[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: New backgammon engine: wildbg
From: |
Philippe Michel |
Subject: |
Re: New backgammon engine: wildbg |
Date: |
Sat, 25 Nov 2023 23:49:17 +0100 |
On Fri, Nov 24, 2023 at 11:21:00AM -0800, Trevor Slocum wrote:
> I have drafted an initial specification of a Universal Backgammon Interface
> (UBI).
>
> This protocol facilitates interoperable communication between backgammon
> engines and clients.
I have some comments on the <state> item and the evaluation commands and
replies.
The <state> doesn't contain the cube value and position. Should they be
transmitted earlier by another command ? It seems more natural to me
that they belong to the state.
A similar issue exists for the match score (in "away" format ? 0-0 for a
money game ? how to flag a Crawford game ?)
The OkDouble and OkMove replies don't contain the estimated equity ?
As described, maxmoves looks more like a minimum number of expected
moves (if there are enough legal moves).
XG, GNUbg, and probably BGBlitz and Snowie use move filters, evaluating
all moves at a low ply then doing a deeper evaluation of the most
promising ones. It may be more natural to use 0 for "respond with the
moves evaluated at the higher ply or rollout for the selected level" and
other values for "respond with <n> moves", possibly implying "expand
your move filter if needed".
Whether you agree with the latter or not, should it be expected that all
returned moves have been evaluated at the same level ? Otherwise that
level should be in the response (for each play). Maybe it should be
there anyway.