Roulette Outside the BOX

Very well stated.

Ken

 

The loss limit is very easy to calculate . No sweat .



ND

 

Wally, if you only know when to jump off by "intuition" which you can't explain in terms of "triggers" (perhaps because there are none, or they seem to change every time) then you're right - that can't be coded. But then if success depends on such feelings there's no point in having a system in the first place, at least not one that can be shared, because it would be totally subjective. And in that case forums would be redundant because what worked for you might not work for someone else.

 

Jerome, it would be difficult to argue with that. Some of us old timers may have learned a skill I consider to be a spin-off of emotional intelligence ... that ability to judiciously determine what’s in our best interest given any set of circumstances. How many times have members of this forum “had a feeling” things are just not quite right should they continue to play, regardless of whether up or down at the moment.

 

More results will likely shed light on the challenge with this variant of when to walk, and when to stay. It can be made mechanical, but as in this example, you'd leave money on the table if you played by a "first large profit, reset" rule as it was down, then recovered to +647u at spin 226. Good walk away point there, surely most would be happy with that session, especially how hard it was in the red....Except then it caught fire and went to +2113u by spin 246.

So the exit point for losses can be set at some hard cap, lets say -1500u if you have access to 0.25u or 0.50u play, but when it's winning it's extremely hard to figure out when it's about to turn for you or against you.

With the other version I've been testing it's easy, you know from the spin count how you're doing and if you need to eject or press, and you're right the majority of the time both ways.

With this one though I'd need more information than the dozen tests I've done. Will keep at it, it's interesting to say the least.

Might be a way to tame the Lion, but if not, the original version is currently sitting at 32 session wins, 5 session losses, with net units at +14577. Total of 4892 spins.

 

what maybe be an idea is the following;

When we got to a point where we have lets say 5 unhits left (no show yet) we bet those 5 from now on as a group and use a +1 l -1 progression until we are in profit and then start over. usualy between spin 70-100. just an idea.

 

Damn brothers, i've tried to hit the unhit that have hit. Next time i'll try to introduce this feeling you're talking about. We could bet maximum when we feel it and no play when we don't feel it. Let's make money! Whewww

 

Eddy, are you willing to share or provide a link to this?

 

Finally you're on the right track ...

 

Yeah bro, that sentence makes so much sense: Try to hit the unhit that have hit...

Hopefully he will, before we wonder why he never lost with something, but not use it anymore and searching for something else... Because it is a drag? But you don't lose, so why not play maximum of the table...

 

I've done a few tests using the data generated from the program for frodo's system (you can download it in post #157). From 100 sessions (which equates to about 3000 bets), the average bank for Frodo's bet selection was -11.91 and for the random selection it was -5.07. However, before you start thinking that betting randomly is better than Frodo's method, a means test showed that there was no significant difference (in other words, the difference can be attributed to chance).

Code:

Null hypothesis: Difference of means = 0

Sample 1:
n = 100, mean = -11.19, s.d. = 105.444
standard error of mean = 10.5444
95% confidence interval for mean: -32.1123 to 9.73232

Sample 2:
n = 100, mean = -5.07, s.d. = 92.7479
standard error of mean = 9.27479
95% confidence interval for mean: -23.4732 to 13.3332

Test statistic: t(194) = (-11.19 - -5.07)/14.043 = -0.435805
Two-tailed p-value = 0.6635
(one-tailed = 0.3317)

As usual, the number of interest is the "p-value", which you can think of as the probability that the result occurred by chance (a small p-value means that the result is "significant"). This shows the danger of jumping to conclusions based on intuition. I did another test comparing the means of the number of wins for both methods and it also gave no significant result.

However, things were different when it came to the variance of the wins for the two methods -

Code:

Null hypothesis: The population variances are equal
Sample 1:
n = 100, variance = 8.83424
Sample 2:
n = 100, variance = 29.9984
Test statistic: F(99, 99) = 3.39569
Two-tailed p-value = 3.804e-09
(one-tailed = 1.902e-09)

Sample 1 is the variance for Frodo's method and sample 2 is variance for betting randomly. The p-value is close to zero, meaning a highly significant result. You can clearly see the difference in volatility from a chart of the wins over the 100 sessions - the blue chart show the wins for Frodo's selection and the red chart shows the wins for random selection :



I must admit I was surprised at this result because I assumed that results would be pretty similar for variances as well as means.

 

Hi Jerome,

First off, thanks for the comprehensive analysis.
Let me see if i understood the results:
Bet selection:/numbers returning to the average EV over time:
If we could get the blue line horizontal, you would have just coded the HG?
The tested method stops at 185 in your data. Can we remake this test, only this time the cycles go to spin 740- for a sleeper can and will match that interval on a larger scale.
In theory, the variance should be more steady.
The problem is in my opinion the attack interval, one that we are trying to overcome with progressions.
Let's try to flat bet.
Your results can change the way we redesign the system.
Give me a system that FAILS on the long run/ moderate and i'll create a winner!

Can you test this?
740 spins
Flat bet
Wins graph versus random.

Regards,
John

 

I need explain here.
we are still playing the same way as i posted before, only now up till we have reached only 5 unhit numbers left. Because i found out that from this point on, the hitratio will drastic decrease.
so maybe, just maybe it's better that from this point on, when there are still 5 numbers unhit, we stop witj what werw are doin and focus us only from now on until we have reached a new high on the 5 numbers still unhit as a group betting.
this will lower our drawdown with a huge amount.
first tests are possitive. keep you all posted.

 

Problem here is, BM playtime.
I for one can't see me sitting at a table for more then 4 hours straight (240 spins). and this is also alot of time, including bathroom, smoke, eating breaks.

 

We should focus on getting our job done within this timefram.

BM
Max 4 hours playtime
Max 240 spins.
Min. profit of +100
Max. Loss of -500

 

I don't have a link anymore wally, is been quite some time. but you can search it up for yourself.
Palestis dozen.
was on the old roulette30 forum.

 

Making it horizontal would be impossible IMO, but maybe it could be flattened a little more. As for the HG, well not if HG means making a profit flat betting, because it doesn't according to my tests, at least in its current form. The significance of the reduced variance is that if it can be kept low enough then you may be able to use some kind of progression which stays within the house limits. Geez, I'm starting to sound like a system player

Yes, more bets results in reduced variance when taken over the entire session, but in general that won't result in reduced variance in the short term (ie over the current session length of 185 max). The way to reduce the variance is to come up with an even better bet selection. I've yet to code the latest version so we'll see how that pans out.

Anyway, I've just tested another bunch of sessions; 200 this time, just to make sure that the previous result wasn't a fluke. Here are the results :

Code:

Null hypothesis: The population variances are equal
Sample 1:
n = 200, variance = 10.7207
Sample 2:
n = 200, variance = 34.9706
Test statistic: F(199, 199) = 3.26197
Two-tailed p-value = 5.315e-16
(one-tailed = 2.657e-16)

Again a p-value of nearly zero, even smaller than the previous result. Here's the chart -


To be honest, I'm pretty blown away by this, I keep thinking I must have made a mistake. What it shows is that there seems to be something to systems after all. I'm still of the opinion that they can't win flat-betting in the long term, but they don't need to if you can reduce the swings enough to stay in the game long enough to make a profit.

 

Anytime you run a progression you will have larger variance. When testing to see if a system works it makes zero sense to test it with a progression.

Form a hypothesis...why should it work
Define test parameters.

Reduce the degrees of freedom as much as possible to increase statistical relevance and reduce testing size requirements.

Run a random control bet along side.

Run an out of sample test to replicate the first test. A three standard deviation result from a single test isn't all that relevant without an out of sample test with defined parameters.

Testing only hundreds of spins is silly and absurd for the system described.

 

By the way, the standard deviation for the bet selection win should grow as the spin sample grows.. if the results are real. Shoot for five standard deviations in out of sample testing.

 

Right, but there's no progression in this system - flat bets only.

That's exactly what I've done.

I've haven't tested only hundreds of spins, but hundreds of SESSIONS (each session being up to 37 spins/bets). Also, see my post #225 in relation to testing small sample sizes.