Experimental Glaurung version

Discussion of chess software programming and technical issues.

Moderators: hgm, Harvey Williamson, bob

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Post Reply
Tord Romstad
Posts: 1808
Joined: Wed Mar 08, 2006 8:19 pm
Location: Oslo, Norway

Experimental Glaurung version

Post by Tord Romstad » Thu Dec 04, 2008 8:17 pm

Hi all,

After the discussion in the thread "Hardware vs Software", where (among other things) the values of recursive null move pruning and late move reductions were discussed, I have made an experimental Glaurung version with UCI options for enabling/disabling null move pruning, late move reductions and the tapered "super qsearch" I use in the transitory phase between the main search and the qsearch. The three new options are called "Late Move Reductions", "Null Move Pruning" and "Super Qsearch", and they all have the default value "true".

The source code and a polyglot.ini file with the new options can be downloaded here.

This Glaurung version is published only for the benefit of those (if any) who wants to experiment with the three new options (setting any of them to "false" will almost certainly reduce the playing strength, but I have no idea by how much). Of course, you are also free to do whatever else you like with this version, but don't expect it to be significantly stronger than Glaurung 2.1. Almost nothing has changed (I don't do chess programming at the moment, and haven't touched my source code since May or June), and the strength should be more or less the same as for 2.1.

Tord

bob
Posts: 20358
Joined: Mon Feb 27, 2006 6:30 pm
Location: Birmingham, AL

Re: Experimental Glaurung version

Post by bob » Thu Dec 04, 2008 8:40 pm

Tord Romstad wrote:Hi all,

After the discussion in the thread "Hardware vs Software", where (among other things) the values of recursive null move pruning and late move reductions were discussed, I have made an experimental Glaurung version with UCI options for enabling/disabling null move pruning, late move reductions and the tapered "super qsearch" I use in the transitory phase between the main search and the qsearch. The three new options are called "Late Move Reductions", "Null Move Pruning" and "Super Qsearch", and they all have the default value "true".

The source code and a polyglot.ini file with the new options can be downloaded here.

This Glaurung version is published only for the benefit of those (if any) who wants to experiment with the three new options (setting any of them to "false" will almost certainly reduce the playing strength, but I have no idea by how much). Of course, you are also free to do whatever else you like with this version, but don't expect it to be significantly stronger than Glaurung 2.1. Almost nothing has changed (I don't do chess programming at the moment, and haven't touched my source code since May or June), and the strength should be more or less the same as for 2.1.

Tord
If you want, when I have time, I can try that version in my testing and play an 8K match vs other programs with each of the three parts disabled, one at a time and all at once...

Tord Romstad
Posts: 1808
Joined: Wed Mar 08, 2006 8:19 pm
Location: Oslo, Norway

Re: Experimental Glaurung version

Post by Tord Romstad » Thu Dec 04, 2008 10:06 pm

bob wrote:
Tord Romstad wrote:Hi all,

After the discussion in the thread "Hardware vs Software", where (among other things) the values of recursive null move pruning and late move reductions were discussed, I have made an experimental Glaurung version with UCI options for enabling/disabling null move pruning, late move reductions and the tapered "super qsearch" I use in the transitory phase between the main search and the qsearch. The three new options are called "Late Move Reductions", "Null Move Pruning" and "Super Qsearch", and they all have the default value "true".

The source code and a polyglot.ini file with the new options can be downloaded here.

This Glaurung version is published only for the benefit of those (if any) who wants to experiment with the three new options (setting any of them to "false" will almost certainly reduce the playing strength, but I have no idea by how much). Of course, you are also free to do whatever else you like with this version, but don't expect it to be significantly stronger than Glaurung 2.1. Almost nothing has changed (I don't do chess programming at the moment, and haven't touched my source code since May or June), and the strength should be more or less the same as for 2.1.

Tord
If you want, when I have time, I can try that version in my testing and play an 8K match vs other programs with each of the three parts disabled, one at a time and all at once...
Thanks, that would be very interesting, of course! If it is difficult to find the time to test all 2^3=8 combinations, focus on the null move and LMR settings, which are the most important ones.

Tord

Edsel Apostol
Posts: 759
Joined: Mon Jul 17, 2006 3:53 am
Full name: Edsel Apostol
Contact:

Re: Experimental Glaurung version

Post by Edsel Apostol » Fri Dec 05, 2008 2:09 am

Tord Romstad wrote:
bob wrote:
Tord Romstad wrote:Hi all,

After the discussion in the thread "Hardware vs Software", where (among other things) the values of recursive null move pruning and late move reductions were discussed, I have made an experimental Glaurung version with UCI options for enabling/disabling null move pruning, late move reductions and the tapered "super qsearch" I use in the transitory phase between the main search and the qsearch. The three new options are called "Late Move Reductions", "Null Move Pruning" and "Super Qsearch", and they all have the default value "true".

The source code and a polyglot.ini file with the new options can be downloaded here.

This Glaurung version is published only for the benefit of those (if any) who wants to experiment with the three new options (setting any of them to "false" will almost certainly reduce the playing strength, but I have no idea by how much). Of course, you are also free to do whatever else you like with this version, but don't expect it to be significantly stronger than Glaurung 2.1. Almost nothing has changed (I don't do chess programming at the moment, and haven't touched my source code since May or June), and the strength should be more or less the same as for 2.1.

Tord
If you want, when I have time, I can try that version in my testing and play an 8K match vs other programs with each of the three parts disabled, one at a time and all at once...
Thanks, that would be very interesting, of course! If it is difficult to find the time to test all 2^3=8 combinations, focus on the null move and LMR settings, which are the most important ones.

Tord
If Bob have more time to test and would like to validate results from other engines besides Gaurung and Crafty, I suggest my engine Twisted Logic. It already has these UCI options to enable/disable Transposition Table, Null Move Pruning, Zero Pruning (revised Razoring/Futility Pruning), Speculation (revised LMR).

At Tord, I think Futility Pruning/Razoring is important also. I just don't know if its more important than LMR o not but I think Null Move is the most important of them all.

User avatar
hgm
Posts: 22909
Joined: Fri Mar 10, 2006 9:06 am
Location: Amsterdam
Full name: H G Muller
Contact:

Re: Experimental Glaurung version

Post by hgm » Fri Dec 05, 2008 9:55 am

Actually, I think it would not be very elucidating to test all kind of engines that are all equally advanced and designed according to almost identical principles, all of nearly equal strength. It would be much more useful to test an engine at the completely opposite end of the spectrum, which has almost nothing, like micro-Max. Bob might not have a suitable gauntlet for it, though...

Uri Blass
Posts: 8309
Joined: Wed Mar 08, 2006 11:37 pm
Location: Tel-Aviv Israel

Re: Experimental Glaurung version

Post by Uri Blass » Fri Dec 05, 2008 12:52 pm

hgm wrote:Actually, I think it would not be very elucidating to test all kind of engines that are all equally advanced and designed according to almost identical principles, all of nearly equal strength. It would be much more useful to test an engine at the completely opposite end of the spectrum, which has almost nothing, like micro-Max. Bob might not have a suitable gauntlet for it, though...
I doubt if Crafty and Glaurung are designed according to almost identical principles.

Glaurung is still clearly better program than Crafty inspite of the fact that Tord did not have a cluster for testing so I guess that Glaurung is the program with the superior search relative to Crafty and I guess that tord is better than Bob in using search tricks so they give Glaurung more rating points.

Uri

User avatar
Jim Ablett
Posts: 1327
Joined: Fri Jul 14, 2006 5:56 am
Location: London, England
Contact:

Re: Experimental Glaurung version

Post by Jim Ablett » Fri Dec 05, 2008 9:43 pm

Image
Glaurung 081204 (experimental) by Tord Romstad

Windows Intel p.g.o Win32/X64 builds here >

http://www.mediafire.com/?mdjngqzz2dq

Jim.

Post Reply