vpFREE2 Forums

Suspect video poker results

#1

Edmund Hack wrote :

?*************************************************************

?So, by your theory, you would need 235 trip aces, or on average about 580,500
hands. At 700 hands/hour, you are talking about 829 hours, or almost 21 weeks of
full time play.

In other words, it's nearly impossible to tell.

···

**************************************************************

Edmund,? I was thinking more of 10 times 23.5 ( success of?filling in trips) ?times 650 ( occurrence of trip aces) or about 15,000 hands.? That is still a lot of hands.? The less frequent the event you are trying to study, the more samples you need.? The finer the level of deviation from average you are trying to detect, the more samples you need.? The higher the confidence level you want to have, the more samples you need.??Unfortunately, the commercial programs out there don't let you break out hand type in their autoplay features.

As for Nyquist, I threw that out as a starting point.? It is true that video poker as a?whole is decidedly non normal and is better represented by a Poisson distribution.? This is due to some highly infrequent events ( RF in?particular).? An individual hand type should have a more normal distribution.

I don't recall seeing information on sample size for Poisson distributions.?

This is why the problem is interesting.? Running the numbers is the easy part.? Figuring out which numbers to run and what to do with the results is the challenging part.

I would agree with Bill Coleman on casino cheating at video poker.? I think the threat is overstated and there is very little data to support the allegations.? Once we establish a plan of attack for suspect machines, I think it will further support video poker being a fair game.?

More to follow.

________________________________________________________________________
Email and AIM finally together. You've gotta check out free AOL Mail! - http://mail.aol.com

[Non-text portions of this message have been removed]

#2

greeklandjohnny wrote:

An individual hand type should have a more normal distribution.

Individual hand distribution is normal - not merely "more" normal.

What makes overall video poker results skewed is the disparate
frequency with which each hand occurs. As you build a distribution
chart over a limited number of hands, the distribution is exceedingly
irregular. It smooths out as you increase the number of hands, but
can only be said to approach "normality" -- never achieving a true
normal distribution.

- Harry

#3

All this is pretty amusing to me. Given that the cumulative distribution function for all
(well almost all) video poker games is known, you can (trivially) construct any confidence
test you want. And you don't have to limit yourself to a particular "hand" (RF, trips,
whatever).

I'd recomend you start with the so called "K-S" test (kolmogorov-smirnov) which give
statistically meaningful results when you know the cdf (what you do with those results is
up to you). The K-S test is (so) easy to apply to for video-poker (though not partiucalrly
sensitive at the tails, but that is another matter, which can be addressed with KS variants)
[See http://en.wikipedia.org/wiki/Kolmogorov-Smirnov_test]

If you want a challenge (fun?), you might want to try adapt one of the condifence tests you
might already know about for video poker. A good place to start might be to take a look
at the chi-squared distribution. The chi-sqrd dist. looks a bit like a video poker PDF and it
converges nicely to a gaussian. You could find a correspondence between a chi-squared
distribution and the pdf of a certain # of hands of video poker, etc.

On the other hand, if you don't want to record every hand that is played, and only want to
look at a particualar hand-- you can find an easy test for that also-- but if you start with
an assumption (an assumption) then, well, the meaningfullness of your test is brought into
question (the ks test and its variants make NO assumptions given that you know the CDF).
BTW, assuming that an individual hand frequency follows a normal distribution is an
assumption itself, since, you don't know that is indeed the case (In other words, the
"gaffed" game could be gaffed in a way to alter the shape of the distribution, or just its
mean, etc). Hence its best to establish the distributions are the same (expected and
actual) first.

I guess if you wanted a real challenge you might want to first set out to prove that video
poker is really an ergodic (or at least wide sense stationary) randomn process (when its not
gaffed!) using real data. Then go on to condsider what the data from a gaffed game might
do to your analysis. LOL.

Watch our for people who recommend that you "fetch rocks".

Edmund,?

[snip>

It is true that video poker as a?whole is decidedly non normal and is better represented

by a Poisson distribution.? This is due to some highly infrequent events ( RF in?particular).?
An individual hand type should have a more normal distribution.

I don't recall seeing information on sample size for Poisson distributions.?

This is why the problem is interesting.? Running the numbers is the easy part.? Figuring

out which numbers to run and what to do with the results is the challenging part.

···

--- In vpFREE@yahoogroups.com, greeklandjohnny@... wrote:

#4

cdfsrule wrote:

All this is pretty amusing to me. Given that the cumulative
distribution function for all (well almost all) video poker games is
known, you can (trivially) construct any confidence test you want ...
Watch our for people who recommend that you "fetch rocks".

You're surprisingly late to the party ... being in your back yard, I'd
looked for you sooner.

I feel like you tossed a fast ball -- I missed your take on whether
you find it feasible for a player to make a call on machine
"fairness". For the sake of example, say a player wanted comfort that
the machine they play hasn't been gaffed to reduced ER by more than 3%
(reasoning that a casino wouldn't bother for anything less).

Within the practical resources of a single player, is it possible to
construct a test for this with a given confidence? Suggestions for
best means, if so?

Your "fetch rocks" slid right past me too. Don't pick me for team
practice today.

I may be way off base here ... I usually have you pegged as someone
who shares the strength of their knowledge to help others better
understand a concept. This post smacks of simply sharing that
cdfsrule "better understands".

- H.

#5

You aren't calculating things correctly. You get trip aces 1/650 hands, fill
it 1/23.5 hands, so you fill every 15,275 hands. So to see it happen 10 times
is 152,750 hands. So at 700 hands/hour you're looking at 218 hours of play.

Edmund

···

On Nov 9, 2007 8:43 AM, <greeklandjohnny@aol.com> wrote:

Edmund Hack wrote :

?*************************************************************

?So, by your theory, you would need 235 trip aces, or on average about
580,500
hands. At 700 hands/hour, you are talking about 829 hours, or almost 21
weeks of
full time play.

In other words, it's nearly impossible to tell.
**************************************************************

Edmund,? I was thinking more of 10 times 23.5 ( success of?filling in
trips) ?times 650 ( occurrence of trip aces) or about 15,000 hands.?

#6

I'd recomend you start with the so called "K-S" test

(kolmogorov-smirnov) which give

statistically meaningful results when you know the cdf (what you do

with those results is

up to you). The K-S test is (so) easy to apply to for video-poker

(though not partiucalrly

sensitive at the tails, but that is another matter, which can be

addressed with KS variants)

[See http://en.wikipedia.org/wiki/Kolmogorov-Smirnov_test]

Dang! Pardon the interruption of this thread from a rookie, who tries
to play perfect strategy, but is still in the learning mode....but
this stuff takes out all the fun for me!! LOL

Kurt

···

--- In vpFREE@yahoogroups.com, "cdfsrule" <vpfree_digests@...> wrote:

#7

Dang! Pardon the interruption of this thread from a rookie, who tries
to play perfect strategy, but is still in the learning mode....but
this stuff takes out all the fun for me!! LOL

Kurt

Thus the beauty of the delete key.

#8

The delete key? No way. I hope I wasn't taken as complaining about the
thread, as I am enjoying it. As a recreational player, I read
everything on this list, hoping to gain some insight to make my play
worthwhile.

I am just amazed at the depth of mathematical knowledge that some put
into this.

Kurt

···

--- In vpFREE@yahoogroups.com, "ozymandias77096" <ugwf@...> wrote:

>
> Dang! Pardon the interruption of this thread from a rookie, who tries
> to play perfect strategy, but is still in the learning mode....but
> this stuff takes out all the fun for me!! LOL
>
> Kurt
>
Thus the beauty of the delete key.

#9

1) Your example. The 3% ER change is not enough info to answer the question. Imagine
that is acheived by changing the payout for common hands OR for uncommon hands...
What makes up the "test" for the ER change would depends on the choice-- hence it
depends on the distribution-- assuming of coarse that the game continues to play the
same (and always plays the same). [If you must, as for a test that will always work, just
record every hand for a while, and appply the KS test or one of its variants. As a check, you
might want to try if for a non-gaffed machine first. You may find out quickly enough that
your ugaffed machine is likely gaffed. C'est la vie]

2) I've played on a couple of gaffed machines-- gaffed both for and against the player.
Some had been rigged simply to output too much money (coins) sometimes when you win
(or get a payout at all). You could have monitored the hand freuquencies all day (all week,
lol) and never detected the gaff. You had to get the machine to payout coins. I've played
on other machines that had a two-way gaff on. One that hurt the player and one that
helped (in the player new the secret). The funny thing about both these games is that
although they didnt' last long but people kept playing them (as if they were un-gaffed). I
played on other machines that supposeddly were gaffed in the way that is being discussed
here-- but...

3) It's like fetching rocks. [being told to fecth rocks, then when you do get some, being
told to fetch some more; the only thing you get is tired and a pile of rocks]. It's not really
about the sexiness of the math. There's probably a conservation principle of sorts acting
here: The time a "gaffed" machine exists is inversly proportional to its effect. Likewise,
the smaller the effect, the more data you will need. And of course add into the "secret
knowledge; inside job" issue, adn the fact that your test subject-- the machine-- is
probably not staying the same-- we'll you can get the picture. There's a certain amount of
luck involved in finding that gaffed machine....

For the sake of example, say a player wanted comfort that

···

--- In vpFREE@yahoogroups.com, "Harry Porter" <harry.porter@...> wrote:

the machine they play hasn't been gaffed to reduced ER by more than 3%
(reasoning that a casino wouldn't bother for anything less).

Within the practical resources of a single player, is it possible to
construct a test for this with a given confidence? Suggestions for
best means, if so?

Your "fetch rocks" slid right past me too. Don't pick me for team
practice today.