Engine release Exacto 0.e

Exacto · Post by **Exacto** » Mon Mar 10, 2014 9:00 am

Hello,

Thanks to a few people on here for testing out Exacto 0.e. While it shares the same name as my last program, Exacto 0.d, this new version is really a brand new version. Is there anything I need to do to get the new version onto the CCRL?

Anyway, here's the website. It contains a list of the features:

http://math.berkeley.edu/~sparks/exacto/

The source code is included and covered under a BSD type license (but use it at your own risk - I am neither a chess player nor a programmer). The engine is meant to use with WinBoard/XBoard/Arena.

Best,
Dan

Graham Banks · Post by **Graham Banks** » Mon Mar 10, 2014 9:34 am

Exacto wrote:Hello,

Thanks to a few people on here for testing out Exacto 0.e. While it shares the same name as my last program, Exacto 0.d, this new version is really a brand new version. Is there anything I need to do to get the new version onto the CCRL?

Anyway, here's the website. It contains a list of the features:

http://math.berkeley.edu/~sparks/exacto/

The source code is included and covered under a BSD type license (but use it at your own risk - I am neither a chess player nor a programmer). The engine is meant to use with WinBoard/XBoard/Arena.

Best,
Dan

Could you give us a rough idea of some engines that you feel are around the same strength?

Graham.

Adam Hair · Post by **Adam Hair** » Mon Mar 10, 2014 11:08 am

Exacto wrote:Is there anything I need to do to get the new version onto the CCRL?

You do not have to do anything. I will test it for the 40/4 list.

Sylwy · Post by **Sylwy** » Mon Mar 10, 2014 11:40 am

Exacto wrote:Hello,

Thanks to a few people on here for testing out Exacto 0.e. While it shares the same name as my last program, Exacto 0.d, this new version is really a brand new version. Is there anything I need to do to get the new version onto the CCRL?

Anyway, here's the website. It contains a list of the features:

http://math.berkeley.edu/~sparks/exacto/

The source code is included and covered under a BSD type license (but use it at your own risk - I am neither a chess player nor a programmer). The engine is meant to use with WinBoard/XBoard/Arena.

Best,
Dan

Doesn't work under Uncle XP !!!

SilvianR

SzG · Post by **SzG** » Mon Mar 10, 2014 12:47 pm

Exacto wrote:this new version is really a brand new version. Is there anything I need to do to get the new version onto the CCRL?

Hello Dan,

Thanks for your engine.

Maybe a different version number would show that this Exacto is not that Exacto.

Now that Adam is going to test it, which I myself intended to do, I can wait and see in which range it belongs.

Gábor

Gerd Isenberg · Post by **Gerd Isenberg** » Mon Mar 10, 2014 3:46 pm

Exacto wrote:Hello,

Thanks to a few people on here for testing out Exacto 0.e. While it shares the same name as my last program, Exacto 0.d, this new version is really a brand new version. Is there anything I need to do to get the new version onto the CCRL?

Anyway, here's the website. It contains a list of the features:

http://math.berkeley.edu/~sparks/exacto/

The source code is included and covered under a BSD type license (but use it at your own risk - I am neither a chess player nor a programmer). The engine is meant to use with WinBoard/XBoard/Arena.

Best,
Dan

Hi Dan,

you may easily quarter your plain magic tables sizes, i.e. 2MB instead of 8MB for rooks, if you allow constructive index collisions, where redundant outer one-bits of the occupancy may share the same attack set and entry.

Gerd

Gerd Isenberg · Post by **Gerd Isenberg** » Mon Mar 10, 2014 4:18 pm

Gerd Isenberg wrote:Hi Dan,

you may easily quarter your plain magic tables sizes, i.e. 2MB instead of 8MB for rooks, if you allow constructive index collisions, where redundant outer one-bits of the occupancy may share the same attack set and entry.

Gerd

oups, you already do that...

Code: Select all

    if&#40;rook_moves&#91;square&#93;&#91;quickhash&#40;x, magic, bits&#41;&#93; == all_set&#41;&#123;
        // A new hash entry
        rook_moves&#91;square&#93;&#91;quickhash&#40;x, magic, bits&#41;&#93; = y;
    &#125; else if&#40;rook_moves&#91;square&#93;&#91;quickhash&#40;x, magic, bits&#41;&#93; != y&#41; &#123;
        // A bad collision

Then why not 12/9 rook/bishop bits only?

Exacto · Post by **Exacto** » Mon Mar 10, 2014 11:30 pm

Thanks guys!

- The engines I test against are usually Clarabit, Firefly, Joker, Adam, Neurosis, and a few others between 2200-2330 on CCRL. I estimate the rating for Exacto 0.e should be higher than 2100, but lower than 2200. (Compared to the CCRL ratings.)

- This version is supposed to be called 0.e. I guess I didn't put that in the title of the executable, but it does say which version it is when you boot it up.
[EDIT: Oh! You probably mean like... 1.0 instead of something so consecutive... yeah, that might have been a good call.]

- It doesn't work on XP at the moment, which is something I plan to address in the next version. Unfortunately, the program searches different game trees when compiled in GCC vs Microsoft, but if I can sort that out, I'll compile in Mingw-w32 for people on XP.

- As far as my magics - good eye catching that. Yes, I allow constructive collisions. But, for whatever reason, my code wasn't able to generate magics for smaller tables. It would just sit their looping seemingly indefinitely (20-30 minutes, until I closed it and decided to just use 14/11 bit magics).

Gerd Isenberg · Post by **Gerd Isenberg** » Tue Mar 11, 2014 5:35 am

Exacto wrote: - As far as my magics - good eye catching that. Yes, I allow constructive collisions. But, for whatever reason, my code wasn't able to generate magics for smaller tables. It would just sit their looping seemingly indefinitely (20-30 minutes, until I closed it and decided to just use 14/11 bit magics).

I guess your randoms are not sparse enough. See this proposal by Tord Romstad:

Code: Select all

uint64 random_uint64&#40;) &#123;
  uint64 u1, u2, u3, u4;
  u1 = &#40;uint64&#41;&#40;random&#40;)) & 0xFFFF; u2 = &#40;uint64&#41;&#40;random&#40;)) & 0xFFFF;
  u3 = &#40;uint64&#41;&#40;random&#40;)) & 0xFFFF; u4 = &#40;uint64&#41;&#40;random&#40;)) & 0xFFFF;
  return u1 | &#40;u2 << 16&#41; | &#40;u3 << 32&#41; | &#40;u4 << 48&#41;;
&#125;

uint64 random_uint64_fewbits&#40;) &#123;
  return random_uint64&#40;) & random_uint64&#40;) & random_uint64&#40;);
&#125;

Exacto · Post by **Exacto** » Wed Mar 12, 2014 2:54 am

Ruxy - following up on the XP issue. I've made a separate set of compiles for XP (32 bit) and Macintosh (64 bit) platforms through GCC. An important note: in order to switch into GCC I had to track down and squash a bug, which means that this version may search slightly different trees than 0.e. On my website (link above) it is listed as 0.e.x.

But, 0.e.x is not a "release," rather, it's just something for people on XP32 / Mac to be able to play against my engine. So, if testing, please test 0.e on Win64.

[Future versions should be distributed with identical XP32(GCC)/Win64(MS)/Mac64(GCC) compiles.]

TalkChess.com

Engine release Exacto 0.e

Engine release Exacto 0.e

Re: Engine release Exacto 0.e

Re: Engine release Exacto 0.e

Re: Engine release Exacto 0.e

Re: Engine release Exacto 0.e

Re: Engine release Exacto 0.e

Re: Engine release Exacto 0.e

Re: Engine release Exacto 0.e

Re: Engine release Exacto 0.e

Re: Engine release Exacto 0.e