|
BGonline.org Forums
XGR+ looks stronger than I'd hoped ! and...
Posted By: neilkaz In Response To: Bot Comparison 140 / 500 games. (Michael Depreli)
Date: Friday, 27 November 2009, at 8:23 p.m.
When I proposed the XGR+ concept to Xavier I'd hoped that we could gain about 5 elo over XGR at a cost of being about 5 times slower (aimed for high end computers for near real time, and for longer more accurate match analysis). I am pleased with the end result when I see it is 9 elo stronger and 6 times slower. I was certain that XGR+ wouldn't take some of the bad cubes that XGR occasionally takes, but I really didn't expect such an improvement in checker play.
Any user except those with quite slow computers can make good use of XGR+ for detailed match analysis. Simply select the new World Class analysis option and the first pass will be made with 4 ply and the 2nd pass with XGR+. This analysis may take an hour or two on an older computer for a typical 11 pt match, but wil finish in a decent time frame for those with dual cores or better yet multi cores.
Those with i7 chips may enjoy playing against XGR+ as a very strong opponent.
I remain impressed by the strength of XG's results as these tests continue onward. XG is the clear winner on every ply. XG 3 ply checker play is stronger than GNU 3 ply and XG 4 ply checker play is almost as strong as GNU 4 ply, while XG 4 ply cube is clearly stronger than any other bot. It may make sense for those with fast computers and running an overnight RO to deeply study a position to use 3 ply play/4 ply cube with XG.
I do see some oscillation effect with XG where the odd plies (3 and 5) seem prone to the occasional bad take, but overall my tests show that this effect is less than with GNU.
Note that the rollouts for this study are being done with GNU. Therefore, if there's any bias in the results, it should be towards GNU, however, I'd expect it only to be slight if at all.
.. neilkaz ..
|
BGonline.org Forums is maintained by Stick with WebBBS 5.12.