCRL or CEGT with more than 1000 long time control games, so the ratings are very accurately known.

Rank Name Elo + - games score oppo. draws
1 Rybka 3 1-cpu 32-bit 165 27 26 512 73% -21 15%
2 Cyclone 2.0.163-1cpu 32 24 24 512 55% -4 20%
3 HIARCS 11.1 UCI 26 24 24 512 54% -3 18%
4 Shredder 11 UCI 18 24 24 512 53% -2 16%
5 Naum 3.1 9 24 24 512 51% -1 17%
6 Toga II 1.4.1SE -14 24 24 512 48% 2 19%
7 Fruit 2.3.1 -34 24 24 512 45% 4 19%
8 Glaurung 2.1 (JA) -70 24 24 512 40% 9 18%
9 Spike 1.2 Turin -133 25 26 512 31% 17 16%

The engines I chose are listed above. You'll see that they are staggered approx 32 ELO point apart. I then created a reference database of 1792 games (the 8 engines above in a 64 game round robin), all games starting from 32 known test positions (Noomen, most frequent lines, etc). Marc provides his own MFL start position PGN file for this pourpose. Each engine plays white and black for each of the 32 test positions. Then, make a change to the test engine, and run a 64 game gauntlet using the 32 test positions against the 8 opponents above…this results in 512 games (at average 34 secs per game...it takes around 3.5 hours). When the gauntlet is finished, add these games to the 1792 game reference database. the new database is 2304 games. analyze this file with bayeselo and output the results to a text file.

The result is proving to be amazingly accurate. (i.e. the results mirror closely the ratings and results that CEGT and CCRL get with long time control games)…but these games are played with as little as 6 seconds per side! i usually run my tests at 15 secs. any game is adjudicted as a win if both engines indocate that one side has acheived an advantage of 200 centipawns for 5 half moves in a row, so in this manner, one can get an accurate result of 512 games in 3.5 hours.!
In order to play games at such a fast ime control you'll also need as couple of dos utilities to kill the running processes quickly and efficiently, and to delay the beginning of the next game until the start position has be set. Marc has provided these....delayexec.exe, and pv.exe, along with his own executables that do the bulk of the work.

If you're interested in reading further, Marc has provided lot of info here:
http://chessbazaar.mylivepage.com/file/164_MLmfl_tests
and here:
http://users.skynet.be/mlcc/chessbazaar/index.html

Good Luck withyour testing!
Norm

[geots]

Im not sure about shit happening as you described- but for me it already has with this version. I am running it in chessbase as UCI, and it will not use over 64MB. I run the exe, and that is what it is set for. I run the exe for 1.0, and it is set for 128mb. And it runs at that. 64mb for 2.0 is basically useless for me. How do i change the hash in the exe when i run it- which it looks like i will obviously have to do. Secondly, when i run the 1.0 exe, the egbb probe for Daniels egbbs is already loaded, and 1.0 uses the egbb. the probe is not already loaded for 2.0 and of course it DOES NOT access egbb. I know how to load a probe in an ini file for a winboard engine- but there is no ini file here. So how do i load the probe so 2.0 will use egbb. Either i am a bit stupid, which is highly possible- or this version is sort of a mess.

Best,

[kranium]

Hi George,

there's no INI file, you must use the UCI config dialog to change the settings.

cyclone uses dynamic hash by default....(with 500000 nodes per second and 5 seconds per move = 64 MB as default).
if you know the NPS acheived, and the time control...you can enter it in the UCI config dialog under Nodes Per Second and Seconds per Move.
the hash size will be automatically calculated and set. (version 1.0 used 128 MB as default).

if you don't want to use dynamic hash...it can be disabled in the UCI config dialog.
uncheck the Dynamic Hash check box to disable...then simply check the Manual Hash check box...and set the hash size as you prefer.

there's more info here:
http://www.cyclonechess.com/dynamic_hash.html

the egbb access in 2.0 is disabled by default...simply go to the UCI config dialog and enable it. the default path is c:\\egbb. Of course, this can be changed...if need be.

good luck,
Norm

[mclane]

Cyclone 2.0 is now available

2.0 is a major departure from 1.0, which had an emphasis on features and configurablility (with 60 configurable UCI parameters).

thank you norman. how can we configure the 60 parameters ?
when i look into the UCI parameters that can be chose or setup by cyclone, there are only a few.

how to get 60 parameters ?

[geots]

Hi George,

there's no INI file, you must use the UCI config dialog to change the settings.

cyclone uses dynamic hash by default....(with 500000 nodes per second and 5 seconds per move = 64 MB as default).
if you know the NPS acheived, and the time control...you can enter it in the UCI config dialog under Nodes Per Second and Seconds per Move.
the hash size will be automatically calculated and set. (version 1.0 used 128 MB as default).

if you don't want to use dynamic hash...it can be disabled in the UCI config dialog.
uncheck the Dynamic Hash check box to disable...then simply check the Manual Hash check box...and set the hash size as you prefer.

there's more info here:
http://www.cyclonechess.com/dynamic_hash.html

the egbb access in 2.0 is disabled by default...simply go to the UCI config dialog and enable it. the default path is c:\\egbb. Of course, this can be changed...if need be.

good luck,
Norm

Thanks Norm. But there is no UCI config. dialogue in 2.0 . Are you saying copy this file from 1.0 to 2.0? If your answer is yes, i will do it- but i dont think chessbase will recognize or pay any attention to any config. files in UCI engine folders.


Best,

[mclane]

george there are a few parameters. and in the same way as in version 1.0 you can enter the NPS and time that is computed and the engine will automatically choose the appropriate amount of hash for this time control and the hardware you use.
just enter the NPS it does and the computation time you test.

[kranium]

Hi Thorsten,

i did quite alot of work to create the multitude of UCI parameters in 1.0...
but I got no feedback at all on them, and they created many lines of 'fat' code. if they were popular and being frequently used, i wasn't aware of it. therefore, they were removed for speed and ELO points.

if i find out that they are indeed wanted and utilized by many, ...i can add them back in for 2.1. <:

Norm

[geots]

Hi George,

there's no INI file, you must use the UCI config dialog to change the settings.

cyclone uses dynamic hash by default....(with 500000 nodes per second and 5 seconds per move = 64 MB as default).
if you know the NPS acheived, and the time control...you can enter it in the UCI config dialog under Nodes Per Second and Seconds per Move.
the hash size will be automatically calculated and set. (version 1.0 used 128 MB as default).

if you don't want to use dynamic hash...it can be disabled in the UCI config dialog.
uncheck the Dynamic Hash check box to disable...then simply check the Manual Hash check box...and set the hash size as you prefer.

there's more info here:
http://www.cyclonechess.com/dynamic_hash.html

the egbb access in 2.0 is disabled by default...simply go to the UCI config dialog and enable it. the default path is c:\\egbb. Of course, this can be changed...if need be.

good luck,
Norm

Thanks Norm. But there is no UCI config. dialogue in 2.0 . Are you saying copy this file from 1.0 to 2.0? If your answer is yes, i will do it- but i dont think chessbase will recognize or pay any attention to any config. files in UCI engine folders.


Best,

UCI config dialog is the window that pops up when you click on parameters in CB or engine options in other GUI's.

Got it. Thanks to Norm, Thorsten and Gabor. After Gabor's words, i made the changes in parameters. Never thought to look there


Best,