BGonline.org Forums

31P-11, data

Posted By: Nack Ballard
Date: Wednesday, 21 July 2010, at 4:34 a.m.

In Response To: 31P-11, data (David Rockwell)

What I meant is, "I don't see how this sort of same-seeding can net gain."

The issue is that same-seeding reduces the effective number of trials. (Or at least that's our assumption -- if it doesn't, then both sides of our discussion are moot.) It sounds clever to isolate the effect of the higher ply by same-seeding, but IMO the best way to do that is to maximize the (effective) number of trials.

Consider the extremes. If you roll 1k on each ply, you're desperate to have twice as many trials in order to reduce the variance and obtain a meaningful result, and different seeds guarantee you've done that: clearly that would be preferable to same-seeding. OTOH, if you roll a billion trials on each ply, you've squeezed all the variance out anyway and there's no need to use the same seed.

In other words, you're already seeing the effect of the higher ply by rolling out the position a sufficiently large number of trials on different seeds. We can't have it both ways by saying that we've rolled it out so many trials that the variance is gone and therefore the .003 difference we are looking means something special (i.e., it's not just variance) and yet somehow we couldn't say that the .003 is meaningful while using different seeds.

I much appreciate your offer to reroll, but no need -- I doubt same-seeding makes much difference. IMO, mainly what it does is reduce the effective number of trials, but not by much. OTOH, if after due consideration you disagree with what I've said above and want to keep same-seeding parallel plies, I won't complain, having said my piece. Mostly, I'm happy to have the data. :)

Nack

Messages In This Thread

• 31P-11 4-ply
  David Rockwell -- Monday, 19 July 2010, at 11:01 p.m.
  • 31P-11 4-ply
    neilkaz -- Monday, 19 July 2010, at 11:14 p.m.
    • 31P-11 3-ply
      David Rockwell -- Tuesday, 20 July 2010, at 12:09 a.m.
      • 31P-11, data
        Nack Ballard -- Tuesday, 20 July 2010, at 1:00 a.m.
        • 31P-11, data
          Bob Koca -- Tuesday, 20 July 2010, at 1:27 a.m.
          • 31P-11, data
            David Rockwell -- Tuesday, 20 July 2010, at 1:54 a.m.
            • 31P-11, data
              Nack Ballard -- Wednesday, 21 July 2010, at 4:34 a.m.
        • 31P-11, data
          David Rockwell -- Tuesday, 20 July 2010, at 1:57 a.m.
          • 31P-11, data
            Nack Ballard -- Wednesday, 21 July 2010, at 4:38 a.m.
  • 31P-11 4-ply
    Christian Munk-Christensen -- Tuesday, 20 July 2010, at 11:33 a.m.
    • 31P-11 4-ply
      Leonardo Jerkovic -- Tuesday, 20 July 2010, at 12:05 p.m.
    • 31P-11 4-ply
      Stick -- Tuesday, 20 July 2010, at 12:19 p.m.
      • 31P-11 4-ply
        Christian Munk-Christensen -- Tuesday, 20 July 2010, at 1:24 p.m.
        • 31P-11 4-ply
          Stick -- Tuesday, 20 July 2010, at 1:48 p.m.
          • How to organize your study time
            Timothy Chow -- Tuesday, 20 July 2010, at 11:30 p.m.
            • How to organize your study time
              Christian Munk-Christensen -- Wednesday, 21 July 2010, at 6:16 a.m.
      • 31P-11 4-ply
        Nack Ballard -- Wednesday, 21 July 2010, at 4:46 a.m.
      • 31P-11 4-ply
        mtuhtan -- Wednesday, 21 July 2010, at 5:26 a.m.