bug-gnubg
[Top][All Lists]
Advanced

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

[Bug-gnubg] Question about movefilters and the best tractable gnubg perf


From: Jonathan Elesha
Subject: [Bug-gnubg] Question about movefilters and the best tractable gnubg performance settings
Date: Fri, 25 Nov 2005 12:52:44 +0000

Hello,

I've been trying to work out what the best settings for gnubg are in
terms of plys and move filters for my specific PC. I was wondering if
I could get some clarification as to what you people think are the
best settings.

I assumed that using the move filter Huge
set movefilter 1 0,0,20,0.440
set movefilter 2 0,0,20,0.440
set movefilter 3 0,0,20,0.440
set movefilter 3 2,0,6,0.110
set movefilter 4 0,0,20,0.440
set movefilter 4 2,0,6,0.110

and 3 ply for chequer and cube decisions would produce the best player
(that my machine can handle in a reasonable time) but now I'm not so
sure.

>From looking at the hints I can see that virtually all the moves are
only being evaluated at 1ply.
In which case the settings I am using will make Gnubg very astute with
very close good moves which seems to happen infrequently.

Now I'm starting to think that the best setting for gnubg in terms of
movefilters might be to automatically accept moves in. As in the
default settings for the gnubg-no-gui:

Move filters:
  Move filter for 1 ply:
    keep the first 8 0-ply moves
  Move filter for 2 ply:
    keep the first 2 0-ply moves and up to 3 more moves within equity 0.1
    keep the first 0 1-ply moves
  Move filter for 3 ply:
    keep the first 16 0-ply moves
    keep the first 4 1-ply moves
    keep the first 0 2-ply moves
  Move filter for 4 ply:
    keep the first 8 0-ply moves
    keep the first 0 1-ply moves
    keep the first 2 2-ply moves and up to 3 more moves within equity 0.1
    keep the first 0 3-ply moves

  This way moves which seem ok will get pushed through and it may turn
out that they are much better in further branches. Although this seems
to increase the performance massively, and since the top moves are
pushed in, we may still miss out on lower moves which turn out to be
better further in the game tree. So we widen the breadth of the search
tree but limit the depth (as 2 ply is more realistic with the default
settings above than 3ply).


Sorry for waffling on there. I hope that makes some sense. So
ultimately I'm asking what do people think are the best move filter
settings that are reasonably tractable.Whether you think more depth or
more  breath is better for backgammon and gnubg?

My second question is that if a movefilter is set to:
movefilter 3 0 0 0 0.0

Does this count as there being no pruning at the relevant ply, IE the
filter effectively being disabled at that ply.

Thanks
Andrew Cunningham




reply via email to

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