Forum Archive :
Programming
In his report on the World Computer Olympiad here, Achim mentions that
gnubg's 2-ply is equivalent to "the NON-UNIX world's 3-ply"
(paraphrased). Is there any backgammon software besides Snowie that
messes this up?
At one time I thought Jellyfish might have had this wrong, but that is
clearly not the case. Jellyfish has pulldown functions for "Evaluate"
("level-5" no lookahead), "Lookahead" ("level-6"), and "2-ply" (or
"level-7").
I think Snowie/Oasya get's the credit for starting this mislabelling,
and I hope it stops there.
|
|
Robert-Jan Veldhuizen writes:
I guess there's some extra complicating factors which causes the
different views:
In backgammon, there are also rolls to be considered, not just moves
You can evaluate a position with or without a given dice roll
(similary: you can evaluate cube and play decisions)
Given a certain roll, is determining all valid moves and evaluating
their resulting positions a way of looking ahead or not?
Personally I agree very much that GNUBG's 0-ply (static) cube
evaluation is indeed 0-ply. It seems silly to me to call this 1-ply.
The actual position is fed directly into the NN; no lookahead
whatsoever.
However, when you roll something and then analyse this at GNUBG's
0-ply, one could say that you're looking ahead, since you play all
valid moves and evaluate the resulting positions to give the full
analysis results. GNUBG's doesn't consider this as looking ahead,
probably with the idea that all INDIVIDUAL valid moves are just played
then evaluated statically, so 0-ply. But there's something to say to
call the whole operation of finding the best move at 0-ply, actually a
1-ply analysis (consisting of multiple 0-ply evaluations?)...
Confusing stuff for sure..!
|
|
Jordan Lampe writes:
When I was a kid and we had computer chess programs, there was no such
thing as 0-ply. 1-ply meant that the program looked at each possible
move (that's the ply), evaluated the position, and then returned the
result (i.e. the best move). 2-ply meant that it looked at each
possible move and each possible response, etc.
Between backgammon and chess, there is much more reason in backgammon
to evaluate a position statically, on its merits alone, as compared to
chess where I've never had the experience of asking a computer to say
who is winning and by how much (remember I was a kid back then).
It seems that nowadays, when you ask a backgammon program to make a
move, it does a search of all legal moves ("1-ply") and then evaluates
each position statically ("0-ply"). A 1-ply search but a 0-ply
evaluation, if you will. When it comes to a doubling decision, the Gnu
count is clearly more accurate, because "Snowie 1-ply" doesn't do any
lookahead at all when deciding.
So, while it's possible to argue that Gnu's count is more accurate, I
also think that Snowie set the precedent. Snowie's count isn't
horrible, and a whole generation of backgammon players has been
confused by Gnu's insistence on the "correct" method.
|
|
Jørn Thyssen writes:
I don't think it's possible to find a consistent scheme that matches
Snowie's way of counting, so the pragmatic solution is just to change
gnubg to match Snowie, and forget about finding such a scheme. Perhaps
the gnubg developers just have to accept that somebody came up with
counting schemes that are now industry standard, and just abide to it
:-)
I've previously refused to consider changing gnubg, but I'm starting
to soften up, and I'll consider suggested that we take the pragmatic
solution and change gnubg to follow the same counting scheme as
Snowie... To my knowledge this is the solution chosen by the Zbot
team; internally counting starts from zero but externally it starts
from one. However, I don't think this means that Zare agrees with the
numbering used by gnubg :-)
|
|
Douglas Zare writes:
I don't feel Snowie's conventions are necessarily the most natural.
From one perspective, I would prefer to use Snowie's count for checker
play decisions and gnu's 0-based count for cube decisions. (A 1-ply
checker play search evaluates the moves on 0-ply.) Will we impose that
hybrid on Zbot users? Of course not. As far as the users are
concerned, Zbot will follow Snowie's convention, at least for
evaluations. The point is to communicate.
|
|
|
|
Programming
- Adjusting to a weaker opponent (Brian Sheppard, July 1997)
- Anticomputer positions (Bill Taylor+, June 1998)
- BKG 9.8 vs. Villa (Raccoon+, Aug 2006)
- BKG 9.8 vs. Villa (Andreas Schneider, June 1992)
- BKG beats world champion (Marty Storer, Sept 1991)
- Backgames (David Montgomery+, June 1998)
- Blockading feature (Sam Pottle+, Feb 1999)
- Board encoding for neural network (Brian Sheppard, Feb 1997)
- Bot weaknesses (Douglas Zare, Mar 2003)
- Building and training a neural-net player (Brian Sheppard, Aug 1998)
- How to count plies? (Chuck Bower+, Jan 2004)
- How to count plies? (tanglebear+, Mar 2003)
- Ideas for improving computer play (David Montgomery, Feb 1994)
- Ideas on computer players (Brian Sheppard, Feb 1997)
- Introduction (Gareth McCaughan, Oct 1994)
- Measuring Difficulty (John Robson+, Feb 2005)
- Methods of encoding positions (Gary Wong, Jan 2001)
- N-ply algorithm (eXtreme Gammon, Jan 2011)
- Neural net questions (Brian Sheppard, Mar 1999)
- Pruning the list of moves (David Montgomery+, Feb 1994)
- Search in Trees with Chance Nodes (Thomas Hauk, Feb 2004)
- Source code (Gary Wong, Dec 1999)
- TD-Gammon vs. Robertie (David Escoffery, June 1992)
- Training for different gammon values (Gerry Tesauro, Feb 1996)
- Training neural nets (Walter Trice, Nov 2000)
- Variance reduction in races (David Montgomery+, Dec 1998)
- Variance reduction of rollouts (Michael J. Zehr+, Aug 1998)
- Variance reduction of rollouts (Jim Williams, June 1997)
- What is a "neural net"? (Gary Wong, Oct 1998)
- Writing a backgammon program (Gary Wong, Jan 1999)
From GammOnLine
Long message
Recommended reading
Recent addition
|
| |
|