BGonline.org Forums

XG 4 ply RO of 64P-62 completed

Posted By: Daniel Murphy
Date: Thursday, 28 January 2010, at 8:36 p.m.

In Response To: XG 4 ply RO of 64P-62 completed (Gregg Cattanach)

it feels like I'm being told that the plays SHOULD be called, one better than the other, if we just did more rollouts.

"Too close to call" means that after X trials with Y parameters and seed Z, the confidence intervals are so large relative to the difference in equity between the plays that the rollout has not determined with sufficient confidence that one play is better than the other.

That's all you're being told by "TCTC."

But if the "too close to call" rollout is extended indefinitely, it's very unlikely that the equities of the two plays will end up being equal. So, most likely, the indefinitely extended rollout will indeed show that (with parameters Y and seed Z) one play is better than the other.

The confidence intervals in Neil's rollout are ~±0.003. These will get smaller if the rollout is extended. A mere 699,840 more trials should reduce them to ~±0.00075! At which point, however, the rollout may still be "too close to call."

I'm not saying you'd have to do 746,496 trials, or more, before concluding that the plays are "tied." It'd be reasonable to conclude after only 46,656 trials, with confidence intervals of ~±0.003 and a difference of ~0.001, that the plays are very likely to be very close. Call them "tied" if you like, but acknowledge that by "tied" you mean not that the plays are equal but are very unlikely to differ by more than some small amount.

Messages In This Thread

• XG 4 ply RO of 64P-62 completed
  neilkaz -- Saturday, 23 January 2010, at 10:16 p.m.
  • XG 4 ply RO of 64P-62 completed
    Andreas -- Sunday, 24 January 2010, at 8:33 a.m.
    • XG 4 ply RO of 64P-62 completed
      Nack Ballard -- Sunday, 24 January 2010, at 9:11 a.m.
      • ty for the ATS link ! *NM*
        Andreas -- Sunday, 24 January 2010, at 1:02 p.m.
  • XG 4 ply RO of 64P-62 completed
    Gregg Cattanach -- Tuesday, 26 January 2010, at 8:57 p.m.
    • XG 4 ply RO of 64P-62 completed
      Matt Cohn-Geier -- Tuesday, 26 January 2010, at 9:10 p.m.
    • XG 4 ply RO of 64P-62 completed
      Stick -- Tuesday, 26 January 2010, at 9:13 p.m.
    • XG 4 ply RO of 64P-62 completed
      Stick -- Tuesday, 26 January 2010, at 9:39 p.m.
      • XG 4 ply RO of 64P-62 completed
        Gregg Cattanach -- Thursday, 28 January 2010, at 1:42 p.m.
        • XG 4 ply RO of 64P-62 completed
          Daniel Murphy -- Thursday, 28 January 2010, at 8:36 p.m.
          • XG 4 ply RO of 64P-62 completed
            Gregg Cattanach -- Friday, 29 January 2010, at 1:30 p.m.
            • XG 4 ply RO of 64P-62 completed
              Daniel Murphy -- Friday, 29 January 2010, at 6:07 p.m.
              • XG 4 ply RO of 64P-62 completed
                Nack Ballard -- Friday, 29 January 2010, at 7:08 p.m.
                • XG 4 ply RO of 64P-62 completed
                  Daniel Murphy -- Friday, 29 January 2010, at 9:43 p.m.
    • "Tied" vs. "TCTC"
      Timothy Chow -- Friday, 29 January 2010, at 10:30 p.m.