"ah...Clem" <
[email protected]> writes:
On 9/26/2022 7:03 PM, Stick Rice wrote:
You can see the Keith modifications that I'm referring to about
3/4ths of the way through this page. And yes, it gets it right.
http://bgonline.org/jf.html
There's a lot to digest on that page, but my reading is that the short version of the "modified" Keith count is to expand the doubling window
for longer races. Am I missing something?
Not IMHO. I tested Stick's ideas some months back. Where my Isight
method (in its standard version) and the Keith count ignore the race
length and Trice classifies into short and long races, Stick uses 4
cases:
l <= 80
80 < l <= 100 (4 % of the races)
100 < l <= 120 (0.3 % of the races)
120 < l (exactly 1 (!) of 50000 positions in Tom's database)
On Tom Keith's endgame database the total equity loss is as follows:
Trice (with "my" pip adjustments): 1424
Keith : 1262
Stick's modified Keith count : 1182
Isight (standard) : 1064
Isight (short/long, see page 40) : 1041
Since the Keith count requires more adjustment work than mine, my method
is more accurate AND needs less effort, see paragraph 2 on page 17.
And, if you care about races longer than 80 pips, see page 40. For money sessions and "long" races above 50 pips this boils down to (in your
Python code terminology)
POLT = LC + LC/10 + 2
, which with a window size of 4 is the good old "10 % +/- 2 pips"
rule. For "short" races below 50 pips you get
POLT = LC + 2*LC/10 - 3
Both are very simple to calculate. For exactly 50 pips both formulae
give the same POLT of 57.
The reason that this is better than "standard Isight" becomes clear from
figure 13 (Thorp is "10 % +/- 2 pips"), and the reason that this is only slightly better becomes clear from figure 14.
But this short/long modification gets all three of Stick's examples
right.
Best regards
Axel
--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)