[ View Thread ] [ Post Response ] [ Return to Index ] [ Read Prev Msg ] [ Read Next Msg ]

BGonline.org Forums

Bot Comparison after 60/500 games.

Posted By: neilkaz
Date: Thursday, 29 October 2009, at 12:47 a.m.

In Response To: Bot Comparison after 60/500 games. (Michael Depreli)

Some trends are emerging. While this is only 60 games, it is several hundred plays, although for some, the differences are small. Obviously when a hundred or two games are done, this will likely be a more accurate representation of the bots' strengths and weakness.

XG does seem to be really strong, and that is quite impressive when one notes how fast it is.

XG 4 ply and XGR are very strong. XG 4 ply cube play is most impressive.

GNU 4 ply is very strong as one would expect for such a long lookahead. Its checker play is excellent, but it's cube play is a bit off the mark, perhaps from overvaluing positions with single anchors?

Snowie 4 is showing good cube play so far, but less accurate checker play.

GNU 3 ply checker play is more ahead of 2 ply than I'd have expected while 3 ply cube play is worse than I'd have expected.

BGB cube play needs some more work, especially the frequent errors from cubing too soon.

I'll let a couple of the cats out of the bag now....

Work is continuing on XG for quite a few enhancements targetted at high end users...ie the serious student. Anyone with a multi-cored/threaded computer will enjoy the fact that XG will soon be able to use more than 2 cores..ie all of them. Feel free to look at how quickly I did a 57K rollout of the opening 21 using all 8 cores. .. I started it when I left work, and it was finished by the next morning when I got into work.

Work on parallel rollouts is being done, and many users will like following along and seeing the rollout's progress for all plays.

For XGR fans, and especially useful to those with a fast computer, work is being done to have an additional setting for XGR which is stronger than the former, while being no more than 4 or 5 times slower. I've done lots of testing of settings and what we have now is definately superior to XGR for cube decisions and doesn't seem to have nearly as many bad takes as the few XGR has.

Messages In This Thread

 

Post Response
Your Name:
Your E-Mail Address:
Subject:
Message:

If necessary, enter your password below:

Password:
 

 

[ View Thread ] [ Post Response ] [ Return to Index ] [ Read Prev Msg ] [ Read Next Msg ]

BGonline.org Forums is maintained by Stick with WebBBS 5.12.