Many thanks for this, Eugene

Joe

On Thu, 2009-01-08 at 21:50 -0800, Eugene Deon wrote:
> If you're into studying hooks lately, you might find this useful:
> 
> 
> I took the idea of playability and tried computing the relative
> imporance of all the words with respect to needing to know and find
> the hooks of that word to make the best play possible.  This list is
> based on (only) 2 million games of quackle speedy self-play.  No
> surprise the 2s win out.  No matter what letter is hooked, the word
> accumulates equity difference, so the more hooks a word has, the
> higher it tends to be in the list.
> 
> 
> And the winner is..... WE!
> 
> 
> http://www.eugenedeon.com/OWL2_hookability_jan_2009.txt
> 
> 
> exact details:
> 
> 
> -let A be the set of all words hooked in the best play
> -if A is nonempty, find the first subtoptimal play which hooked the
> set of words B such that A - B is non empty
> -accumulate the equity difference of the two plays to each word in A -
> B
> 
> 
> example:
> 
> 
> quackle's top 3 moves are:
> 
> 
> 1) DARK, (HINGE)R, A(TONE) 30.2109 
> 2) ARK, (HINGE)R, A(TONE) 27.789
> 3) KA(N)TAR, A(TONE) 24.9023
> 
> 
> the best played used an R hook on HINGE, and an A hook on TONE.  So
> did the next play, so skip it.  the 3rd best play didn't hook anything
> on HINGE, so accumulate 30.2109 - 24.9023 to HINGE.  A(TONE) was still
> made on the 3rd play, so don't accumulate points to it.  This is
> obviously an arbitrary choice, and I don't claim for it to be correct
> (or for any of the data to be accurate).  I'll keep running it for a
> while.  Maybe until 30 million games or so.  Then the middle section
> should be more accurate.  The top is fairly stable.
> 
> 
> Eugene d'Eon
> 
> 
> 
> 
> 
>  

Reply via email to