Cubeful rollout bug
Peter Schneider
X cannot win gammon since 0 has 4 checkers off. Without the 6% gammons, it's
perhaps no double any more.
Has anybody else had bugs related to cubeful Snowie rollouts? I have useless
results nearly every time.
Greetings,
Peter aka the juggler
================================================================================
====
File: C:\Snowie Documents\My exported files\vesko_2001_05_31_2b_game2_mv28.txt,
01:36:46, 05.06.01
Snowie Professional Edition Version 3.2 Output (Export v2.10)
--------------------------------------------------------------------
| juggler (X) vs. vesko (O) |
| 5 point Match |
--------------------------------------------------------------------
Match to 5. Score X-O: 0-0
-------------------------- Move 28 X -------------------------
X on roll, cube action
+24-23-22-21-20-19-------18-17-16-15-14-13-+
| O O O X | | |
| O O O | | |
| O O O | | | S
| O | | | n
| | X | | o
| |BAR| | w
| | O | | i
| | | | e
| X X | | |
| X X X X X | | |
| X X X X X | | X |
+-1--2--3--4--5--6--------7--8--9-10-11-12-+
Pipcount X: 102 O: 44 X-O: 0-0/5 (1)
Men Off X: 0 O: 4
CubeValue: 2, X owns Cube
28. X: Double
O: Pass
Rollout Money equity: 0,399
0,0% 6,0% 71,6% 28,4% 9,4% 0,0%
95% confidence interval:
- money cubeless eq.: 0,399 ą0,007,
- live cube no redouble: 0,883 ą0,008,
- live cube redouble take: 0,851 ą0,009.
Rollout settings:
Full rollout,
1000 games (equiv. 66416 games),
played 3-ply score-based (medium, 33%), cube 3-ply,
seed 1, with race database.
Evaluations
1. Redouble, take 0,885
2. No redouble 0,858 (-0,027)
3.* Redouble, pass 1,000 (+0,115)
Proper cube action: Redouble, take
Live cube
1. No redouble 0,883
2. Redouble, take 0,851 (-0,032)
3.* Redouble, pass 1,000 (+0,117)
Proper cube action: No redouble, take 21%
neghe Onegu
I inserted the position into my Snowie and it doesn't show the gammons.
roll lucky and plya skillfull
"Peter Schneider" <schne...@gmx.net> wrote in message news:<3b1ff47d$0$26420$6488...@news.comundo.de>...
Peter Schneider
thanks for your effort. I wondered why nobody answered. I think that the bug
occurs only if
-- the rollout is 3-ply *cubeful*, cube 3-ply (the bug does not appear in
cubeless rollouts, not sure about the 3-plies)
-- Some higher number of games are already performed in the rollout. The bug
does sometimes not occur before a number of games are rolled out. This number
seemed to vary but was always smaller than the necessary number of games.
If that was true for your rollout, it would be interesting ...
BTW, the bug seems to be position independent; I have not found a way to avoid
it yet (other than un-checking the cubeful checkbox).
Anybody else? Am I the only one who performs cobeful rollouts?
Greetings, Peter
Chase
I perform cubeful rollouts and have never run into the problem.
However, I seem to remember reading somewhere that this bug only shows
up under certain conditions (as you state in your post). I also seem
to remember that one of the conditions necessary for the bug to appear
is that it be a FULL, not a truncated rollout. I'm not entirely sure
this is the case, but that's what I recall.
Chase
_____________________________________________________________________
Vist GammonQuest, for a HOST of online tournaments!
http://www.mindspring.com/~dgay/GammonQuest/
To respond via email, replace "USERNAME" with "demiga" in my address.
Peter Schneider
I could only see your posting via deja/google, so I answer via it.
[I reported a Snowie rollout result which gave me some percent gammon
chances although my opponent already had borne off some checkers (!),
and asked whether other Snowie users had made similar observations.
Chase said:]
[...]
> I also seem
> to remember that one of the conditions necessary for the bug to appear
> is that it be a FULL, not a truncated rollout. I'm not entirely sure
> this is the case, but that's what I recall.
[...]
Indeed, I do perform untruncated rollouts (usually over night ;-)).
I'll check your idea, do some more research and post the results in a
week or so.
Still, every additional observation or idea would be welcome.
Greetings, Peter
aka the juggler on FIBS
Christopher D. Yep
>
> Hi Chase, hi all,
>
> I could only see your posting via deja/google, so I answer via it.
>
> [I reported a Snowie rollout result which gave me some percent gammon
> chances although my opponent already had borne off some checkers (!),
> and asked whether other Snowie users had made similar observations.
Hi Peter,
If you're using Snowie 3, I believe this problem specifically only
occurs when you perform a rollout with checker play according to score
(this specifically also means that you are performing a cubeful rollout,
otherwise this checkbox won't even be available).
Specifically sometimes the rollout will show one side winning gammons
when gammons are impossible. Other times one side will be shown winning
*zero* gammons, when clearly gammons are possible (or even likely).
I don't often perform "checker play according to score" rollouts since
the rollouts take considerably longer than normal cubeful rollouts (with
checker plays based on cubeless money equity) without improving accuracy
very much. However the general consensus among several users I have
talked to is that the overall *equities* are likely correct even though
the breakdown (backgammons, gammons, normal wins) is incorrect.
For example, here are the rollout results that you reported on June 7,
2001:
Rollout Money equity: 0,399
0,0% 6,0% 71,6% 28,4% 9,4% 0,0%
95% confidence interval:
- money cubeless eq.: 0,399 ą0,007,
- live cube no redouble: 0,883 ą0,008,
- live cube redouble take: 0,851 ą0,009.
The above hypothesis states that the money cubeless equity is 0.399
despite the fact that the displayed numbers (0.0%, 6.0%, 71.6% 28.4%
9.4% 0.0%) actually imply a 0.415 money cubeless equity. Similarly, the
hypothesis states that the 0.883 and 0.851 numbers are in fact valid.
One hypothesis is that Snowie always (accurately) keeps track of the
overall equity, but that there is some type of display flaw that causes
it to erroneously report some of the wins as gammons or vice-versa.
To test this hypothesis, run the same position *without* checking
"checker play according to score." In the position that you posted on
June 7, 2001 (5-away/5-away with cube on 2), this will likely not make
much difference in the final results. See if the equity comes close to
equity that Snowie reports with "checker play according to score."
Chris
Peter Schneider
A long time ago, I said:
>> [I reported a Snowie rollout result which gave me some percent
gammon
>> chances although my opponent already had borne off some checkers
(!),
>> and asked whether other Snowie users had made similar observations.
Chris said:
> If you're using Snowie 3, I believe this problem specifically only
> occurs when you perform a rollout with checker play according to score
[...]
> However the general consensus among several users I have
> talked to is that the overall *equities* are likely correct even though
> the breakdown (backgammons, gammons, normal wins) is incorrect.
[...]
> One hypothesis is that Snowie always (accurately) keeps track of the
> overall equity, but that there is some type of display flaw that causes
> it to erroneously report some of the wins as gammons or vice-versa.
> To test this hypothesis, run the same position *without* checking
> "checker play according to score." [...] See if the equity comes close to
> equity that Snowie reports with "checker play according to score."
Thanks for your suggestion! First of all it's reassuring that it's not
some installation or operating problem of mine but happens to other
people too...--
The idea of a display flaw is nice but I think I remember to have seen
plain wrong results in positions with high gammon proportions... but I
have no position at hand, need some research.
And, provided that everything work alright (what it apparently does
not), a score-aware rollout should yield noticeable differences if a
gammon doesn't matter, specifically if the player only needs the
present cube value to win the match. Score-aware moves would then
optimize for overall wins and ignore gammon chances. Additionally, a
position normally sufficient for doubling the opponent out would not
be of any particular value in such a situation and might be rejected
in favour of a more solid position.
But, after all, this discussion does not affect the majority of games
very much, and I could live pretty well with none-score aware
rollouts.
Greetings, Peter
aka the juggler on fibs