[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Bug-gnubg] Zadeh and mec mets

From: Joseph Heled
Subject: Re: [Bug-gnubg] Zadeh and mec mets
Date: Sat, 22 Jun 2013 17:19:09 +1200

Hi Rick,

It has been a long time. The mec26 was obtained by first getting the post crawford probabilities from rollouts, then from the program mec.cc (i.e. "first principles"). I can't remember where the other tables came from :(

However, I remember running a fair number of rollouts and evaluations of specific positions with different equity tables, and was always surprised how little effect that had, if at all.
A position has to be very marginal in some sense if an equity table change changes the move or cube action.

It has been a while, so perhaps someone can take a fresh look and see if my observation above is correct or not - by examining on a much larger scale (which was not feasible at the time).
Alternatively, one can build an accurate table today based on rollouts alone, and compare it to the rest.


On 22 June 2013 02:39, Rick Janowski <address@hidden> wrote:



I wanted to ask how the match equities for Zadeh and mec  METs (mec not mec26) are handled in gnu – the met files have no array of probabilities so it looks like the computer calculates from first principles. If so something has gone badly wrong with the parameters because the Zadeh equities are often more than 3or 4% different than the values tabulated in his paper of 75.  Checking those values in his paper I found that incredibly there was quite negligible error with the rolled out mets produced by Rockwell and Kazaross – less than 1% at all pre crawford scores in a 15 point match and when crawford and post crawford are also considered no error greater than 1.2%. Unfortunately the values you see in Gnu make it look like the worst MET of all.



Bug-gnubg mailing list

reply via email to

[Prev in Thread] Current Thread [Next in Thread]