Can you find "pattern recognition" in casino software?

[Peep]

Guy who beat Hero's for 35K claims you can.

Quote:
Originally Posted by Patrick McIrish
......This guy found some sort of pattern recognition that he could take advantage of, instead of being smart he went overboard IMO. The old saying you can sheer a sheep forever but you can only butcher him once probably applies here. I'm not a BJ expert but best I can tell he's broken no rules, probably going to end up getting paid if the place is legit. They all throw out the "audit" card before paying out decent sized scores in the casino, keep us posted if you can, thanks.

Pattern recognition is not so hard, especially if you get this: WLWLWLWL
Or this: LLLWLLLWLLLWLLL You just have to be willing to be wrong that the pattern will change but still do a progression bet. Many times I see this: LLLLWLWLLLL- two lone wins in the middle of a string of losses. After a while, I got a lot of those at Heroes and I just learned to anticipate them and be willing to bet big with expectation of the second one win. It is things like that that I use my brain for and I win on most of them. Also, within a session of 25 hands, you are bound to get a two-win streak or longer. So I do a progression bet anticipating the second-in-a-row win. Also, if you get a very low percentage win game, you could anticipate a high percentage game in the next two or three sessions and bet bigger (but not always). I had plenty of big losing games (when I lost my coolness!) but I come back eventually. Just pen and paper and my head.

Did I go overboard? Yes, admittedly I did but Heroes advertises $50K a roll in craps so I thought it was not a problem with them. They seemed to have a fair game of blackjack instead of feeling like the randomness was overridden on the bigger bets.

And, if so, would you guys consider this a legit win?

[Data]

[reno cool]

There probably are some sophisticated ways of cracking a random number generator used in some machines, but it sounds like this guy is talking about table games.
Anyway, this kind of simple pattern recognition is practiced by many slot players who are mostly hopeless.

[pico]

i think the only person that recognized a casino pattern is that computer programmer works for the gaming comission and figure out the seed in the random number generator used for the keno and got caught because his buddy was being stupid.

[curious]

I write sophisticated software for a living. There is no such thing as a "random" number generator. The best you can hope for is that the non randomness isn't too predictable.

Well, you CAN create an almost "random" number generator by doing very sophisticated things like sampling "random" things in nature such as star noise, etc. Or sampling keyboard strokes across lots of computers. There are various techniques for using data inputs for the random numbers, but these too will suffer from the fact that nothing is truly random.

And I can pretty much guarantee you that the writers of the gaming software don't use such sophisticated techniques because 1) most of them don't have the skill sets, 2) the sources of the "random" data usually cost money, and 3) building almost random number generators is very time consuming and expensive.

So, yes you can use pattern recognition on online gaming software and find patterns. The type of pattern recognition software you would need would be sophisticated and you would have to be skilled at using it.

Finding the seed used for an "off the shelf" random number generator would be fairly straightforward, most programmers tend to use the same code libraries and you can get the source of the random number generator and figure out an algorithm to reverse engineer the seed from a given string of numbers. Doesn't mean the seed wasn't changed at some point.

I have run statistical analysis on several online BJ programs and all of them cheated, usually when you had a big bet out. By cheated, I mean that the house won more often than the probabilities of blackjack would predict and the number of wins for larger bets was even more skewed in the house's favor.

[Peep]

I have run statistical analysis on several online BJ programs and all of them cheated, usually when you had a big bet out. By cheated, I mean that the house won more often than the probabilities of blackjack would predict and the number of wins for larger bets was even more skewed in the house's favor.

Wow.

That is really interesting Curious, thanks.

There is a guy in the sportsbook/industry section who just started a thread claiming the same thing.

[Ganchrow]

Well, you CAN create an almost "random" number generator by doing very sophisticated things like sampling "random" things in nature such as star noise, etc. Or sampling keyboard strokes across lots of computers. There are various techniques for using data inputs for the random numbers, but these too will suffer from the fact that nothing is truly random.

And I can pretty much guarantee you that the writers of the gaming software don't use such sophisticated techniques because 1) most of them don't have the skill sets, 2) the sources of the "random" data usually cost money, and 3) building almost random number generators is very time consuming and expensive.

Most if not all major casino software providers either require the purchase of truly random number generators or provide them to their clients. These typically utilize quantum events that are in fact "truly random".

Random.org provides truly random data over SSL for free based on atmospheric noise. John Walker's (no relation) HotBits website/toolkit, OTOH, pull truly random data (also over SSL) from phenomenon based on radioactive decay. (Not that I'd trust random data provided by unknown sources, mind you, but I'm sure you get my point.)

The Perl programming language, as a a further example, also provides a Math::TrulyRandom module (slow but effective) based on CPU timing discrepancies. Using a method such as this this to seed a high-period PRNG (such as the Mersenne twister algorithm, with a monstrous period of 2¹⁹⁹³⁷ − 1) to generate a relatively small quantity of data (say a couple hundred values) would be virtually as effective a solution as any.

I have run statistical analysis on several online BJ programs and all of them cheated, usually when you had a big bet out. By cheated, I mean that the house won more often than the probabilities of blackjack would predict and the number of wins for larger bets was even more skewed in the house's favor.

I'd be exceedingly anxious to examine such analysis as it relates to any major casino software provider, and if what you're saying proves statistically tenable I am very much in a position to do something about it.

[Ganchrow]

There is a guy in the sportsbook/industry section who just started a thread claiming the same thing.

Many posters have made such claims with respect to major software providers, and while I couldn't say that none are accurate, I can say that I've never seen any evidence whatsoever in support thereof.

There was of course that English Harbor issue (although hardly a a "major" software provider, btw) a few years back but my friend Mike Shackleford ("The Wizard of Odds") got to the bottom of it quite quickly.

[curious]

Most if not all major casino software providers either require the purchase of truly random number generators or provide them to their clients. These typically utilize quantum events that are in fact "truly random".

You know this for certain? I know this did not used to be the case.

I would doubt that one could prove that atmospheric background noise is truly random, or that radioactive decay is truly random. Has anyone (that you know of) did enough testing of this that it can be said that these two sources have been "Proven" to be truly random?

[Mr Handicapable]

I have run statistical analysis on several online BJ programs and all of them cheated, usually when you had a big bet out. By cheated, I mean that the house won more often than the probabilities of blackjack would predict and the number of wins for larger bets was even more skewed in the house's favor.

I totally agree!! I've goofed around w/Blackjack on Bodog for several years with maybe 40-50 bucks at a time and it will NEVER let you triple up and rarely double up!! I use perfect basic strategy with a progressive betting pattern like $4/4/8/8/16/16 and it nails me hard w/a losing streak EVERY time I get to that 5th bet!
Whats funny is that I can sit down 2-3x/week (no more than that) with $50 and run it up to $75 more often than I drop to $25! I'm up $400 in the last 3 months playing that way? There is something in their program which keeps you from ever making a run like you might in a casino and once I quit trying then it seems like I can take advantage of the teaser the program lets you win? Then again I could be wrong?

[Peep]

Back to the Hero's guy for a minute.

He starts with $400. Runs it up to $35,000.

What are the chances he could do that with the software being honest?

[Ganchrow]

I totally agree!! I've goofed around w/Blackjack on Bodog for several years with maybe 40-50 bucks at a time and it will NEVER let you triple up and rarely double up!! I use perfect basic strategy with a progressive betting pattern like $4/4/8/8/16/16 and it nails me hard w/a losing streak EVERY time I get to that 5th bet!
Whats funny is that I can sit down 2-3x/week (no more than that) with $50 and run it up to $75 more often than I drop to $25! I'm up $400 in the last 3 months playing that way? There is something in their program which keeps you from ever making a run like you might in a casino and once I quit trying then it seems like I can take advantage of the teaser the program lets you win? Then again I could be wrong?

With all due respect, anecdotal "evidence" of this sort hardly passes the test of statistical rigor.

[curious]

With all due respect, anecdotal "evidence" of this sort hardly passes the test of statistical rigor.

Ganch, there used to be a watchdog organization that gathered and analyzed this type of statistical information on online betting software to "prove" if the sw was honest or not. I cannot remember who they are. Is such a thing still around? The thing that always made me skeptical about the old group was that they got their data from the software vendors and the casinos. I think to do an honest test the data would have to be collected independently of them somehow using either screen scraping or bots or some other tactic.

[Ganchrow]

Back to the Hero's guy for a minute.

He starts with $400. Runs it up to $35,000.

What are the chances he could do that with the software being honest?

I'm not going to spend the time right now going through the Brownian calculations (as I did for a similar problem here), but it would be very much a function of his betting style and size.

No doubt that this represents an unlikely outcome, but given the number of online blackjack players, this seems well within the realm of believability.

Think about it this way ... were he to double his bet after each win, then ignoring blackjacks, doubling, and splitting (which drastc ally increase his probability of an outsize win at the expense of greater risk), then given the roughly 47.75% conditional probability of winning a blackjack hand he'd have a 47.75%⁷ ≈ 0.57% of turning $400 into $400 * 2⁷ = $51,200.

Lastly I'll point out that just because the player may have turned $400 into $35K from deposit to cashout, doesn't mean he didn't make earlier deposits.

So in other words, still assuming flat betting, the player may have in fact actually turned, say, 10 units into 87.5, a decidedly more likely outcome.

[Peep]

Thanks Ganchrow. 2000 to one is a close enough answer for me. Sounds like he did do outsized bets as well.

[Ganchrow]

Back to the Hero's guy for a minute.

He starts with $400. Runs it up to $35,000.

What are the chances he could do that with the software being honest?

OK, I ran a Monte Carlo simulation of this using the rules from Microgaming Vegas Strip:

Trials: 7,370,000
Win Prob: 0.9750%
Std. Dev. of Win Prob: 0.0036%
E(per session): -0.235
Std. Dev. per session: 8.729
Mean loss per losing session: -1.101
Mean win per winning session: 87.681

Assumptions:

• player always plays perfect composition-dependent basic strategy
• session terminates at the end of any hand if the player has either:
  • lost a total of 1+ units on the session or
  • won a total of 87.5+ units on the session
• player always bets 1 unit (except as follows below)
• player will never attempt to overshoot 87.5 units with his initial bet
• player will temporarily deposit more funds only as required to complete splits/doubles

[Peep]

Sorry Ganch, as usual, I got my facts wrong.

Here is what the player himself says took place.

To give exact numbers, I deposited $200 and won $34,724 in the end! Heroes Casino gives .3% rebates on the day's wagering on all games! One day I wagered somewhere around $165,000 and they gave me $495 back!

[Ganchrow]

Sorry Ganch, as usual, I got my facts wrong.

Here is what the player himself says took place.

OK, well I just stopped the simulation.

I'm not going to run it again based on the updated information as I think the point remains clear: Given the universe of online players, results of this magnitude are fully expected even without "online blackjack pattern recognition software".

[Peep]

Yes, agree it is well within the boundaries of possibility.

As we like to say in sportsbetting.... "shit happens".

[Wheell]

I have run statistical analysis on several online BJ programs and all of them cheated, usually when you had a big bet out. By cheated, I mean that the house won more often than the probabilities of blackjack would predict and the number of wins for larger bets was even more skewed in the house's favor.

Have you turned this analysis over to Ganchrow?

[curious]

Have you turned this analysis over to Ganchrow?

No

[Wheell]

Will you turn this analysis over to Ganchrow?

[Ganchrow]

You know this for certain? I know this did not used to be the case.

Yes. While it might not have been true in the past I do know it to be true now for the major casino software providers.

I would doubt that one could prove that atmospheric background noise is truly random, or that radioactive decay is truly random. Has anyone (that you know of) did enough testing of this that it can be said that these two sources have been "Proven" to be truly random?

Random.org provides a good discussion of "randomness" here. (The only part which I'd seriously question, although admittedly as at absolute best a physics dilettante, would be the concept of hard determinism as applied to quantum phenomenon.)

The problem is you can't take a given set of data and prove it to be random any more than you can prove the nonexistence of Russell's teapot. In fact for any given set of data one can always come up with an (uncountably) infinite number of deterministic algorithms that would produce that exact same data result.

Of course we'd potentially be able to disprove that any given algorithm were the correct one after the next item on the list, but even at that point we'd still be left with an (uncountably) infinite number of algorithms that would produce that exact (new) data set.

What one can do is perform statistical analysis on a (randomly selected) set of supposedly randomly generated data and determine the likelihood of it not being random based on statistical tests such frequency analysis or the degree of statistical independence between any two randomly selected nonoverlapping subsets. Nevertheless, plenty of pseudorandom generators (again I'll refer you to the Mersenne twister) will pass these tests as well.

The issue is that with knowledge of the actual seed and the gnerating process one could always produce the same pseudorandom series as well as accurately subsequent elements in that process. Just because one might not yet know the seed+process doesn't mean that such doesn't exist (and couldn't be determined at the worst of all possible times).

So this brings us back to Random.org's discussion of "randomness". We can either use a sufficiently chaotic process that has so complex a seed and algorithm as to render it virtually truly random, or we can take a quantum process (such as radioactive decay or the results of a two slit experiment) that produces (the "hard quantum determinism" mentioned on Random.org notwithstanding) truly random data.

Anyway, I'm hardly an expert in this field and I'd encourage anyone interested to check out the wealth of data and links on both the Hotbits and Random.org sites mentioned above.

[Ganchrow]

So just a quick thought as to how one might go about relatively easily and cheaply go about creating a truly random random number generator (albeit a very slow one).

Take a sufficient amount of a mildly radioactive material (such as, say, those old glow-in-the-dark radium watches, I say "mildly" so as not to seriously injure oneself) and place in a radiation and light shielded, temperature and humidity controlled box with a Geiger counter. Measure the number of "clicks" over two unit intervals of time.

If the first number of clicks is odd and the second even, then assign the '0' bit.
If the first number of clicks is even and the second odd, then assign the '1' bit.
If both numbers of clicks are odd or both number of clicks are even then repeat the experiment.

This, mechanical failures notwithstanding (and also I suppose neglecting any possible bias resulting from severely dwindling radioactive material supply), would then give you a single random bit. Repeat as necessary to determine a truly random sequence of desired bit length.

[Dark Horse]

What if true randomness doesn't exist, but only exists in man's mind and its limited capacity to recognize system?

Wouldn't it be relatively simple to set up a number of asymmetrical shapes, rotating against each other, that would create combinations well beyond man or machine's capacity to recognize underlying order?

[sqvirrel]

True randomness isn't necessary anyway. What is necessary is near randomness with an unknowable input. For example, one component could be server-side timing of packets out and in. Internet delay would make user-side manipulation impossible and delay itself is not predictable from packet to packer, even if the entire algorithm is known.

[MonkeyF0cker]

Agreed, sqvirrel. Acquiring the seed from third party or external sources would increase the security risk to cracking the RNG algorithm rather than insure it.

[Lets_Get_Money]

Online Casino Cheat ... Stay Away from them at all costs