| |
BGonline.org Forums
GNUbg proper cube action: what does the (n.m%) figure mean?
Posted By: Henrik Bukkjaer In Response To: GNUbg proper cube action: what does the (n.m%) figure mean? (Bob Koca)
Date: Wednesday, 4 May 2011, at 4:15 p.m.
Yes, exactly the same hypothesis that the "No double, Take" part of the same but result is based on!
BTW: in gnubg, if you use a skewed ME table, you're removing some a that oddity you mention, because both the pass and the take variants will include errors from one side in the rest of the match. However, I'm aware that all subsequent evaluations does not take into account the difference - just the ME calculations.
Eg. the score is -1/-2 post Crawford. You're the "favorite" in the match by say 200 rating points (which equals around 54% in DMP if I recall correctly). You start with a good roll, like 5-3. Now your opponent doubles. I think gnubg would evaluate the position to you having in the neighborhood of 53% (correct for evenly matched play), and then drop because the ME table say you have 54% going to DMP!!
If these bots are to play perfectly against players who do not they need to be constructed a lot different from what developers have been aiming at so far. The nets should be trained in play vs. opponents with different traits, and evaluations should be adjusted before being put into the ME calculations. There's a lot of stuff that could be done and improved upon, and the bots could benefit a lot from this and become MUCH stronger vs. the imperfect opponents!
But the percentage shown in the line below (4,2%) is a step in the right direction! I find it useful for analyzing real life play.
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.