BGonline.org Forums

54-s RO

Posted By: Philip Cherrington
Date: Monday, 14 February 2011, at 9:19 p.m.

In Response To: 54-s RO (neilkaz)

Comparing this result with Claudes shows emphatically why caution is essential in interpreting rollout results.

Neils rollout has confidence of less than .001 for the two moves. Adding them and doubling the result gives a maximum variance of .004 that the result would almost certainly be repeated within that limit. As they are .008 part now it is virtually certain that (in XGs eyes at least) S is (at least slightly, possibly moreso) better than R. Add to the fact that Neils rollout is of a significant length.

On the other hand Claudes rollout has confidence of .002 for maximum variance of .008 and rollout that is short enough to be known to give questionable results anyway. As the moves are returned as equal it is distinctly possible that the moves could easily be .008 apart if the rollout were continued long enough.

Rollouts should be continued until the confidence intervals are small enough with regard to the move separation that we can indeed be confident that the result would be repeated by another rollout (and therefore as accurate a result that we can get for now within the limitations of the current bots)

Messages In This Thread

• 54-s RO
  neilkaz -- Monday, 14 February 2011, at 3:43 p.m.
  • Data for 54-s
    Nack Ballard -- Monday, 14 February 2011, at 5:33 p.m.
  • 54-s RO
    Philip Cherrington -- Monday, 14 February 2011, at 9:19 p.m.
    • 54-s RO
      Nack Ballard -- Tuesday, 15 February 2011, at 3:05 a.m.