Re: [Bug-gnubg] What MET is loaded by default with standard installation?

[Joseph Heled]

Sorry, I take back what I said. I mistakenly thought I was using Zadeh, 
but actually I typically use the table generated by mec.c

// (Copyright (C) 1996  Claes Thornberg (address@hidden)), using
//    Gammon rate 0.2600
//    Winning %   0.5000

It is somewhat close to Snowie 2.1 table.

For post crawford I use my own table from GNUbg rollouts. Here are the 
first values

{ 0.50000,
  0.484988779204,
  0.3195,
  0.302313595788,
  0.18935485,
  0.175375581897,
  ...
}


I am close to completing 500000 7 point matches between Woolsey and 
mec26 (above) (at 0 ply of course). Result is near mec26 50.06% vs. 
woolsey 49.94, So I am pretty confident in my statement the woolsey 
table is not up to the task.

-Joseph

Robert-Jan Veldhuizen wrote:
> I use the SW 2.1 MET, I think it's probably the best or second-best MET 
> GNUBG provides. The Jacobs MET is the other one.
>
> The Zadeh MET doesn't seem very realistic to me, with f.i. 28.75% MWC at 
> -2,-1Cr where simulations with GNUBG get a MWC percentage of well over 
> 30; probably also over 31. The SW 2.1 MET gives 31.5% MWC for this 
> particular score; Jacobs gives 31.0%.
>
> Kit Woolsey's MET really lacks at least a third digit to the entries. 
> This can sometimes produce pretty inaccurate results.
>
> For scores where both players are more than 15-away or even 25-away, 
> this only matters when someone plays a 15/25+pt match, and at these 
> scores the MET used probably won't make much difference anyway (they'll 
> likely be very close to each other all). The interpolation GNUBG uses in 
> these cases is probably sufficient for almost anything.
>
> It might be interesting to hear Neil Kazaross' opinion on this, as he's 
> a very good and experienced match player.