[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Bug-gnubg] Even, odd and half plies (WAS: New Contact Net Er ror Ra
From: |
Jim Segrave |
Subject: |
Re: [Bug-gnubg] Even, odd and half plies (WAS: New Contact Net Er ror Rates) |
Date: |
Tue, 25 Feb 2003 22:45:38 +0100 |
User-agent: |
Mutt/1.2.5.1i |
On Wed 26 Feb 2003 (10:25 +1300), Joseph Heled wrote:
>
> I think this is what Doug Zare calls "bot confusion", and he uses it to
> measure exactly that - how good is the bot understanding of that
> position. Obviously when ply0 == ply1 *for all positions" you have a
> perfect bot (by induction).
I just re-read that particular article again, but I'm still
curious. Can you tell me which of the following is the best
description of the behaviour in those positions where odd plies are
least desireable:
let t = 'true' equity of a position (or even the
equity from an extnsive rollout.
let e(n) be the absolute value of the difference between the equity
given by an n ply evaluation and t (abs value of error)
Is the case that:
1) e(2) < e(0) < e(1) - the error increases on odd evaluations
2) e(2) < e(1) < e(0) - odd plies are better than the lower level
even ply, but e(1) - e(0) is much smaller
than e(2) - e(1) - implying that odd
evaluations are more accurate than lower level
ones, but the difference is negligible.
or 3) something else?
And can anyone give me an explanation of why this would be so?
--
Jim Segrave address@hidden