HAL9000

HAL9000
"It just isn't conceivable that you can design a program strong enough to beat players like me."

November 28, 2017

Android UCI engine update: Cfish 2017.11.27.ACH

This time, it's shared on time! The build is still hot and don't get your fingers burned. Thanks to Alexei..

GET Cfish 2017.11.27.ACH for arm7 and arm8

In case of trouble downloading from PC, use your mobile or refer to THIS POST.

105 comments:

Alex said...

Update for Komodo 10 & 11 - Fix Syzygy tablebase bug.

Unknown said...

K11 is such a trashy engine. Its beyond words you cannot install cfish into chessbase

Unknown said...

You can still use Cfish with Droidfish Chess for Android and every other Gui

Unknown said...

Thans for info Alex

Unknown said...
This comment has been removed by the author.
Rahul said...

SCTR1.1 chess engine available.

Unknown said...

Thanks for update Rahul.Alexei has already ported SCTR 1.0 (arm8 & arm7).

chysiddh14 said...

Hawk chess location of pgn needed help

Alex said...

"Hawk chess location of pgn needed help" - storage/emulated/0 (on my device)/Download/hawk-chess/pgn-game-libraries
You can also set your own path for games.

Unknown said...

Dear Alexei. can u compile the newest brainfish arm8 for android. thq b4

Unknown said...

Yes hello Alexei also reminding you SCTR 1.1e is massively improved! 3057 so far in CCRL 40/4 031217 (list).I hope you find time to port it.Thank you

Unknown said...

Hello Bram.hope you are well and enjoying chess on your mobile ��

Unknown said...
This comment has been removed by the author.
Unknown said...

Hii Lazar, thank u s much,, i hope u r well 2. n God always bless u :)

Alexei Chernakoff said...

My compilation Brainfish-031217(arm7,arm64v8) here:
https://app.box.com/s/s4fbjpzpsjda5jubg4znp3c63qr7ruds

Book to lay down along the way:/sdcard/book/Cerebellum_Light_Poly.bin

Unknown said...

Alexei,,u r the legend, thz so much :)

Unknown said...

Thanks Alexei !

Unknown said...

Alexei can you please update both stockfish and cfish to the latest
" A better contempt implementation for stockfish " 05/12/17 ? It looks exciting for Stockfish!

Alexei Chernakoff said...

My compilation Cfish-041217(arm7,arm64v8).
New:
A better contempt implementation for
Stockfish
https://app.box.com/s/r2mmdy0fappmbunf50cg2s0n478g8ubr

Unknown said...

Woohoo!! Thanks Alexei! Hope you do Stockfish too.By the way your latest Brainfish 03/12/17 is excellent.Very strong !

Unknown said...

Gingko engine fluctuates from showing uci option threads 1-6 to then showing 1-8 under droidfish when reloaded.Never seen this kind of change in uci options for an engine before.Gurcan mentioned something about this in his post.Alexei or Gurcan can you explain this ?

Unknown said...

Thank you for the newest Cfish 041217 Alexei and Gurcan. The dynamic contempt: is it set by default even though the value is physically set to 0 or should the contempt value be manually adjusted?

Thanks for your hard work

Alexei Chernakoff said...

Against the weaker engines, contempt can be up to 40. Developers still argue which contempt to set by default. Therefore, they have so far left this indicator = 0.

Alexei Chernakoff said...

Here the version of cfish-041217 is aggressive with contempt = 20 by default.
https://app.box.com/s/s2sqq8yd86wng355v15f6581yyj9ci3o

Unknown said...

Thanks for the explanation Alexei.i read the link with the discussion about comtempt provided in the commit.So above Cfish (contempt 20 default) is same as other cfish except for 20 contempt default? So if i use Cfish under droidfish i dont need to redownload your latest cfish? I can adjust it myself? I guess its useful for testing under CFA with default comtempt 20 ;-)

Alexei Chernakoff said...

Yes, I made it specially for games in CFA. Of course the best results of the patch with dynamic disdain showed at contempt = 40. I made more careful contempt = 20.

Unknown said...

Yes i saw the results from the contempt 40 tests.impressive.yes contempt at 20 should be interesting.It will also be interesting to see the default comtempt that will be finally decided upon in future Stockfish versions.Thanks for uploading;-)

Unknown said...

I wish i could explain. Rapidroid games set to 4 threads via engines.xml seem to run okay. Ginkgo will be rated around 3150 vs Andscacs which will be 3200-3210.

Unknown said...

Yes at least it works troublefree when set at 4 threads or less in both arm 7 & 8 and in fixed time per move too.Gurcan i must admit that a difference of 50 elo vs Andscacs in arm 7 rapidroid is disappointing and more than it probably should be looking at the rating lists for PC versions.Also on my device Andscacs arm 8 vs arm 7 has a speed difference of 90% vs 60% for Ginkgo arm 8 vs arm 7 so its all the more surprising.It would seem that Andscacs is preforming very well in Android whilst Ginkgo - arm 7 at least- is underperforrming.Alexei compiled some great binaries for Andscacs !

Unknown said...

Hello this is an open-ended question for all:

I would like to ask you for your opinion on a topic was was discussed on chesstroid 2016 = allocating hash. I still have a hard time figuring out what the best option is and now for the Android platform. I currently use 128 04 256 mb hash.

There are four methods mentioned:

1. give as much hash as you can or whatever is the threshhold of your device (mine is 4000mb)

2. houdini formula (T x S / 100) MB
For example if you use a Time Control of 10 minutes for 40 moves repeating, the average move time T = 15 seconds. On hardware that produces about 2,000 kN/s the optimal hash size would then be approximately (15 x 2,000 / 100) = 300 MB, in other words 256 MB or 512 MB would be the recommended values.

3. Komodo team says: Take the minutes as second and add the increment. Since Rapidroid uses 900+1 TC, for me that makes a test of (15+1) x 3 = 48 seconds. Therefore i run Komodo 10 on Galaxy Note 2 with hash display version of Droidfish. With 256MB hash, the usage trend goes like the following:
12" 16%
18" 24%
24" 32%
31" 41% => limit exceeded
33" 44%
35" 46%
39" 54%
42" 55%
52" 65%
61" 73%
70" 80%
84" 86%
104" 93%

According to Komodo practice, my 256MB setting is low and i need something like 48 / ( 31 / 41 * 40 ) * 256 = 406 MB for 900+1 TC.

4. compare knps output:
You may simply conduct a series of fixed time analysis from the starting position, using a different hash setting for each run, from the smallest to the biggest, one by one. You will notice a quick increase in smallest settings but it will diminish somewhere. You must then stop and take the setting where the knps don't increase anymore. The test duration of each run must match the TC you want to use.

Which method is the most accurate for the Android platform?

Thanks for your time!

Unknown said...
This comment has been removed by the author.
Unknown said...

Thanks Lazar for your advice! What device do you use? I use a Galaxy s7 with android 7.1

Unknown said...

I too have an S7 only at the moment..as my tablet is being repaired..broken screen.im still on Android Marshmallow.Biggest issue with s7 is cpu throttling due to heat.Extensive android testing especially in medium time controls atm is only really feasible with a rooted underclocked device..as Im sure you and Gurcan & Alexei etc know too well.Its this throttling by the way that prompted me to ask P.Osterlund to implement a max nps uci option for texel- which he kindly and surprisingly did ;-) I can now set texel to about say 65% the speed of a typical middlegame node speed (example 420k to 450k) on 1 core and play at 40/120 if i so wish with consistent speed and reproducible games.its great and only the various JA & AP rodent engines had the same feature previously.Ive also emailed the author of Chiron and Pedone with same request.its also very fun just to play against say a much slower Chiron thats still playing best play-no randomness.i firmly believe the obsession with strength will abait in coming years as engines reach "celestial" strengths...i like engines to play interesting/ humanlike chess more than sheer strength.I welcome any possible paradigm shift that AlphaZero may bring to the minimax models we have.exciting times for Computer Chess fans!!

Kevin said...

I like how AlphaZero tested with Stockfish. This could be the break Stockfish needs before releasing Version 9. It can learn a few gains of coding and strength from its AI enemy and come out way stronger.

Stockfish 8 dominated the floor for over a year, and now imagine Stockfish 9 thanks to the match with Alpha Zero, with the idea of it knows what to work on to become stronger.

Unknown said...

It sounds good in theory Kevin but im not sure thats how Stockfish development works.They will no doubt look at he games but im not sure they will be able to extrapolate anything to actually improve Stockfish's play in the next month or so.Of course Stockfish 9 may be delayed even longer than a month so you never know.. ;-)

Alexei Chernakoff said...

Hi Gurcan.Here my compilation fruit-reloaded 3.2.1(arm7,arm64v8):
https://app.box.com/s/fcdmtkcwok977m8kwfz53f0gipizgzx9

Unknown said...

alexei, if it didnt bother u. can u compile the newest brainfish. 06.12.17. thq s much

Unknown said...

Great Alexei! Thanks! May i ask, is SCTR 1.1f easily portable to Android? Have you looked at the Source Code at all.I know you ported version 1.0.Hoping you might do the same with version 1.1f as its much stronger ;-)

Unknown said...

Alexei is fruit arm 8 built with Static libraries or not necessary? Arm 8 binary is smaller than arm 7 ;-)

Alexei Chernakoff said...

My compilation Brainfish-061217:
https://app.box.com/s/xfhcm00lga56x713zv9v1mnj1enl27p0

Alexei Chernakoff said...

Dynamic library.

Alexei Chernakoff said...

Gurcan, the other day will be a pleasant surprise. I managed to persuade the author of CFA (Aart Bik) to make the awarding rule as in Cuteches. Now the shell will be able to decide whether to win or draw. This will save the test time. I think next week will release a new version of CFA.

Unknown said...

Ok nice.Alexei Brainfish without Cerebellum book is exactly like normal Stockfish?

Alexei Chernakoff said...

Yes.

Paulzzz said...

Hi Alexei,thank you for compiling Fruit Reloaded and the latest Brainfish!

Alexei Chernakoff said...

Write how the arm64v8 version of the fruit works (or does not work) and the type of processor if a problem occurs.

Unknown said...

Well it works perfectly on galaxy s7 exynos 8890.I was asking out of curiosity of course..though for other engines i like your compiles with Static
Libraries ;-)

Rahul said...

Is "laser chess" update available for android? and what happend about senpai2.0?

Unknown said...

I remember Alexei saying Senpai 2.0 was buggy.

Alexei Chernakoff said...

Sorry. I found an unpleasant bug in Fruit. The drift sometimes flies when it plays black.

Unknown said...

Lazar, what are your thoughts on the tournament conditions of the match between AlphaZero vs Stockfish? I am referring to the fact that no opening books or tablebases were used for Stockfish. From what I understand, Stockfish needs the opening book and tablebases to play at optimum level. The cpu Stockfish was running on at the time I think was 64 cpus and 60 sec per move for both entities. I have no idea as to the combined computing power behind Deep Mind.

Anonymous said...

Alexei,can you compile the new Sugar-061217 or the newest Shymphosodon?.Thanks so much.

Alex said...

Some results from my test - Cfish contempt=40/ against Cfish 8
Program Elo + - Games Score Av.Op. Draws

1 cfish-041217-arm64v8 : 3351 10 10 1000 57.3 % 3299 63.7 %
2 cfish_8_ACH_arm64v8 : 3299 10 10 1000 42.8 % 3351 63.7 %

Alex said...

Btw, i've done a lot of tests about Cfish, my good friend Alexei knows it :). Alexei buddy, don't stop provide to us new compilations!! :)

Unknown said...

Well im not sure i have anything original to say on the match conditions..lol.In a nutshell i think the conditions heavily favored AlphaZero.Due to reinforced learning A0 "knew" what optimal opening lines to play whereas Stockfish had to play the openings "on the fly" Endgames were same story.2nd point is A0 was running on much faster hardware and with enough hash to accomodate its 80k/s positions searched wheras Stockfish was given just 1gb hash to deal with 64 cores at 70 million positions per sec.Stockfish needed at least 40gb hash.Finally Stockfish benefits from managing its own time but this too was denied with fixed time.
Point is however that A0 has possibly matched the play of Stockfish by learning by from self-play and by calculating far fewer pos/sec.Its a paradigm shift that will come to fruition in the coming years.the hardware to emulate what A0 did is a few years away as far as the average comluter chess user is concerned.AI in future will change everything anyway..Chess Shogi and Go are just part of beginning of the Story...stay tuned :-D

Unknown said...

really thq s much Alexei. :)

Paulzzz said...

Fruit Reloaded arm8 works well in marshmallow flare s5 octacore,both in DroidFish and CEPA.

Unknown said...

It appears that Fruit R builds have both issues wlth SMP. I played two games with 2 cores: it resigned after 1-2 moves. In single core config, no issues.

Alexei Chernakoff said...

I'll think about this problem.

Rahul said...

Ethereal8.60 new update.

Rahul said...
This comment has been removed by the author.
Alexei Chernakoff said...

Gurcan, try this version of fruit:
https://app.box.com/s/x3vv108tsdqerlkk0dbv0akkezz7y644

Alex said...

New fruit on android 6, arm7 working fine with 3 cores, 1sec per move, with white and black, without resignations!

Alexei Chernakoff said...

Thanks for the test.

Unknown said...

Hello Alexei.Could you please compile Stockfish 11/12/17 "Simplify other checks" #1334.We haven't had a pure Stockfish compile (Brainfish excluded) since 18/11/17 i think?

Alexei Chernakoff said...

My compilation Cfish-111217 (arm7, arm64v8) now supports the book Cerebellum_Light_Poly.bin! The book is put in the [book] folder. The path to the book: /sdcard/book/Cerebellum_Light_Poly.bin.Download the engine here:
https://app.box.com/s/3q4pwmg27fvcva19kquishyt1ia0lnd1

Unknown said...

Thanks Alexei! Is that chessman3 cfish or original cfish?
Also there is compile of Armfish 231117 on mz chess forum posted by yonatan with download link :
https://app.box.com/s/o2h36n7bihv8h0ntwrywwbf79ocehk70
Do you know who yonatan is? Its not Lantonov right?
 Also any chance you can compile Hakkapaalitta 210416 tcec (dev branch)  from commit with same date from this webpage?:
 https://github.com/mAarnos/Hakkapeliitta/commits/06df898acfee7f07ae3eac7bae98422f3abbebc9

Its stronger than hakka 3.0 according ro ccrl 40/4
Thanks either way !

Unknown said...
This comment has been removed by the author.
Unknown said...

Texel 1.08a5 download package now available with arm8(pie) and arm7 (nopie) compiles only.Link is:

https://www.dropbox.com/sh/v16dye7ww33x7oe/AAAXmGY0phMkB7XlR-vaOxasa/test?dl=0&preview=texel108a5.7z

Rahul said...

Hi, Alexei there is a Amoeba chess and its source code avaivable. Can you make a android version of it.It has good strength.
https://github.com/abulmo/amoeba

Alexei Chernakoff said...

There [d] source, I do not know how to compile them.

Alexei Chernakoff said...

Cfish-131217(fix Cerebellum book support):
https://app.box.com/s/5wzqqoidxsoqgd4ipq06hvalv65jsixn

Kevin said...

@Alexei Chernakoff youre a genius!

BrainCfish is strong like BrainAsmFish

Unknown said...

Alexei, there is new brainfish 12.12.17. if u dont mind,can u compile it, thq Alexei

Alexei Chernakoff said...

Cfish-131217 with the support of Cerebellum books is stronger and better than Brainfish.

Rahul said...

https://drive.google.com/file/d/0B3p3UQUO0d2UYU42U09DV3NMeFlXTlI3dlF4aDBKcE90V3ln/view?usp=drivesdk follow this link and tell me that senpai2.0 run flalessly or not.

Unknown said...

Thanks Rahul.Seems Aprijal is back.

Unknown said...

...but i still prefer Alexei's arm8 compiles as im not sure Aprijal is using Static libraries.SL privided the best compatibility going forward with updayed OS and devices ;-)

Unknown said...

Just looking now senpai 2.0 arm8 under droidfish has a significantly lower node count then senpai 1.0 (ape) on one thread.doesnt neccessarily mean anything but its interesting as you first see the engine in action....

Unknown said...

i love brainfish alexei, n its playing style. if u dont mind, can u compile it 4 me, thq b4 alexei

Rahul said...
This comment has been removed by the author.
Alexei Chernakoff said...

My compilation Brainfish-121217:
https://app.box.com/s/zrjvi67hji07jwnqn9nhyd4i35brry59

Unknown said...

thq so much alexei,am so glad :)

Paulzzz said...

Is Aprijal the compiler of Senpai 2.0 for android?It's not yet posted in his blog.Btw,thanks to Lazar and Rahul for sharing the links to the new versions of Armfish,Texel and Senpai.Namaste!

Paulzzz said...

Also thanks to Alexei for keeping us updated with his compilations of cfish and brainfish.

Moch Wawan said...

Cfish with cerebellum book is stronger a lot. I moved on now :D

Unknown said...

Enjoy Paulzzz.Yeah AP compiled Senpai 2.0.he doesnt update his site anymore.i havent yet tested Senpai 2.0 in engine tourmaments under CFA
Yet..but works fine with Droidfish

Rahul said...

Hi,Sir Alexie please compile sctr 1.1 source code is able here.I know that you know.
https://github.com/nitrocan/sctr/releases

Rahul said...

New chess engines update like ticano7.0,scopian 2.7.9 and others.

Rahul said...

One more thing is that "new scarpio2.7.9" is based on "Manto carlo tree searh" which used by Alphazero go. It will be interesting now.

Alexei Chernakoff said...

I tried it, it does not work well on android. I assembled this version, but this algorithm still works poorly on mobile platforms.

Unknown said...
This comment has been removed by the author.
Rahul said...

Hi,Alexie can you compile Arasan latest version there is source code & thanks
https://www.arasanchess.org/downld.shtml

flither said...

So the new Chess for Android is out! But I've run 5 Komodo-Cfish games and one game K lost from nowhere when 700cp ahead, and the other one CFish won by mate but CfA didn't adjudicate it before. I don't know if it's bugs or I've done something wrong (I used tablebases for both engines and CFish also used Cerebellum)

Alexei Chernakoff said...

At me all normally works.

Unknown said...
This comment has been removed by the author.
Alexei Chernakoff said...

Medulla ?? What is it and where is Cfish?

Unknown said...
This comment has been removed by the author.
Alexei Chernakoff said...

In droidfish it is visible it is possible. But to use it in CFA, it is necessary to place the name of the book in source codes Cfish before compilation.

Unknown said...

i got it alexei, thank u s much. n btw, is there any cfish update??

flither said...

It's me again. Seems like it's Komodo 11 having problems with my Syzygy in CfA, I disabled it and K is not crashing in the middle anymore. But it seems adjudication in CfA doesn't work if engine is reporting "mate in x", with normal values it's ok (I set it to 650cp)

Alexei Chernakoff said...

I have a CFA adjudication that works well without errors. Just in settings [resigned adjudication] it is necessary to adjust the setting of [move count], to make it = 4. And also in the matches Cfish (stockfish) and Komodo do not 650 cp, but 400 cp, this is enough. If you use this setting, the solution will be rendered if the worst engine, within 4 moves, evaluates => 400 cp.
You must properly configure the shell.
I did it and she makes decisions exactly as I pointed out. Everything works just fine.

Rahul said...

Hi,Can anyone tell me that firenzina2.4.3 run on android 7.0(naugat) or not???bcoz I tried to install in c4a but it not install.

flither said...

Ok, thanks Alexei! I will try your settings.

Unknown said...
This comment has been removed by the author.