Paper: Four shuffles is enough

FLASH1296

Well-Known Member
#2
random shuffling <?>

Four shuffles is adequate? Maybe it is.

Do they take into consideration that most good experienced dealer's riffles are (virtually) perfect;
that is to say that they intertwine the cards as single cards and not as random clumps of 1, 2, 3, or more cards?
 

Sonny

Well-Known Member
#3
johndoe said:
Of course, they didn't consider any AP techniques in this analysis...
That’s true. Just because something is randomly distributed doesn’t mean that it isn’t predictable. That's where the money is.

-Sonny-
 
#4
FLASH1296 said:
Four shuffles is adequate? Maybe it is.

Do they take into consideration that most good experienced dealer's riffles are (virtually) perfect;
that is to say that they intertwine the cards as single cards and not as random clumps of 1, 2, 3, or more cards?
A perfect riffle is trackable. A really bad riffle is trackable. The least trackable is when a dealer riffles really well or really poorly, at random.
 

bjcount

Well-Known Member
#5
Automatic Monkey said:
A perfect riffle is trackable. A really bad riffle is trackable. The least trackable is when a dealer riffles really well or really poorly, at random.

A bit of sensitive material? :rolleyes:
 
#6
bjcount said:
A bit of sensitive material? :rolleyes:
Sonny will bust it if so. But I don't think so- in order for the casino to exploit that info they'd have to tell the dealer to kind of just do whatever he wants for the shuffle procedure. There is actually an AC casino that does shuffle like that! The dealer splits the shoe into 4 parts and the dealer grabs from them at random. But I doubt that's really the procedure, probably just lax enforcement of the procedure.
 

Guynoire

Well-Known Member
#7
Sonny said:
Just because something is randomly distributed doesn’t mean that it isn’t predictable.
Yes it does, if a deck of cards is completely randomly distributed then all 52! permutations are equally likely. You can't predict the order of the cards then.
 

alienated

Well-Known Member
#9
Different concern

I only took a quick glance at the link, but I think the concern in the paper is with how much shuffling is required to ensure "good as random" results for computer simulations or other computer applications that require random features. For blackjack, obviously the shuffle requirements for "good as random" results depend on what you want to simulate. I think the mathematicians have found that for situations where "suit doesn't matter" (e.g. basic strategy, counting, or any strategy for which suit is irrelevant), four shuffles is sufficient to give "good as random" results.

I think they are concerned with a wider implication that extends far beyond blackjack. Namely, when a problem does not require all aspects of randomness to be satisfied, the randomizing procedure can be simplified and shortened (less computer time expended).

So applying the idea to blackjack, if you want to analyze basic strategy, the paper indicates that you only need four shuffles, because not all aspects of randomness need to be satisfied (e.g. suit doesn't matter). However, a strategy for which additional aspects of randomness were required (e.g. suit does matter) would require more shuffles.
 

Sonny

Well-Known Member
#10
Guynoire said:
…if a deck of cards is completely randomly distributed then all 52! permutations are equally likely. You can't predict the order of the cards then.
Why not? It doesn’t matter that each outcome is equally likely if you can recognize which outcome just happened. Just because the average outcome is randomly distributed doesn’t mean that each individual iteration is not predictable.

For example, the top card of the deck should be an ace (or any rank for that matter) 1/13th of the time. If each rank is randomly distributed then it will be an ace 1/13th of the time, but we might still be able to predict when that happens and when it doesn’t. If the dealer flashes the top card after every shuffle then we will know exactly what rank it is every time, but the distribution (the average number of times the top card is any given rank) can still be completely random. Shuffle tracking, sequencing, glimpsing and reading marks are other ways to predict the final outcome of a shuffle even though it will, on average, be randomly distributed.

As another example, I’ll flip a fair coin several times. Each outcome is equally likely on every flip, but if I watch closely I might be able to determine which side it just landed on. Each flip is completely random and the average result at the end of the experiment will be 50/50 but I can still predict what the results are. There's a great story about how Ken Uston fell for that very scam. He lost about $9,400 to a guy in AC who could predict coin flips to some degree. The guy didn't influence the coin to be biased so the results were completely random, but he still got an advantage over Uston.

-Sonny-
 
Top