Author Topic: Data Collection for Machine Learning  (Read 1556 times)

0 Members and 1 Guest are viewing this topic.

vitorwally

Data Collection for Machine Learning
« on: April 03, 2018, 08:12:04 PM »
This week I managed to briefly get in touch with a friend of mine whose field he's involved had him working with Machine Learning. Complementarily, a few weeks ago, I've got myself "travelling" within the paradigma of creating a statistical predictive model apply-able to roulette. Ones might find it feasible, others the opposite. That's not the topic here. Unsurprisingly, to create a predictive model for roulette I have most likely to enroll in Machine Learning as well as it is the most used tool in the moment to design predictive models. Unfortunately this might turn out a very time-consuming effort as I'll have to gather a large database of spins and detail them as much as possible to make easier the task of the software to identify patterns. The thing is, I have a couple of questions. Here they go:
  • Which should be the "nature" of the spins?
  • From B&M casinos or from live streaming gambling studios?
  • Dealer driven wheels or airballs?
  • May RNG be considered as well or should it be excluded? If included is a random sequence generated in Excel as valid as a list of numbers taken off a RNG roulette from an online casino?
  • Do I want the predictive model to be specific and only input data from one of the above examples or it's better to have a more in-depth "piece of equipment"?
If you're getting your hands ready to start bashing me please have in mind the intentions I have with this predictive model. Of course I believe in randomness. I've seen myself "weird" outcomes, I'm not a newcomer to roulette. What I ask you is to don't forget that large data usually don't lie. There are certain behaviors that we all have observed during our game-play. A bunch of you guys have use them to strengthen your ways of playing. Some guys call them red flags. You can name them whatever you want. The point is, I don't want this to be some kind of wizardry that would deliver the next spun number correctly time after time. If this alone manages to get a near to profit hit rate picking groups of numbers, I would say between 18 and 24, we can do the rest with money management. At the very least, I'm creating this topic with educational purposes, as we all learn new things everyday.
 
The following users thanked this post: Reyth, MrPerfect., MickyP

MrPerfect.

Re: Data Collection for Machine Learning
« Reply #1 on: April 03, 2018, 08:43:57 PM »
Victor... in your particular case " large data " will lie.
  If you come to the point that mashine will collect ball and rotor timings from video , let me know..  lm more then curios.
 

vitorwally

Re: Data Collection for Machine Learning
« Reply #2 on: April 03, 2018, 09:35:08 PM »
@MrPerfect.
I kind of suspect where you're trying to lead me. The undeniable fact that randomness will always have something to say and in some way make a fool out of the predictive model. I can't deny it. But let me make you a question. As an AP, do you think you consider all the variables that could affect the outcome of a spin? I don't want to sound like a preacher of the butterfly effect and neither call you ignorant, but the more you focus on some variables there might be something, somewhere, that could affect your prediction. I'm not regarding the degree of sureness of your picks. Allegedly they are top notch. I believe your words. What I want to bring up is that even your selections might be affected at some degree by randomness. So, I don't discard the power of Machine Learning just because there's unpredictability within the sphere of influence we are regarding. Artificial intelligence has given us the ability to build predictive systems with unprecedented accuracy. Machine learning is being used in virtually all areas in one way or another, due to its extreme effectiveness. In addition, if in any case, randomness succeed to make a fool out of our predictive model, there's Bayesian modeling to help us "teach" our child and let him know of this new situation. Machine learning isn't just a mere observation of the frequency of the pockets. Tending to infinite, the frequency is equal to each pocket. It is an universal truth. Large data would fail if the analysis has in that frequency the only focus. However, are the permutations of the spins evenly distributed? That's where Machine Learning may perhaps have a word.
« Last Edit: April 03, 2018, 09:40:52 PM by vitorwally »
 
The following users thanked this post: MrPerfect.

MrPerfect.

Re: Data Collection for Machine Learning
« Reply #3 on: April 03, 2018, 11:30:11 PM »
Victor, there are simple things about roulette...
 Results mean anything only in set of conditions they are taken. If conditions setts are mixed- results are mess, but it will never show expected averages due to overlaps of results on some numbers/ zones.. ets.
   I will give you simple example... you got some idea where to bet, be it yourself or with your " child". ... they change ball ... and all info you got there become outdated. It doesn't mean that data become zero value, it's still valuable with other ball ( if/ when they put it back in game).
   Or other one... you got data on some particular set of conditions when some  variable take some determined values...
   Now when you start playing, they see you wining and start to change variables they got control ... that particular variable getting new values that are different from before...  your actions??
   In both cases, if you follow what was before, result is more likely to be - loss.
   So what you should be focusing on ? Make creepy skynet or understand how to beat the wheel?
   Your mashine will be as intelegent as it's programming, if you can not give it ability of math modeling of physical systems, results will be...  hm... how do you think?
   Especially considering that you apriory wanna give your mashine wrong algorithm. ... there are no perfect wheel, l studied many, yet no one of them was perfect... so no reasonable fitting of reality in your math abstraction of it. Some numbers always go ahead... qwestion is if you can determine wich numbers and when or can not.
   Besides... you probably will not bother to imput data to this mashine manually,  wouldn't you? So , when you figure out how to make it collect full profile of variables( or at least ball and rotor timings) , please let me know. It's something lm missing as a tool for my studies. It takes so much time....
 

vitorwally

Re: Data Collection for Machine Learning
« Reply #4 on: April 04, 2018, 07:24:33 AM »
@MrPerfect.
As it seems you answered one of the questions I had:
Do I want the predictive model to be specific and only input data from one of the above examples or it's better to have a more in-depth "piece of equipment"?
In your words, results mean anything only in the set of conditions they are taken. If sets with different conditions are mixed, results are a mess, they will never show expected averages due to overlaps of results on some numbers/zones. It makes sense to me, can´t reject that. Yet, it sounds as all the work done over the years with the CLT (Central Limit Theorem) of the Normal Distribution could be flush down the toilet. I believe you have that judgement due to the high rigor you input into your game-play, you don’t afford to lose even a 0.01% edge. You want your picks to have max accuracy.
The approach with the modeling can float within the range of numbers you want to wager in. I might want to have meticulous predictions of single numbers or I might want to guess the next coming dozen. I think this is the point where we should take a moment before starting collecting data. For more narrowed picks it would be required to have a more restricted dataset.
I strongly accept as true the control of physics over an outcome of roulette and stepping up the game would be as you say being able to feed our dataset with the visual ballistic variables of each spin.
 

MrPerfect.

Re: Data Collection for Machine Learning
« Reply #5 on: April 04, 2018, 12:45:30 PM »
https://www.roulettelife.com/index.php?topic=1750.0
    Victor, look the topic l posted for people who wanted to see what is possible and what is not.
   If you wanna create mashine to predict outside bets, how to put it in nice words..... you probably should be looking for more specific targets. No need to make super robot to automatically perform your mistakes,  if this thing may work, no need to limit it for bets with negative expectation.
   About precision of my edge estimations... it's around +/-10% from what l expect it to be on majority of situations.
    On my personal opinion , such a mashine would be nice to have all info avaliable to operate with, don't you agree?
 Key variables that may affect game already have been listed all over the place in my posts some even have pictures with examples...  hope it helps to give you some ideas you could use...
« Last Edit: April 04, 2018, 12:53:24 PM by MrPerfect. »
 

Reyth

Re: Data Collection for Machine Learning
« Reply #6 on: April 04, 2018, 01:38:40 PM »
My reasearch clearly shows that RNG will produce bias and adhere to certain expected characteristics, all on its own, without any physical cause necessary.

Can a programmed data collection and analysis help us to interpret this information over a very large sample (millions of spins)?  Most definitely!

The problem as it has been described to me and as I have understood it, is properly analyzing the data in order to draw our conclusions.

My goal with Machine Learning was to predict the onset/ending of hot streaks.  So far I haven't put together even any completed flowcharts.
 

MrPerfect.

Re: Data Collection for Machine Learning
« Reply #7 on: April 04, 2018, 03:09:03 PM »
That's a clear example of what lm tolking about!!!
  If you lost something in dark room, there is no point to look it somewhere else just because lights there is better.
  Wrong tasks offer wrong solutions.
 

Janusz

Re: Data Collection for Machine Learning
« Reply #8 on: April 04, 2018, 03:20:10 PM »
All know the arguments against it. Nobody knows anything to help.

Only selfish vain people here. Especially the AP-Maniacs.

To wally:

Datas collected from anywhere will predict the outcomes from where?

In another forum we collect datas only from 1 Table and wanna transfer the results to the sam table.
I think this is very important.
It considers that there might be a pattern in the same dwell.
 
The following users thanked this post: Reyth

Mike

Re: Data Collection for Machine Learning
« Reply #9 on: April 04, 2018, 07:32:30 PM »
Quote
Only selfish vain people here. Especially the AP-Maniacs.

Jeez, all you ever do is whine.  ::)
 
The following users thanked this post: MrPerfect.

Reyth

Re: Data Collection for Machine Learning
« Reply #10 on: April 04, 2018, 08:00:55 PM »
I am not being a critic but the ONLY basis for taking stats from one TABLE OR RNG SESSION to another is personal permanance; i.e. each random sequence has its own unique characteristics that will affect the random flow.  Changing tables or RNG sessions is like beginning a new session.

Maybe we can't see the causes of certain events because we don't have the visual/cognitive scope that a computer can emulate -- MAYBE this can work to teach us things that we otherwise simply would not see or imagine could exist.
 

vitorwally

Re: Data Collection for Machine Learning
« Reply #11 on: April 04, 2018, 09:23:24 PM »
@MrPerfect.
Thank you for replying and leaving a link to a utter useful topic. I got another look on the kind of detail I should get on my dataset. I understand what drives you in roulette, you're never satisfied. Convenience is a highway to failure. I recognize why you choose the AP path, even in roulette data analysis. I need to reject outside bets in favor of harder to predict straight bets because If we work out our way to successful modeling the success has to be represented as well on outside bets. Furthermore, I recognize how advantageous it would be to increase the detail of our data with the physics side of the problem.

@Janusz
What you are asking me is kind of embed in the questions I wrote at the beginning of the topic. I'm new to roulette data collection on a context different from everyday statistical or probability analysis. Machine learning is a few steps above. It messes with statistical inference and prediction and those are new lands to me. MrPerfect. sort of directed me to his point of view, which I agree with. Have a read and make up your mind.

@Reyth
I decided to dip into Machine Learning exactly for what you said, for the sake of getting some level of abstractness. Frequently, one is too wrapped around a bunch of concepts and can't see meaningful aspects worth of reasoning.
Another thing you referred I was already expecting. The RNG bias. Quick question: MrPerfect. indicates that for "real life" wheels we should not discard an AP data analysis. What do you suggest for RNG roulette tables? There's no such thing as rotor speed or ball type on RNG.
 
The following users thanked this post: Reyth, MrPerfect.

Reyth

Re: Data Collection for Machine Learning
« Reply #12 on: April 05, 2018, 12:56:09 AM »
For RNG I can only recommend keeping records from EVERY spin in a session.  Spin hundreds and even thousands of times.  The Bias will always be present because it MUST be.

If you change to a new session, the bias will automatically change.

Here is an example of me working with long-term RNG Bias:

https://www.youtube.com/watch?v=fJXFWmcMeCY
 

MrPerfect.

Re: Data Collection for Machine Learning
« Reply #13 on: April 05, 2018, 12:59:28 PM »
Technically speaking l realised this possibility long ego..  instead of advanced computer algorithm l use physics...  instead of pc learning l learn mysel...
   What can l say...  it sucks to match reality with expectations,  but it's better the way it is comparing with blind " pattern following" , even if backed up by advanced algoritm.