|
| ||||||||||||||||
I very recently posted two Snowie truncs (in separate posts). The first trunc is the one I originally discussed from memory here of 51$-43S-44 (left-hand position). The second trunc I just ran today of 61P-54S-51H-43S-44 (right-hand position).
Below is data from five bot sources.
Key: S = Split (Bar/21 24/20(2) 13/9), K = Kill (Bar/21 13/5*/1*), Y = Yoga split (Bar/21 24/20 13/5*), c = cross (Bar/21 24/20(2) 8/4), C = Cross (Bar/21 24/20 8/4(2), M = Mayfair split (Bar/21 13/9 8/4(2)). Error sizes are in thousandths. After the brackets is a bot symbol ( ` Jellyfish 1-ply, ~ Snowie, / GnuBG, < XG 3-ply) and the number of trials in thousands, rounded down (an asterisk repeats a number), with t standing for trunc.
For example, "[K Y53 S59] `10t" below means that K is best, Y is -.053 and S is -.059 in a Jellyfish rollout of 10000+ trials.
- 51$-43S-44
[No Jellyfish data]
[S K48 c52 M75 Y84 C90] ~1t**.6*.3 details
[No Snowie full data]
[No GnuBG data]
[S K6 M32 c42] <20*5 details
61P-54S-51H-53S-44
[K Y53 S59] `10 (from Kit/Hal book)
[S K20 Y58 c86] ~1t**.6 details
[K S3] ~5 details
[K S6 Y47] /46*3 details
[No XG data]
Different bots (and trunc vs full) are producing different results. What we are mostly lacking is a fair comparison of the two positions using the same bot. That is why I ran the second Snowie trunc: it finished quickly and provided us with our first apples-to-apples comparison.
Isolating just the Snowie truncs:
- 51$-43S-44 ............... [S K48 c52 M75 Y84 C90] ~1t**.6*.3
61P-54S-51H-43S-44 [S K20 Y58 c86] ~1t**.6
Whatever bias the trunc have, it probably applies roughly to the same extent in both positions. That is cogent evidence that S is relatively better compared to K in the first position than in the second position (primarily because of the extra spare on the midpoint, IMO), which is what I was getting at here.
We have a Snowie full rollout and a GnuBG rollout that closely agree and average [S K4] for the left-hand position. At the same time, we have an XG rollout that yields [K S6] in the right-hand position. That's a difference of .010. The truncs indicate a wider difference of .028, and from experience that suggests to me that an XG full will not quite agree with Snowie or GnuBG, given a rollout of the same double 4s position.
It would be great if someone does a Snowie and/or GnuBG rollout of the left-hand position, and/or an XG rollout of the right-hand (New Ideas) position, which will give us a direct bot-to-bot full rollout comparison.
Nack