BGonline.org Forums

H vs X question

Posted By: Nack Ballard
Date: Tuesday, 23 March 2010, at 6:28 a.m.

In Response To: H vs X question (David Rockwell)

Along with Neil's rollouts of 63S-41 and related positions, Nack mentioned that GNU has a bias for X which is not fully supported by Snowie and XG. Does anyone know if this bias also applies to sequences such as 43S-41 where the hit is not on the bar point? The dynamics are a little different, so it isn't clear to me whether this would follow or not.

Gnu's rollout bias is for H (not X). I assume your reference was to this post. To be sure we mean the same thing, I am stating that Gnu's rollout results have H coming out better than it's really worth (relative to X); I call that a Gnu H bias.

We can never be 100% sure if a Gnu H bias is real or not, even if we've rolled enough trials to sufficiently reduce variance, because it might be the other bot(s) that are off. What I've measured Gnu's result to is the average of Gnu and Snowie, or the average of all three bots when available.

The positions below are of the type where the first player splits with a 2, 3, 4 or 6, and the second player can play H (hit-and-down) or X (hit-and-split). Most follow the (Gnu H bias) pattern, with Gnu's biggest H biases at the top. (Plays other than H or X are omitted).

The reply position is listed, followed by the results of Snowie, Gnu, and XG (usually 3-ply) respectively. For example, in the first row, for 54S-41, Snowie is [H X14], Gnu is [H X38], and XG is [H X18].

54S-41 [H X14] 5k, [H X38] 7k, [H X18] 31k
52S-32 [H X14] 31k, [H X38] 5k, H X19] 62k
32S-43 [H X9] 18k, [H X27] 10k, [H X8] 62k
54S-43 [H X4] 20k, [H X26] 10k
43S-43 [X H23] 31k, [X H3] 46k, [X H11] 62k
64S-63 [X H82] 5k, [X H62] 5k
62S-41 [X H4] 10k, [H X15] 15k, [H X6] 62k
52S-43 [H=X] 20k, [H X12] 15k, [X=H] 62k
63S-41 [X H3] 10k, [H X9] 15k, [H X1] 46k
64S-41 [X H14] 25k, [X H4] 46k, [X H11] 31k
43Z-43 [H=X] 12k, [H X9] 15k
32Z-43 [X H18] 20k, [X H8] 46k, [X H10] 62k
43S-41 [X H8] 10k, [H X1] 46k
64S-21 [X H5] 20k, [X H1] 46k
43S-61 [X H23]31k, [X H13] 15k, [X H9] 62k
62S-62 [X H13] 20k, [X H11] 15k
32S-32 [H X19] 20k, [H X21] 15k
64S-64 [X H62] 5k, [X H60] 5k
32Z-32 [X H10] 31k, [X H4] 46k, [H X1] 62k
63S-62 [X H??] 5k, [X H20] 5k
62S-21 [H X??] 5k, [H X32] 5k
43Z-32 [H X10] 20k, [H X10] 18k
----------------------------------
63S-21 [H X17] 20k, [H X11] 31k, [H X13] 62k
63S-64 [X H44] 5k, [X H50] 5k
62S-63 [X H37] 5k, [X H50] 5k

Nactation key: H = hit-and-down, and X = hit-and-split. As always, the better play is listed first, then the other play with its error size in thousandths, and the number of thousands of trials (rounded down) after the brackets. For example, [H X14] 5k means H beat X by .014, after 5000+ trials (actually, 5184).

Nack

Messages In This Thread

• H vs X question
  David Rockwell -- Monday, 22 March 2010, at 5:39 p.m.
  • H vs X question
    Nack Ballard -- Tuesday, 23 March 2010, at 6:28 a.m.