Forum Archive :
Snowie
Vince Mounts writes:
> 1) I get _Move Error Rate_ as a percentage. What is this a percentage
> of?
Nothing. It isn't a percent of anything.
I believe that it's the average normalized gammon- and backgammon-price
adjusted total cubeless error per 100 unforced moves.
Unpacking this:
- it's a cubeless error rate. The cube doesn't enter into it in any way.
- it's gammon- and backgammon-price adjusted. Snowie uses its
match-equity table and the current cube level to calculate the gammon
and backgammon prices, and uses these rather than the money values.
- it's normalized. An error that costs you the equivalent of the value
of 100% of a plain win is an error of size 1 -- regardless of the size
of the cube or the match score.
- it's the average of this per 100 unforced moves.
Basically, if you defined the error threshold as 0.0 and then went
through and added up all your errors in a match (add up the figures
in parentheses), divided by the number of unforced moves and multiplied
by 100, you would have it.
> 4) What is variance reduction. Why should I believe snowie when it says
> my 288 game roll out is like 30000 games? What is it doing to get
> the "extra" games?
Take a look at the Backgammon Galore newsgroup archive, Programming
Section. There are several articles on variance reduction, although
they are somewhat technical.
If you have rolled out a complete game on 2- or 3-ply, you can
trust the number of equivalent games. If you have truncated, especially
after only a few ply, you should regard the rollout as a little less
reliable than the equivalent games says.
David Montgomery
monty@cs.umd.edu
monty on FIBS
|
|
|
|
Snowie
- Announcement (Olivier Egger, Apr 1998)
- Checker-play-according-to-score bug (Peter Schneider+, June 2001)
- Error rates (Gregg Cattanach, Oct 2000)
- Hints and questions (Achim Müller+, Aug 1998)
- Luck calculation (Gregg Cattanach+, Dec 1999)
- Questions and answers (David Montgomery, Dec 1998)
- Running in low priority (lmfback+, Oct 2004)
- Snowie 4.0 (SnowieGroup Info, Oct 2002)
- Snowie 4.3 update (Gregg Cattanach, July 2005)
- Snowie cube evaluation (Kit Woolsey, Sept 2007)
- Snowie vs GNU (Stanley E. Richards+, Oct 2005)
- Snowie vs. Jellyfish (Mark Driver, Apr 2001)
- Snowie vs. Jellyfish (Daniel Murphy, Oct 2000)
- Snowie vs. Jellyfish (Gregg Cattanach+, Sept 2000)
- Snowie vs. Jellyfish (Wayne Crookes, Jan 1999)
- Snowie vs. Jellyfish (Kenneth M. Arnold+, May 1998)
- Terminology (Alexander Nitschke, Sept 1998)
- Using rollouts (Michael J. Zehr+, Oct 1998)
From GammOnLine
Long message
Recommended reading
Recent addition
|
| |
|