RNG Bias in Shuffle Machines?

I rejoice in the Wiz's Baccarat simulation data as I try to hone my R-code programming skills. And, many thanx to those who have shared their R-code, as I have found some of it very helpful. (But, more IS better. Hint, hint.)

Because the Wiz has carefully crafted many thousands of Baccarat hands for me to use in my research, I don't have to fuss with things like the Mersenne Twister or the Fisher -Yates Shuffle. But, I wonder... Does anyone know whether or not the random number generators in the various shuffle machines are biased because their shuffles do not employ Mersenne Twister routines and Fisher-Yates shuffles?

If shuffle machines don't use those techniques, their shuffles may not be truly random, right? I'm not sure how anyone might exploit that bias, but it would be nice to know whether my efforts are compromised because the shuffle machine isn't completely fair.

Quote: LuckyPhow

If shuffle machines don't use those techniques, their shuffles may not be truly random, right? I'm not sure how anyone might exploit that bias, but it would be nice to know whether my efforts are compromised because the shuffle machine isn't completely fair.

Look up the patents for your favorite automatic shuffler. No they are not the least bit "random." While they may use a great RNG (most do), it's the shuffling algorithm itself that is pretty crappy.

This guy, in particular, has done a lot of research on these machines:

https://en.wikipedia.org/wiki/Persi_Diaconis

Here is his most famous paper on the topic:

http://statweb.stanford.edu/~susan/papers/casino.pdf

If you don't want to read all the math, just read the conclusions in section 5.3:

Quote:

5.3. Conclusions and recommendations. The study above shows that a single iteration of a 10-shelf shuffler is not sufficiently random. The president of the company responded “We are not pleased with your conclusions, but we believe them and that’s what we hired you for.”

We suggested a simple alternative: use the machine twice. This results in a shuffle equivalent to a 200-shelf machine. Our mathematical analysis and further tests, not reported here, show that this is adequately random. Indeed, Table 1 shows, for total variation, this is equivalent to 8-to-9 ordinary riffle shuffles.

On the conspiratorial end of this debate, the answer is emphatically "no" - these machines are NOT designed to deal hands in a cheating manner to increase the house edge. Quite the opposite. There is a long history of advantage players reverse engineering these machines and beating the crap out of them.

teliot, would it be fair to say something like this --

Although the shuffling of the cards is not random (ie: a shuffle-tracker's definition of random), the results of the hands played are random.

In other words, it's only "not random" if you're trying to use the information of the shuffle. A shuffle-tracker in blackjack knows the cards are not random (because he has information by tracking the cards), but to a ploppy who has no information -- the cards are random to him.


?

Quote: RS

teliot, would it be fair to say something like this --

Although the shuffling of the cards is not random (ie: a shuffle-tracker's definition of random), the results of the hands played are random.

I don't know what you mean by "random" here. But, to the person whose strategy is perfect basic strategy at blackjack, his long term edge in blackjack against an ASM will be the house edge. In fact, ASM's are slightly more player-friendly than shoes, due to the absence of the cut-card effect.

The edge is the house edge for baccarat and every other game that uses such machines. In other words, these machines would pass a Turing test.

Quote: teliot

I don't know what you mean by "random" here. But, to the person whose strategy is perfect basic strategy at blackjack, his long term edge in blackjack against an ASM will be the house edge. In fact, ASM's are slightly more player-friendly than shoes, due to the absence of the cut-card effect.

The edge is the house edge for baccarat and every other game that uses such machines. In other words, these machines would pass a Turing test.

I'm saying there are essentially 2 different ways to define random, depending on what kind of information (or lack thereof) you have.

To a shuffle tracker, the shuffle is not random.
To a ploppy, the shuffle is random.

Wow! Thank you for such a prompt reply, Lucky sez (opening up a big can of worms...) It highlights several (somewhat related) issues.

Quote: teliot

Look up the patents for your favorite automatic shuffler.

OK. I'm not sure how easy this is to do, but I assume one can do that. Chapter 95 of your Advanced Advantage Play book describes the shuffling algorithm of an older (1997) shuffle machine (which I found most educational). The footnote (page 323) references the patent number for that shuffle machine. Ummm... Do you have a list you can share, perhaps?

Quote: teliot

No they [shuffle machines] are not the least bit "random." While they may use a great RNG (most do), it's the shuffling algorithm itself that is pretty crappy.

Let's talk about this some, OK? Could the (possible) existence of shuffler bias in casino-dealt Baccarat hands affect my confidence in data I generate from Wiz's Baccarat sims. Seems to me shuffle-machine bias might if shuffles do not employ Mersenne Twister and Fisher-Yates functionality, perhaps unintentionally skewing one result versus the other.

I appreciate your pointing me to Diaconis' paper (Analysis of Casino Shelf Shuffling Machines). I had read it previously (those parts that didn't go over my head, since I only had 2 graduate-level stat classes -- just enough to be dangerous, don'cher know?). My (incorrect) original impression was that this research project (testing the randomness of the shuffle of a new shuffle machine) was a one-off and not a concentrated research focus area of the authors. (I look forward to LOTS more exploration here!) It has a great overview of previous research and concludes -- as you note -- the machine needs to perform the equivalent of "8-to-9 ordinary riffle shuffles."

Here is where the rubber meets the road! Do "engineering" shortcuts such as those you describe result in bias (not necessarily favoring one side or the other, since the gambler can bet either Banker or Player) require me (perhaps) to have a more conservative error estimate in my results? (Ummm... Maybe this will help you understand where I am coming from: Suppose I am randomly going to pull sample groups of shoes from Wiz's data for repeated Bayesian analysis.)

Diaconis and others say one must shuffle the deck at least 7 riffles to get a reasonably random shuffle. If I understand correctly, this assumes we are starting with an ordered deck (AAAA, 2222, ... , KKKK). But, why can't the shuffle machine loop through the RNG process electronically multiple times (sorting 52 numbers randomly -- with or without Mersenne Twister & Fisher-Yates shuffle -- each time) before it dispatches the first card to its ultimate output tray?

If the shuffle machine runs the RNG one time, is that satisfactory? Apparently not. But why? With what possible effect on the deck I'm playing against when my (100%-sure-to-win <grin>) Baccarat system is used in a casino, as compared to my simulations using the Wiz data? Or, do they run the RNG multiple times? I dunno.

In so many discussions about how to program various bet-analysis systems, the Probability Pros on this site emphasize the importance of incorporating Mersenne Twister and Fisher-Yates routines, or else one cannot have confidence in the resulting data. OK, so I use Wiz's data (with that "stuff" built in), but the casino doesn't. My concern, simply, is this: Am I trying to hit a moving target that has an unknown, built-in bias?

One last thing. You say:

Quote: teliot

"There is a long history of advantage players reverse engineering these machines and beating the crap out of them."

My understanding is that most casinos lease shuffle machines (including ongoing maintenance). I have tried (unsuccessfully) to get documentation (such as user manuals, etc.) for shuffle machines. How does one go about "getting" a shuffle machine? On occasion I have seen technicians doing maintenance on shuffle machines, but that is as close as I have ever come to getting the "inside" view of a shuffle machine.

Again, many thanx for any info you can share.

LP,

Fwiw, they have classes. With all your interest in this, why not sign up at a vo-tech as a machine repair mechanic and learn how to fix them and how they work?

Quote: teliot

In fact, ASM's are slightly more player-friendly than shoes, due to the absence of the cut-card effect.

Do you mean CSMs? If not, can you explain further? There is still a cut card in the shoe when an ASM is used.

bb,

That is a great idea. I'd probably enjoy that class.

But, I just checked local vo-tech center. N/A in my area, as should probably be expected, living in the biggest city in south Georgia (or thereabouts).

Quote: LuckyPhow

bb,

That is a great idea. I'd probably enjoy that class.

But, I just checked local vo-tech center. N/A in my area, as should probably be expected, living in the biggest city in south Georgia (or thereabouts).

SHFL themselves offer classes in how to fix the machines. I was thinking more of votech in Vegas, perhaps AC or Biloxi for regional machines. I would think they teach those classes in places where there are lots of machines.

Quote: wudged

Do you mean CSMs? If not, can you explain further? There is still a cut card in the shoe when an ASM is used.

You are right, I mean CSMs.