Elf's margin of error
What is the margin of error of Elf's winrates (near 50%)?
OC, there are two errors involved. One is the accuracy of Elf's preferences, i.e., is one play or other closer to Elf's style? The variance of the estimates gives us direct information about that. The second error is whether a play is a mistake or not. We have no direct evidence of that, but that is the more important question.
As I stated I have recently revised my guess about the second margin of error for Elf to 4% from 5%. I have now slightly revised that to 4½%. As if I can tell.
OC, there is no hard and fast threshold, and the number of playouts matter. I am thinking of 10k playouts or more.
Anyway, here is a position to consider.
- Click Here To Show Diagram Code
[go]$$Bc Large knight's enclosure
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . 2 . . . . . . . . . 5 . . . . . |
$$ | . . . , . . . . . , . . . . . , 1 . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . b . . |
$$ | . . . . . . . . . . . . . . . . a . . |
$$ | . . 3 , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . 4 . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]
is not Elf's style. It prefers
a, estimating Black's winrate as 50½% with 30.3k playouts.
, OTOH, gets a winrate estimate of 46% with 19.3k playouts. (Actually,
is not even on Elf's radar, getting 0 playouts. It inherits its winrate estimate from Elf's first choice for
, the enclosure at
b.) Anyway, I have no problem thinking that
is not Elf's style, given the 4½% winrate difference and the 0 playouts, but I do have a problem thinking of it as an error.