STUDY GROUP FOR PRALINES HOLY GRAIL

[schrodinger]

Fine work !
I am focusing on Al as I have already programmed multiple approaches using neural network.

I am actually using multi step architecture for the  generator, first a monte carlo ( hasting for example, look at my github for coin example) to simulate the roulette given its random numbers ( real spins) , with that I get a posterior distribution. then this distrib will be used for the next step when using a BNN bayesian network   link:s://:.cs.toronto.edu/~rgrosse/courses/csc421_2019/slides/lec19.pdf   and tune in such way that my posterior makes it stay in the 95% confidence interval and more setting.. its not complete yet!

[6th-sense]

could you just ask if u dnt understand what we are doing ?  what do u mean by not practical ?! we are dealing with last 180 spins witch just a copy paste from ur casino table history or fill them by hand to the program while u waiting ur play, its not much . then predicting instantly the next 3  in live session or ( test session as im doing now, testing)

Ok I’ll ask..Im talking about a b&m casino..
Not online...
For the online part you are testing when it’s gets to to the generated numbers without the last 6...is it not then possible to go for the prediction then when that’s done ..drop the furthest away same  amount  Of numbers you used and then predict again..
Sort of a stepped rolling basis so to speak..
If it works properly it’ll be just like starting as if you collected results from that new point?
Have you tried this way?
Could your algorithm..neural network accomplish this adjustment?

[schrodinger]

For the online part you are testing when it’s gets to to the generated numbers without the last 6...is it not then possible to go for the prediction then when that’s done ..drop the furthest away same  amount  Of numbers you used and then predict again..
Sort of a stepped rolling basis so to speak..
If it works properly it’ll be just like starting as if you collected results from that new point?
Have you tried this way?
Could your algorithm..neural network accomplish this adjustment?

it doesnt matter if its online or offline,  suppose u collected the first 180 spins, then the next number comes,  you just add it to the data and get the last 180 and u have 180 spins again. (u add 1 and remove 1 so to speak, u dnt have to "remove" in python , just get the last 180)  like u said rolling. and we are testing  next 3 not next 6 ! although u can set it to whatever number u want,  read the last page   and  its a threaded program so in 3 spins you already have a new trained network for the next prediction.  Hope u get it !   u can play it online seamlessly.     

[6th-sense]

Makes sense ..look forward to the end results

[ati]

"guess who is coming to dinner!"

Hehe, but it's "educated guessing"

[ᶦ ᵃᵐ|Ä-łëx]

7
4
0

[schrodinger]

7
4
0

1 number right 

[Tinsoldiers]

1 number right 

Good luck chap. Keenly following this.

[Clf7]

Schrodinger could you explain us on what it is based? Because only on AI it will be difficult to beat roulette, machine learning cant beat the game...

[schrodinger]

Good luck chap. Keenly following this.

cheers mate !

[schrodinger]

Schrodinger could you explain us on what it is based? Because only on AI it will be difficult to beat roulette, machine learning cant beat the game...

the general idea is explained little bit on the top of this page, by taking the Bayesian approach! I based my intuition on the fact that roulette is a physical system and not totally random in the common sense, which can be represented by set of variables. lets investigate some of them:
- we have speed of the wheel and the ball , rotational variables , and lets note the relative distribution diff :  θ
- we have scatters a translation  variables and lets note them by the distribution :  ρ
they can be normally distributed or whatever we choose Î"θ∼N(0,σ1) , Î"ρ ∼N(0,σ2)
from there we can use a Monte Carlo algorithm (refer to this example in my github to have an idea)  to get the convoluted posterior distribution of  θ  and  ρ , which will generate/simulate the roulette.
then using  a Bayesian neural network BNN to get our prediction generated same way as the roulette itself with some uncertainty interval.. the rest is training and tuning.  Hope this helps
cheers

[Clf7]

Thanks   

[ᶦ ᵃᵐ|Ä-łëx]

2
29
1
20
1
30
7
30
33
29
34
28
7
36
36
36
25
16
4
31
31
26
18
34
17
5
27
24
29
10
25
33
9
15
29
29
19
9
3
9
16
24
26
27
25
5
15
3
26
32
25
18
31
23
12
31
5
1
3
0
36
26
18
23
22
10
24
11
6
13
13
11
13
26
1
0
32
29
3
19
34
24
28
21
4
11
26
15
25
11
16
17
9
17
14
17
19
23
24
13
18
16
8
1
17
6
8
25
31
24
28
2
5
22
30
18
34
7
17
10
24
6
4
24
30
1
7
16
9
18
33
13
24
18
15
9
31
35
1
12
13
32
16
11
26
23
32
32
18
23
28
4
33
28
29
21
15
29
16
14
35
5
0
16
28
18
2
19
35
22
26
36
5
8
22
36
24
31
19
28

[schrodinger]

35
22
26
36
5
8
22
36
24
31
19
28

next  : 12 , 0 , 6/36

[ᶦ ᵃᵐ|Ä-łëx]

33
29
33
33