[Top][All Lists]

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

Re: [Bug-gnubg] Cube decisions - 3-ply versus 2-ply

From: Øystein Johansen
Subject: Re: [Bug-gnubg] Cube decisions - 3-ply versus 2-ply
Date: Sat, 19 Dec 2009 18:33:28 +0100

Just a general comment:

I love messages like this one. I really shows that it is close to impossible to make general statements about which bot or which setting is better than an other. You can only make a narrow statement like: in positions like blah blah, we have done blah blah, which indicates that A makes better decisions than B in these positions. 

Even making narrow statements like this take a lot of effort, lots of rollout data, and a lot of time. It seems to me like you have really understood that! (... and it's not many people who understand that.)


On Sat, Dec 19, 2009 at 10:31 AM, Adi Kadmon <address@hidden> wrote:
Hello all,
I'd like to correct a statement I made several months ago on this matter. It then appeared to me (and I was certainly not one of the first to note it) that 2-ply cube decision evaluations were better than those of 3-ply - according to rollouts.
Now, at any rate with GNU 4/2009 version, on the basis of systematic checking of very many positions (with the help of 0-ply-RO, 1-ply-RO, 2-world class-ply-RO and 2-supremo-ply-RO), I think otherwise: 2-ply cube decision evalutaions are definitely better than 3-ply's in openning positions and relatively simple middle-game positions. (3-ply is too "daring" in declaring Doubles, Passes and Too Goods.) But as the game prgresses or becomes more complicated, the tendency lessens and finally turns around: 3-ply is superior particularly in positions where the side trailing in the race nevertheless has the advantage; in those endings where one side primes or closes out the other with most of the other's checkers already borne out; and more types of complicated middle-game or endgame. (There, 2-ply is too "caucious-conservative" in declaring Doubles, Passes and Too Goods).
I should point out that a similar state of things seems to prevail (though it was not as systematically checked) comparing in general odd-ply versus even-ply cube decision evalutaions.
-- Adi

reply via email to

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