Gaviota and 6-men

Endgame analysis using tablebases, EGTB generation, exchange, sharing, discussions, etc..

Gaviota and 6-men

Postby jshriver » Tue May 18, 2010 7:26 pm

When Gaviota was released there was a lot of discussion on how to do the egtb generator (that ultimately was for the benefit of 6men).

Wondering if there has been work on that since or if it was still a goal? Then perhaps even 7-men.
Hope everyone is well :)
-Josh
User avatar
jshriver
 
Posts: 287
Joined: Tue Jan 31, 2006 5:59 am
Location: Toledo, OH, USA

Re: Gaviota and 6-men

Postby mballicora » Thu May 20, 2010 4:18 am

jshriver wrote:When Gaviota was released there was a lot of discussion on how to do the egtb generator (that ultimately was for the benefit of 6men).

Wondering if there has been work on that since or if it was still a goal? Then perhaps even 7-men.
Hope everyone is well :)
-Josh


Hi Josh,

Yes, 6-men are still part of the plan. So far, the goal was to make the probing code as simple/efficient/fast as possible, so the people who decide to use it see a benefit. It is useless to offer something that no one will download. Users/testers will download it if several engines support it, and engines will support it if their authors see an advantage over the previous systems. I think that there is an advantage in every single aspect with no trade offs. Better compression, more options (faster decompression systems for the future SSDs), better cache (more efficient and faster), bitbases on the fly and flexible code that will allow expansion (extremely easy to plug in a new compression scheme), and even a future addition of true bitbases without modifying the probing code in the engine! I think I reached that goal, and now is up to the engine developers to use the code. Hard to say if they will use it, but I did my part.

Regarding the 6-men, the next phase is to redesign the generator for the 5 men. The current one is primitive and slow. It served its purpose (it was straightforward, which was critical to make it bug-free), but not for the 6 men. However, before going to Mars, we need to go to the moon again, but faster and more efficiently. If I can make the generator for 5 men to use as little RAM as possible and faster, the generator for 6 men will me amenable for current hardware that people have in their homes. Ultimately, I won't generate those, it should be a team effort. There is more chances that people will help if they see concrete proof this is going in the right direction. This will be more evident if your "favorite engine" uses 5 men TBs, and now you would like it to use 6 men. I hope my first paragraph explains why having a good probing code is essential before going into 6 men.

I have some algorithmic ideas, but the way to test them is with the 5 men first. I may fail, but I believe it is worth trying. So, this may require some patience, but so far progress has been steady considering this is just a hobby. The only think I can promise is that I will keep having fun doing it :-)

Miguel
mballicora
 
Posts: 15
Joined: Mon Dec 07, 2009 4:05 am
Location: Chicago, IL, USA

Re: Gaviota and 6-men

Postby dann corbit » Thu May 20, 2010 9:07 pm

I have a notion:
When you rewrite the 5 man and generalize it for larger counts, how about making it totally general so that 7,8,9 etc. man files can also be generated.
It may sound silly now, but 5 years from now we'll have petabyte hard drives for $100.
dann corbit
 
Posts: 16
Joined: Fri Jan 25, 2008 9:10 am

Re: Gaviota and 6-men

Postby jshriver » Mon May 24, 2010 4:19 pm

Thanks for the update and appreciate all your hard work so far. Not in a rush, just a little anxious to start generating some but understand the wait. Would rather it be done right the first time since in the end it will save time by being more efficient.

Once I get the 40/40 testing done willing to donate more cpu cycles to other projects.

-Josh
User avatar
jshriver
 
Posts: 287
Joined: Tue Jan 31, 2006 5:59 am
Location: Toledo, OH, USA

Re: Gaviota and 6-men

Postby jshriver » Mon May 24, 2010 4:20 pm

dann corbit wrote:I have a notion:
When you rewrite the 5 man and generalize it for larger counts, how about making it totally general so that 7,8,9 etc. man files can also be generated.
It may sound silly now, but 5 years from now we'll have petabyte hard drives for $100.


Sounds like a good idea to me :)
User avatar
jshriver
 
Posts: 287
Joined: Tue Jan 31, 2006 5:59 am
Location: Toledo, OH, USA

Re: Gaviota and 6-men

Postby redpawn » Thu Jun 03, 2010 4:48 pm

Hi Miguel:
I Just want to say that I have high hopes for your efforts in the tablebases field, and knowing that you're planning for bitbases also with the same probing code is just great,

Optimizing the generation code is very wise and a nice step from you, if the resulting generator will work faster even with less ram and is optimized for the new generation of CPU's with multi cores & is equipped with the capability of saving work & continuing later is great news for me, but still I hope someone will come out & will provide a GUI for the generator.

But since we had already generated the 3+4+5 men Gaviota tablebases our natural next step will be hoping for the release of the long awaited six men generator or even better the universal generator for any number of men. With both DTM or WDL knowledge.

Best Regards,

Wassim
redpawn
 
Posts: 169
Joined: Thu Dec 06, 2007 8:32 am

Re: Gaviota and 6-men

Postby redpawn » Fri Aug 20, 2010 3:39 pm

Sooooooooo

What happened with Gaviota tablebases
redpawn
 
Posts: 169
Joined: Thu Dec 06, 2007 8:32 am

Re: Gaviota and 6-men

Postby mballicora » Tue Aug 24, 2010 7:02 am

redpawn wrote:Sooooooooo

What happened with Gaviota tablebases


I have been working on the engine, but after I release a new version, I will bet back to the TBs. I have been writing down some ideas to try, though.

I wish I can clone myself :-)

Miguel
mballicora
 
Posts: 15
Joined: Mon Dec 07, 2009 4:05 am
Location: Chicago, IL, USA

Re: Gaviota and 6-men

Postby redpawn » Tue Aug 24, 2010 1:01 pm

Hi Miguel:

Best wishes for the engine

I hate to ask this question but how much elo increase do you expect from the next release

I hope it will get to the elite tier soon

as for the generator would you please keep us updated more often

Best Regards

Wassim
redpawn
 
Posts: 169
Joined: Thu Dec 06, 2007 8:32 am

Re: Gaviota and 6-men

Postby mballicora » Mon Sep 06, 2010 10:35 pm

redpawn wrote:Hi Miguel:

Best wishes for the engine

I hate to ask this question but how much elo increase do you expect from the next release


About 100 Elo points according to the tests I am currently running...
Still a long way to go to catch the elite, but still there is a lot of room for improvement.

I hope it will get to the elite tier soon

as for the generator would you please keep us updated more often


Certainly.

Miguel

Best Regards

Wassim
mballicora
 
Posts: 15
Joined: Mon Dec 07, 2009 4:05 am
Location: Chicago, IL, USA

Re: Gaviota and 6-men

Postby redpawn » Tue Sep 07, 2010 2:13 pm

Thank you Miguel :D
redpawn
 
Posts: 169
Joined: Thu Dec 06, 2007 8:32 am

Re: Gaviota and 6-men

Postby jshriver » Sat Oct 15, 2011 4:31 am

Just bumping this up the list :) It's now been 17 months since my initial post so still curious what's in the pipeline. I've been saving money for a while for a top of the line chess machine, and if 6men code for gaviota was available I'd be willing to build and dedicate it to this. While the 6-men egtb are done, I really respect Mr Ballicora and his engine. In fact I finally bout Deep Fritz, and ran a game against the latest Gaviota and Deep Rybka 4 SSE4 in Chessbase and it drew :)

Realize as time goes by people might lose interest, so mostly doing a modest probe to see what's going on? If this might be a project of the past, we all understand and respect life moving on, but this is a unique project with a lot of possibilities :) So if it's still in the works will wait, but if generating efficiency is the only draw back then I'd recommend releasing since if it were possible then, we'd already have over a year to burn CPU's. If it's not a big priority, which I humbly respect, perhaps releasing the tbgen code as an open source project so others might try and get the 6+ barrier broken. Plus the 3-5men already appears to beat the Nalimov generator due to the very slick parallel code. Generating on a quad core is amazing, but we can get a 16core now for less than I paid for my first computer.

I'm willing to dedicate CPU time and bandwidth, but need the code. Miguel if you read this I mean no disrespect, I just want to see what you started blossom and suspect many others are curious as well.

Peace be with you,
-Josh
User avatar
jshriver
 
Posts: 287
Joined: Tue Jan 31, 2006 5:59 am
Location: Toledo, OH, USA


Return to Endgame Tablebases

Who is online

Users browsing this forum: No registered users and 2 guests