63R at GS, data
Posted By: Nack Ballard In Response To: Huge varience in your two RO's running with a 9 at GS (Stick)
Date: Friday, 1 January 2010, at 9:34 p.m.
In Response To: Huge varience in your two RO's running with a 9 at GS (Stick)
You may have forgotten that David and I strongly suspected and/or proved that the Gnu 31k rollout of 63R at GS is corrupt. However, according to my records, only the first 20k were done by HH (not his fault -- bad older version). The last 10k were done by GH, and should be trustworthy.
You can cleverly extract 10k of valid data from the corrupt file by finding out the margin that existed after the corrupt 20k (attached to an e-mail from HH or attached to the e-mail you sent GH to extend to 31k). Then subtract twice the 20k margin from thrice the 31k margin.
For example, if at 20k the (corrupt) result favored S by .0092, you can deduce that the final (uncorrupt) 10k must have been (.0047 x 3) - (.0083 x 2) = -.0025. The minus sign tells you that it crossed over and S is no longer better but rather R is; i.e., the uncorrupted 10k result in this example is [R S2.5].
If that explanation isn't clear, just tell me the margin (or send me the file) and I'll do it. (I've performed this adjustment before and the results have always made sense.) In theory, the other uncorrupted numbers (w/g/bg's, etc.) can be similarly extracted, but it's too much work; the margin is the vital piece of information.
Thus far, the Gnu rollouts we have are
[R S5.4] 15k Supremo (DR's rollout)
[S R3.3] 1k (some independent file you once sent me)
We also have the R equity of .2837 from the (5k) 54 at GS file. The R equity from DR's (15k) file is .2864, so if we weight-average them by number of trials (including the 1k result above, that comes to [R S4.3]. Depending on how much extra reliability you give to Supremo (remembering that variance is always afoot), we might raise R's edge some above 4.3 towards DR's 5.4 margin.
If you are able to find the key HH or GH e-mail, we can average in another 10k of data.
It sounds like Neil has added 63R at GS to his XG 4-ply list as well.
Messages In This Thread
BGonline.org Forums is maintained by Stick with WebBBS 5.12.