BGonline.org Forums

New Gnubg build with XG2 MET

Posted By: Henrik Bukkjaer
Date: Thursday, 6 October 2011, at 10:17 p.m.

In Response To: New Gnubg build with XG2 MET (Daniel Murphy)

It's my opinion, that most players don't really understand what constitutes the "best MET" in a given situation (match).

I haven't done a MET myself, but my understanding is that you calculate it iteratively starting at -1, -1, and that all your steps basically is based on the gammon and backgammon percentages and maybe some cube-efficiency stuff...

(1) If the XG MET is better, wouldn't you rather have Gnubg use the better MET when it is playing against you or doing a rollout?

If my understanding of MET calculation is correct, then XG MET is better for XG, but GNU MET might be better for GNUBG? I mean, if they differ, it must be because XG can achive some other (back)gammon-percentages than gnubg?

Turn it around: Say you were playing a bad player, who's errors were plays that were both too aggressive and too safe. Would you rather have that beginner play all his double decisions based on "optimal" XG MET or the appropriate Jacobs MET (say -150). I think he would win more against you playing the -150 MET.

(2) I haven't compared the tables, but I gather from discussion that the differences between the XG-Kazaross MET and the Rockwell-Kazaross MET are very small.

I hope so :-)

(3) In which case, the main advantage that I see is that using the same MET in both bots is eliminating the difference in, say XG and Gnubg rollouts, due solely to a different MET being used.

Why would you eliminate any differences to begin with? And would you like to eliminate any differences in it was at a cost of lower playing strength (ie. worse analysis)? Wouldn't it be better to then simply eliminate the bot (gnubg) and only look at results from XG? Aren't those differences the most interesting you could come across anyway?

----

I'm only asking these questions to get a better MET (usage) understanding. And I think the "right" answers depends on what you want to use the bot for. Here I just assumed you wanted the best playing strength from the bot. But that might not be the case. Maybe you want the bot (gnu) to analyze your play, and tell you how close to XG you play, in which case it gets blurry to me!

Messages In This Thread

• New Gnubg build with XG2 MET
  Daniel Murphy -- Monday, 3 October 2011, at 4:32 p.m.
  • New Gnubg build with XG2 MET
    kruidenbuiltje -- Monday, 3 October 2011, at 7:24 p.m.
    • New Gnubg build with XG2 MET
      Daniel Murphy -- Monday, 3 October 2011, at 8:27 p.m.
      • New Gnubg build with XG2 MET
        kruidenbuiltje -- Tuesday, 4 October 2011, at 5:49 p.m.
        • New Gnubg build with XG2 MET
          Michael Petch -- Tuesday, 4 October 2011, at 6:12 p.m.
        • XG2 MET on older GNUBG
          Michael Petch -- Tuesday, 4 October 2011, at 6:23 p.m.
          • XG2 MET on older GNUBG
            kruidenbuiltje -- Tuesday, 4 October 2011, at 7:44 p.m.
            • XG2 MET on older GNUBG
              kruidenbuiltje -- Tuesday, 4 October 2011, at 7:59 p.m.
  • New Gnubg build with XG2 MET
    Henrik Bukkjaer -- Thursday, 6 October 2011, at 1:27 p.m.
    • New Gnubg build with XG2 MET
      Daniel Murphy -- Thursday, 6 October 2011, at 7:26 p.m.
      • New Gnubg build with XG2 MET
        Henrik Bukkjaer -- Thursday, 6 October 2011, at 10:17 p.m.