RNG Analysis Software

[Mike]

Steve, I don't know much about your systems and methods, but I know that you have some kind of pattern analysing software; I assume you've tried running RNG spins through it and haven't found any patterns that can be exploited in the way that wheels can?

What IS predictable from an RNG? I would say nothing, only the long-term stats predicted by probability, but they don't help.
Then again, the way that the spins are generated must have some affect on the patterns, so to me it doesn't make sense to crack a 'generic' RNG, just as in the same way you wouldn't expect to see the same types of bias on all wheels (regardless of type or manufacturer).

[bombus]

The absolute toughest thing in roulette for an RNG to accomplish is to get through 37 numbers without a repeat.

I believe this makes repeating numbers the most likely focal point for this endeavour.

[Steve]

Yes I may be wasting my time. I said that, but also said I'm still going to proceed. Why? Because what I'm doing has never been done before. It will be a form of AI analysis.

[chippy]

Hi Steve,i dont know if this is the sort of statistic you would like to look further into. Spin the wheel until you get 18 unique numbers,on average this takes about 24 spins. 6 repeaters. The other half of the board which hasnt hit is exactly the same as any ec not hitting for 24 spins ie 24 reds hitting on the trot,therefore the blacks(remaining 18 nos) are due to start hitting. Hope you can find something in this to help.

[schoenpoetser]

Bombus what do you mean with"37 numbers without a repeat"?
Is it   12 26 16 16 33  or 12 26 16 33 2 16 35 ??
The 1/3 rule  tells us there are about 12 numbers repeat in a row of 36 spins.
Every row of 37 spins is very uniquec because it occurs once in 37^37 spins.

Chippi what is an unique number.All 37 numbers are unique.The wheel or the RNG has no preference

[bombus]

Bombus what do you mean with"37 numbers without a repeat"?

I mean RNGs don't have the intelligence to put out 37 numbers (going from 0 to 36 in any order) before repeating one of the numbers already put out.

[bombus]

Here you go.


I got this from a Really Nice Guy I know.



6x6 number groups:

1)  11.14.15.31.32.36

2)  4.8.9.19.22.30

3)  2.10.18.21.23.29

4)  5.7.17.24.25.28

5)  6.12.16.33.34.35

6)  1.3.13.20.26.27


Record the numbers until,

Any two consecutive hits in group 1) = bet YO, 1 unit, 1 time.

Any two consecutive hits in group 2) = bet RO, 1 unit, 1 time.

Any two consecutive hits in group 3) = bet RP, 1 unit, 1 time.

Any two consecutive hits in group 4) = bet BP, 1 unit, 1 time.

Any two consecutive hits in group 5) = bet BG, 1 unit, 1 time.

Any two consecutive hits in group 6) = bet YG, 1 unit, 1 time.

After each win, keep betting until the next number out leaves the trigger group. Then wait for two consecutive hits in one group before betting again.

0 = wildcard for previous group hit, but never for next group hit.



Betting groups:

YO=  4.8.9.11.13.14.15.19.20.22.26.30.31.32.36

RO=  2.4.8.9.10.11.15.18.19.21.22.23.29.30.31

RP=  2.4.5.7.8.10.17.18.21.22.23.24.25.28.29

BP=   2.5.6.7.10.12.16.17.24.25.28.29.33.34.35

BG=   1.3.6.12.13.16.17.20.24.26.27.28.33.34.35

YG=   1.3.6.11.13.14.15.20.26.27.31.32.33.35.36

[bombus]

Here's a typical 500 spin RNG demolition session...

[bombus]

Here's the summary.

[bombus]

So what is this doing?

Well, it's getting your program to hunt for preset conditions then associating those conditions back to predetermined number societies. The idea being that the RNG has no idea what you're doing, which is fair because you have no idea what the RNG is doing.

This example's particular setup uses the colour wheel down to secondary colours. They are placed in order around the wheel then grouped according to colours. This could be correlated to numbers instead of colours. Instead of Y YO O RO R RP P BP B BG G YG you could have 1 12 2 23 3 34 4 45 5 56 6 61.

Around the wheel you say?... Yes, why not?

The RNG has no idea what a wheel is, or in what order the numbers are placed around the wheel, which is fair because you have no idea in what order the RNG will make the numbers arrive.

Maybe you could make a program that randomly distributes the numbers around the wheel after every result, and then reassigns the number societies for each spin according to the new wheel arrangement. 

[Steve]

I'll give an example of one of the functions. It'll generate say 1m spins. The rng will be done over a few times like a so called "true rng" does. Then it will take a sequence of numbers and check for occurences of what sequence over the 1m spins. then over another set of 1m spins etc. Then it will compare the number of spins between those occurences, and check for possible patterns between groupings.

A large part of it will check to see if there is any predictabiliuty to CLUMPING. For example, over 37 spins, about a third will be repeats. Statistically, of course eventually all 37 numbers will be generated in 37 spins. But I'll test that too, over billions of rng spins. Will it all match up to what we expect? I expect it will.

But still is there any order in it? There always is, because nothing is ever random. So its more a question of can the order be figured out to the point where we can increase accuracy of predictions. Thats what I'm trying to figure out.

It is a lot more involved than looking for sequences though. We can even try 3D charting to check for patterns. Imagine forming a 3D chart that we can scroll around and check. Maybe theres something a 2D chart wouldnt show, but 3D would.

If there is to be a chance of success, NEW things need to be thought of.

[bombus]

Don't put the cart before the horse Steve. These RNGs are re-seeding all the time; the algorithms are perpetually altered, so LONG distance forecasting is probably defunct.

Having said that it can only go "36 degrees" so at best you will probably just end up with the most expensive randomizer in history that just breaks even or as schoenposter says... the result is predictable, -2.7%. 

[mr.ore]

It probably won't work, but I believe you can "fit" your AI to win over millions of known spins. For a bet with payout x:1, you should test it over x*1e6 spins,  and maybe it is not enough. You want an AI to do the trick for you? Do it simple - program back propagation network (nolinks://galaxy.agh.edu.pl/~vlsi/AI/backp_t_en/backprop.html), then add some more outputs to them and store them in a variable, which is an input each spin - they are a memory for it. You want to make a set of those variables to be always same - round 0.3 to 0, 0.9 to 1 and make a lot of them. You can also make some like 0,0.25,0.5,0.75,1 by rounding them. Because of this, neural network has a state and feedback - it has a simple "consciousness". Then train it over those spins to see if it works. If there is something, it might find it. Or train it with a genetic algorithm and use backpropagation to just converge faster, backpropagation itself will stuck. I think that this is probably waste of time - too much hustle, you should know why your algorithm works or not. You should also ignore zero in your testing, or do both with and without zero. Make it a filter - measure z-score over original spins and over bets recommended by AI. If it is never higher than some value, you might have something. You said you want to do something like SETI - are you going to use BOINC?

[schoenpoetser]

Bombus the system you describe in reply 21 doesnot belong in this thread.I am busy to analyse the system and later on I`ll give you my remarks.
Back to the thread.
In my opinion are the results of every random row inside small deviations predictable.If some patterns will repeat with a algorithm the sequence is not random.
I take in my strategy relative small samples with specific features from a very large sequence.All small sequences form together all so a new sequence.

Steve there is no mathematically  difference in a random row of numbers and a row of H/l.Better study a bi sequence.
In my opinion you are looking for a Holy Grail.

[bombus]

You're probably right schoey, I only put it here as food for thought.

[bombus]

-