| |
BGonline.org Forums
Nacbracs
Posted By: Matt Ryder In Response To: Nacbracs (Nack Ballard)
Date: Sunday, 27 December 2009, at 12:07 p.m.
Hi Nack,
My suggestion for combining your nacbracs methodology with the B/W error scale was an off-the-cuff response to your explanation. I wouldn't dream of "distancing [myself] from...the mainstream of communication for early game move comparisons"! I'll leave the Z vs. S jihad to radicals like Timothy Chow :-)
In my opinion, a system of communication that only one person buys into is the worst form of solipsism. And while my study habits tend to be cave-like, it's refreshing to go outside once in awhile to experience the sunlight (I'm with Plato on this :-)
However, in case you misunderstood my motive, I wasn't at all suggesting thousands of a point be replaced with a more imprecise 'truncated' variant. In many cases, I would want to communicate bot results at the highest level of precision. My proposal was simply to expand the vocabulary of nacbracs to include the B/W error scale.
Not to belabor my point, but I do think there are contexts where it's more useful to think in terms of error ranges a la the B/W scale than the hard numbers. (It can get overwhelming if you consider the Niagara in terms of its composite H2O molecules.)
For example, I'm currently looking at some older data off Tom Keith's site where the DMP equities were mathematically derived from the cubeless ones. I'm comparing that with much more recent DMP data posted by Stick, rolled out AtS. (My intent is to figure out how useful such mathematical interpolations are in real terms, compared with AtS results.) Given the big differences in the way the data was calculated, I think it would be fairly irrelevant to post the actual numbers. I believe the B/W scale is better suited to highlight some of the aspects of that analysis.
In the early game especially, the equities are by no means set in stone. The numbers wobble this way or that depending on a given bot's particular neural net weightings, the number of rollouts, the truncation settings, the precision of its analysis, whether it reached statsig etc. Given this degree of volatility, does it really make pragmatic sense to 'order' the Slot and Up plays in your example 51S-21 [$ U3 D28 S34]? (In real terms, it would probably be a big mistake for the average player to try $ as it must result in a decidely complex game?)
In general, I feel way too much attention is given to finding the 'best' play, and to the relative 'ordering' of plays. We should rather be examining how close the plays are to each other, not which are superior by a whisker; and that's what the B/W scale communicates brilliantly.
Merries and Happies to you too,
Matt.
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.