| |
BGonline.org Forums
Opening rollouts in DB?
Posted By: eXtreme Gammon In Response To: Opening rollouts in DB? (Stick)
Date: Thursday, 29 April 2010, at 12:28 a.m.
I hear you :)
I may add the combined Rollout in the Opening book database for Money. And maybe add the replies also.
If anyone (Nack?) has that in a format I could use (i will need actual equity not only the differences) it should not be difficult to convert that to my internal format. One question that will arise is how to combine RO from different bot/level? should a XG 4-ply render all XG 3-ply RO obsolete?
Now for the difference of level. Let's take the Depreli study and assume the opening moves are evenly distributed (so about 33.333 time each of 15 roll). Using Stick GnuBG or on this site, XG err for 21 (0.0156) and 64 (0.0037) for total cost over the 500 games of 0.643 equity.
That's equivalent to about 0.6 Elo point, so surely better but not by much.
- For 3-ply (total error of 35.689) it represents 1.8% of the errors.
- for XGR+ (total error of 16.354) it represents 3.9% of the errors.
However, if we take Neil's latest RO using XG 4-ply, 21$ is better by only 0.003 and 64 is now [R=S P3]. So the total error is now only 0.100 (compare to 0.643) and the Elo gain about 0.1.
To summarize my thoughts: yes we should use the most accurate value we have and RO are more accurate than 5-ply (that's what I use for the opening book). But the difference will be minor.
I don't know a way to apply all that ATS (while using 5-ply it was reasonable to run all 9000+ position/score in 5-ply: about a week work on my old core 2 duo)
PS:I actually plan to add, in the next version, an option to pick the opening move randomly within x equity (0.010 maybe) so players can experience more variety when playing
Xavier Dufaure de Citres
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.