Page 1 of 1

GreKo 5.6.1

PostPosted: 11 Apr 2008, 18:45
by Vladimir Medvedev
Hello, World!

New version is available at the usual place:
http://greko.chess.googlepages.com/

Many changes, mostly in search. Some old bugs fixed, some new added.
Have fun :)

Update (23-Apr-2008)
For those who like to test. Small updates in evaluation. Binary only.
http://greko.chess.googlepages.com/greko-5.6.1.zip

Re: GreKo 5.6

PostPosted: 11 Apr 2008, 21:31
by Graham Banks
Thanks Vladimir. 8-)

Re: GreKo 5.6

PostPosted: 12 Apr 2008, 10:24
by Volker Pittlik
Vladimir Medvedev wrote:Hello, World!

New version is available at the usual place:
...
Many changes, mostly in search....


Thanks from my side too. It compiles fine for Linux after making small changes to the Makefile (unix.* instead of win32.*). With a profile guided compilation with the Intel compiler I got a speed up from 19.8 to 14.9 seconds for "perft 6".

If I only could think of a possibility to determine if the new version is stronger than the old one without playing hundreds of games...


Volker

Re: GreKo 5.6

PostPosted: 12 Apr 2008, 10:28
by Vladimir Medvedev
My tests (after ~200 games against different open source opponents) give about +100 points compared to 5.5 in 1'+1" blitz.

Re: GreKo 5.6

PostPosted: 12 Apr 2008, 12:32
by Volker Pittlik
Vladimir Medvedev wrote:...about +100 points compared to 5.5...


Wow, that is a lot! My computer is just busy with other things but maybe I run a test after the other stuff is finished.


Volker

Re: GreKo 5.6

PostPosted: 12 Apr 2008, 17:37
by Tony Thomas
Vladimir Medvedev wrote:My tests (after ~200 games against different open source opponents) give about +100 points compared to 5.5 in 1'+1" blitz.


I knew you were wrong when you said that you expect the programs progress to end around 2400 points. Tony likes the 100 point improvement in 1m+1sec.

Re: GreKo 5.6

PostPosted: 17 Apr 2008, 11:18
by Volker Pittlik
Volker Pittlik wrote:
Vladimir Medvedev wrote:...about +100 points compared to 5.5...


... I run a test after the other stuff is finished.


Unfortunately our results are different (same time control):

Code: Select all
Rank Name                  Elo    +    - games score oppo. draws
   1 Scorpio 2.0           360  138  105    40   91%    10   13%
   2 Arasan 10.3 (1 cpu)   252  106   91    40   83%    10   20%
   3 Crafty-22.1           148   91   84    40   70%    10   25%
   4 Phalanx XXII Reborn   116   93   87    40   64%    10   13%
   5 GreKo 5.6              17   42   41   200   53%    -2   16%
   6 GreKo 5.5               4   43   42   200   51%    -2   14%
   7 Xpdnt 080209          -73   83   87    40   38%    10   25%
   8 Mediocre v0.332      -124   92  100    40   34%    10    3%
   9 OliThink 5.1.3       -124   89   97    40   33%    10   10%
  10 amundsen 0.65.1      -171   87   97    40   25%    10   20%
  11 Beowulf-2.4          -185   89  100    40   24%    10   18%
  12 prophet_2            -220   96  112    40   23%    10    5%


Whatever the reasons are. This are not enough games. To be relatively sure to say version 5.6 is better than 5.5 it might take thousands of games. A reliable test to be completed in a reasonable time (an hour or less) would be a great gift.


Volker

Re: GreKo 5.6

PostPosted: 17 Apr 2008, 12:23
by Vladimir Medvedev
I see. Of course, it can depend on hash settings, opening book, hardware and many other factors. 1000 games would be fine, but where to get the time? :)