Most of the code is already in place. Things to consider: a) move filter - i.e. which positions to roll out b) using GNURoller for plays and cubes in rollouts c) efficiency of the current mt code for fast games in roll outs - we lock all threads during accounting d) variance reduction - my limited experience tells me that variance/time is more or less independent of turning it on/off for 0 ply rollouts
Christian. On Wed, Aug 19, 2009 at 2:53 PM, Ian Shaw<[email protected]> wrote: > > Over at bgonline there are some results of tests of eXtremeGammon > against Gnubg. > http://www.bgonline.org/forums/webbbs_config.pl?noframes;read=46526#4652 > 6 > > The XGRoller setting seems to clearly outperform Gnubg 3-ply over the > 100 positions tested so far. > > I'd be interested in seeing how gnubg gets on with similar settings. > With variance reduction off (as per XGRoller), I'm sure that gnubg is > fast enough for the job. > > The command "set player 0 chequerplay type rollout" and all related > commands for setting the rollout parameters exist, but are not > implemented. Hopefully the code can be taken from the rollout code > without too much trouble. > > I think that XGRoller has demonstrated that modern PCs are fast enough > to perform such rollouts, which was not the case when gnubg was first > coded. > > -- Ian > > > _______________________________________________ > Bug-gnubg mailing list > [email protected] > http://lists.gnu.org/mailman/listinfo/bug-gnubg > _______________________________________________ Bug-gnubg mailing list [email protected] http://lists.gnu.org/mailman/listinfo/bug-gnubg
