| |
BGonline.org Forums
Non-doublet Replies: Two(?) rules for beginners
Posted By: nack ballard In Response To: Non-doublet Replies: Five rules for beginners (Axel Reichert)
Date: Sunday, 23 April 2017, at 3:52 a.m.
The amount of work you've put into this is extraordinary, Axel! I feel like a loafer, by comparison. It had occurred to me that knowing the average error, as well as the extreme error, is valuable, but I didn't think you would want to go to that much work (and I know that I didn't).
I've been basing my data on rollouts I'd compiled from various people a while ago. Embarrassed, I admit that I had been unaware of the XG site. Moreover, it seems this site has rollouts that haven't yet made it into the accessible program -- or at least not into my version. :)
Some of these rollouts use 2-ply checker ply (with 3-ply cube). I can also go along with mixing these rollout equities with higher-ply rollout equities within the same rollout, if that's the best we have. What I eschew (based on tests, though with earlier bots) is the comparison of rollout equities to ++ evaluations. Granted, this only occurs with blunderish plays (else they would be rolled out), but those are the very plays for which you need to determine inclusion/exclusion in order to decide the exact model that will work for the threshold of .06 (or thereabouts).
For example, on the XG site is 51$-64 [R Z53 U61] <31 &e, where only R was rolled out. In such a case, I evaluate all plays on ++ and compare (unchecking "Use Opening Book" under Options/Settings/Analysis if not already done), which yields 51$-64 [R Z64 U72] &e. Then if at least two plays have been rolled out (not the case here), I combine the best of both data sets. For example, if there were a mystery play m and the XG site data was presented as 51$-64 [R m20 Z53 U61] <31*&e, I would use an aggregate of [R m20<31 Z64e U72e]a.
Of course, still better is to add Z and U to the original rollout. This would require either getting ahold of said rollout, or rerolling Z and U against the top one (perhaps until the square root of the sum of a relevant pair's confidence intervals is lower than the equity difference by which you've cleared the .06-or-whatever threshold).
[Update: The above paragraphs were written earlier. Curious, I did a rollout before posting, with the result 51$-64 [R Z64 U64] "<5 (admittedly, not nearly enough trials yet to determine whether Z or U is worse than -.06.]
If there aren't too many such instances, you might want to do just that. Note that you can reasonably exclude plays that (a) ++-eval worse than say -.120, and (b) of the type 21$-21w (regardless of what happens on ++ appeal), because these plays are unlikely to threaten any model you have. Without that work, someone may inevitably challenge your threshold claim. If you decide not to (understandably), I recommend the aggregate method.
Regarding the current model: Your "H > Potential Point" solution is excellent. (I had the feeling I was missing something!) I'm a bit concerned it could sound like you are hitting there with the potential of covering the point next turn. There may be no perfect substitute, but you might consider "H > makeable point" (or "H higher than makeable point"): You are able to make a point (but have a greater priority).However, having been inspired, I've come up with yet another way of handling it, the logic which you can check for yourself later.
By the way, if you use only the letter H, then you are using the entire Nactation meaning of H, including hitting loose and only once, and on the highest point possible, and by convention playing down (to/within the outer board) if there is a leftover letter. Otherwise, K/X/R instances should be accounted for. If that works poorly, I think "H" should be avoided and instead "Hit" should be written out. I haven't checked for violations but thought I should mention it.
I should have been clearer about the "CRMU" convention. Your first interpretation is correct. Going to the 18pt (or higher) is considered a "U" (Up) portion. Going beyond that is considered "R" (Run).
I actually did consider 63R-63 (-60) and 64R-62 (-62) [though I probably didn't notice 64S-63 (-64)], but it must have been when my threshold was .068. I overlooked 64S-64 (-77) for sure. Anyway, let's shelve CMRU, at least until I've looked at your simpler-looking mid/non-mid rule.
Assuming I understand the intent in your rule 3, I prefer using the word "else" instead of "or" in such cases. "D (else U)" is a short way of saying "play D, if it is legal, else play U." For the rest of my comments, for now I'll take the liberty of substituting "else" where you wrote "or."
In Nactation (non-doublets), "S" means a play that brings the large checker up and the other checker down from the mid (or the only play that brings one up and one down from the mid). This includes plays that hit on the far-side and bring a checker down from the midpoint. When a checker hits on the near side and brings a checker up, one uses "X," and although S/Z is acceptable (though not "S" or "Z" if you might mean the other). And for defining a partial play, S/Z can be confusing if not ambiguous.
The uni-areal letters of "U," "D" and "R," are fine for partial plays (e.g., leftover portions), as long as (a) it is clear in context whether one is referring to full-play or partial play, or (b) either full or partial can apply and thus no distinction is necessary -- usually the case. If you've already defined half the play with "hit," then it makes sense to use U, D or R for the other, except if/when you can find a clever way to (say) use H or X or some bi-areal letter to encompass some particular theme and save words.
A "D" portion can be played within the inner board as well as from the midpoint, whereas the down portion of S or Z must be played from the midpoint. For plays that can occur on the second roll, this only matters where 8/7 occurs, but it matters. For 32S-21, if the hitting default is "D," that means 8/7 6/4*. Instead, "down, else U" (or "down, else X") may be clear enough, while "from mid, else U" is crystal clear.
With that prefaced, I think you mean "U (else D)" in the middle of line 3, and "D (else U)" at the end of line 3. However, you have the aforementioned problem that 8/7 is a blunder when it accompanies 6/4*, which includes 32S-21, 43Z-21, and (horrors) 32S-41 and 43Z-41.
One more thing, though. On 6xS-31, X beats D by 13, 13, 32, which slightly more than compensates for the misplays on 6xS-21X and 6xS-41X (15, 4, 11, 14, 4, 6). More importantly, 6xS-31D represents two of our biggest current errors (.0595 and .0534, from your helpful list). Thus, instead of "hit from mid," I recommend "hit in outer board."
So, you might try something like...
3. (Leftover die): U (else D), If hit is in outer board.
.............................From mid (else U), for other hits.There's another problem, though. What happened to 21$-64R, 41$-64R and 51$-64R? We either have to bring back some variant of the CRMU convention, or list more exceptions (which I'll do in the recap). As it happens, though, $-64R can conveniently piggyback onto D-64R.
Moving on, there appears to be an unfortunate problem with where you moved 65R. Consider 52S-63H and 32Z-65H.
I'm dubious about the wording "most similar legal alternative." For 61P-64/63/62, which is "most similar," D or R? I'm not sure what you meant to cover here, but... I did notice the 61P-6xR problem after my last posting, just felt I had posted enough adjustments for the time being. I had planned to expand "else D" to "else R, else D," though thinking on it further I think it's okay to economize on the else's with commas. Assuming 65R is moved, the last line can now read as a rhyme: "43Z, else S, R, D."
Finally, I don't think it's clear that by "a" and "b" you are listing exceptions. I would like to see the word "exception" at least partially spelled out ("exc" is fine), and I don't really see the need for "a" and "b." You might consider indenting on a second line: For rule 1, "Exc: D-64R." And for rule 4, "Exc: D-x1$, 6xS-65K." Perhaps best is to list all three at the end (i.e., after "4," then line space, then "Exceptions: D-64R, D-x1$, 6xS-65K"). That will also give you more flexibility in dealing with 65R.
Another idea is to list the exceptions first, as technically they rank higher. Then, there is no need to list them as exceptions (they're in essence a pre-rule), and it provides a way to unclutter 65R from the hit/point rules.Tweaking further, I suggest the following rule set, until you/we find something better -- or you find holes :)
........$/D-64R, .D-x1$, and 6xS-65K else 65R. Otherwise......(1) Hit any xpt >2pt, unless you can P > xpt.
...(2) P, else 43Z, S, R, D.Convention for leftover die when hitting:
.........U (else D), when hit is in outer board.
.........From mid (else U), if hit is elsewhere.Nack
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.