| |
BGonline.org Forums
The Michael Depreli Bot Comparisons
Posted By: Daniel Murphy In Response To: The Michael Depreli Bot Comparisons (Tom Keith)
Date: Monday, 14 February 2011, at 7:46 p.m.
Very informative. Good work, well presented!
Minor quibble: "The bots played at their best settings ... 2-ply for Gnu Backgammon." 4-ply would have been better, but awfully slow in 2005. 3-ply with 2-ply cube would have been better, too, according to Michael's results, but I don't think we knew that in 2005.
The "Bot Comparison — 2005 Study" is ordered by "Overall Errors." I think I'd change that to "Total Errors," and similarly order the "Bot Comparison — 2010 Study" by "Total Errors" and Maik Staibler's error rate table by "Error rate."
Interesting that Gnubg 4-ply (ER 0.755) ranks nearly as high as XGR (ER 0.705) (but is much, much slower) and, although both Gnubg 2-ply and Gnubg 3-ply performed much worse than that, if a row were added for Gnubg using 2-ply for cube decisions and 3-ply for checker play, that combination would rank nearly as high as eXtreme 3-ply. No idea how that combo would fare in the speed vs. strength table). That combo wasn't tested, but the data can be fairly extracted from the tables, no?
Might add that the XGR data is for v1 and v2 will be out soon; the current Gnubg engine is v.15 (but I'm not sure what if any improvements in playing strength v.15 has over v.14); BG-Blitz is still under development, and Snowie and good old JellyFish are not.
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.