BGonline.org Forums

About SI

Posted By: eXtreme Gammon
Date: Wednesday, 18 July 2012, at 1:26 a.m.

In Response To: Victor Ashkenazi Vs. Jeb Horton (Henrik Bukkjaer)

I do have the number for Gigantic SI for XGR++

I've however the number for Normal SI and Infinite SI (from the Depreli Study where all candidate rolled are evaluated)

XGR++ infinite SI PR=0.115

XGR++ Normal SI PR=0.134

for reference the difference in 3-ply between Normal and infinite is 0.034 PR

Though, your post point "The (PR) difference I find, typically is not from moves that turns out to be errors on XGR++, but rather a different number of decisions counted towards the PR!

I never though of that so I cannot give you any number of that. It's clear that one extra (or one less) decision can affect PR significantly when the PR is already low. I do not think the effect is as much on the doubles (again just a gut felling i cannot back that by actual numbers) but more on a checker play in Race close to the drop point.

---

About you second question Is it possible to control what ply is used in the initial pruning of the moves?

I think there may be some terminology mix-up. What is refer as pruning is usualy the use of a weaker/faster Neural Net within a N-ply evaluation (N>1).

I think you refer as the SI filtering. XG filtering is described at the end of http://www.extremegammon.com/Searchinterval.aspx

XGR++ make first a 4-ply evaluation using the same SI setting than 4-ply, then after that use the same filter than the 4-ply to decide what to analyze in XGR++

---

A few personal notes about SI:

I do recommend to always use "Normal" SI in Rollout, the gain is so low that your time will always be better used to make a longer RO. If you know that 3-ply Normal SI make an immediate mistake that Large SI will catch you are better off to use a setting where the first move use 4-ply.

For direct analyze the only time I will recommend using anything than "normal" SI is for XGR++ as the difference is more important in % (XGR++ infinite PR is 19% greater than XGR++ normal - for 3 ply it's "only" 7%)

Messages In This Thread

• Victor Ashkenazi Vs. Jeb Horton
  Dmitriy Obukhov -- Sunday, 15 July 2012, at 9:38 p.m.
  • Victor Ashkenazi Vs. Jeb Horton
    benf -- Sunday, 15 July 2012, at 11:32 p.m.
    • Victor Ashkenazi Vs. Jeb Horton
      Dmitriy Obukhov -- Sunday, 15 July 2012, at 11:44 p.m.
      • Victor Ashkenazi Vs. Jeb Horton
        benf -- Sunday, 15 July 2012, at 11:50 p.m.
        • Victor Ashkenazi Vs. Jeb Horton
          Dmitriy Obukhov -- Sunday, 15 July 2012, at 11:55 p.m.
          • Victor Ashkenazi Vs. Jeb Horton
            Henrik Bukkjaer -- Monday, 16 July 2012, at 12:14 a.m.
            • Victor Ashkenazi Vs. Jeb Horton
              Dmitriy Obukhov -- Monday, 16 July 2012, at 12:32 a.m.
              • Victor Ashkenazi Vs. Jeb Horton
                Henrik Bukkjaer -- Monday, 16 July 2012, at 10:55 p.m.
                • +1 *NM*
                  Neil Robins -- Tuesday, 17 July 2012, at 11:27 a.m.
                • Victor Ashkenazi Vs. Jeb Horton
                  neilkaz -- Tuesday, 17 July 2012, at 3:37 p.m.
                  • Victor Ashkenazi Vs. Jeb Horton
                    Henrik Bukkjaer -- Tuesday, 17 July 2012, at 6:46 p.m.
                    • Victor Ashkenazi Vs. Jeb Horton
                      neilkaz -- Tuesday, 17 July 2012, at 7:44 p.m.
                    • About SI
                      eXtreme Gammon -- Wednesday, 18 July 2012, at 1:26 a.m.