Neural Net Lottery Picker

A lot of questions

So I choosed The basic settings 09 -03 1300 by commodity (empirism if you like)

Then I remarked that the F have a significant impact on the drawings if i choose between 150 and 50

So then i thought that if I could find the best back testing with a good "latest used" I will therefore amplify it with the F settings( or if you perfer I hoped for it).

8 is due by the fact that on keno the average to get the full spectrum is between 7 and 9 draws so 8 is in the middle .

Then aside  the skipped results you get the indication  "match" if I get at least 3 in one of the 8 line changing the "use latest" cursor I then test it with the backtesting until I achieve the best of it.

pretty simple a some test (on the paper) gave me some good results on 5/6 draws , I was thinking that there can be some use of this strategy.....and here we are to improve it .

It would be nice if someone somewhere could try these settings on an other Keno.

Again 2x 3/4 tonight (10 euros)

Congratulations to you stoopendaal Great master piece.

Today fail but 70 Euros for these 6 trials

Some interesting  "back testing" by tonight with latest 63 (same rules for the rest)

For the 6 next (with 5 top)

F = 7 = 07 08 26 33 56

F = 9 = 38 26 56 07 37

F =11 = 26 07 37 54 21

F = 13 = 26 07 33 53 54

F = 15 = 48 54 26 21 34

F =17 = 26 67 38 21 48

F = 19 = 38 06 39 26  07

let's see

Hi Stoopendaal,

Got it !! thanks a lot !

Gambler667

Here In Ontario A 3 Of 4 Pays 0$

You Need 3 Of 3 ($25 on $1 Bet or 4/4 On $1 Bet For $100)

So ? What happened with the simulation ?

I thought nobody cares ...

So results from the last trial

The 6 last draws

05,09,10,17,22,23,26,31,34,36,41,44,47,50,56,59,60,61,65,70
05,06,18,19,22,25,26,31,36,37,38,39,46,48,54,63,64,66,67,70
04,05,07,12,18,19,27,28,35,40,42,48,49,58,59,60,61,62,63,64
07,08,24,25,27,33,40,41,45,48,49,56,57,60,62,64,65,66,67,68
03,11,13,21,22,25,27,28,32,36,43,44,45,46,48,50,51,57,65,69
03,04,06,11,12,23,25,28,30,32,33,37,40,43,44,49,59,61,63,65

As you can see at least 4-5 time 4/5 (10 Euros as 5/5 are paid 100) and in France we get 2 Euros for 3/5

No 5/5 but as you can see it worth to play .

I've done some simulation for Keno France 20/70 with the settings below (changing F from 7 to 9, 11, 13, 15, 17, 19 )

I think these are good results, giving 2-3 times better than theory. I will test some more to find if this is only pure luck or is something else.

YES !! good to see your back up.

I think we have set the good stategy for keno the F+++ is the key (I may say the best actually .I hope some will be inspired to try on other keno as well)

But as I told you before I have no clue for the latest as I have a very good back up as well with a lower figure (51)

And it's good as well to see that you come as well with a maximum of 6 draws (you have uploaded new sets each time I guess)

Now it could be interesting to calculate the return if it worth to play 4 or 5 or 6 numbers (yesterday an other 4/4....)

I like you excel file (mine is in collumn )if you don't mind to share (i am lazy isn't it?), by the way how many draws do you use , me I use the fdj backup in full (cvs file).

I 'd like to see up to 25 the average of each F setup (10 lines from 7) I will push as well to 7 numbers for a while

Thanks you

Hi winnerlotto,

I see you have put a lot of effort testing the application .

In the latest version of the tool  (1.5) you have the option to adjust the number of neurons in the hidden layer (with a maximum of 20 neurons).

Perceplotron version 1.5 ANN2 (64 bit version):  http://intelbet.somee.com/tools/perceplotron_64.rar or http://intelbet.somee.com/tools/perceplotron_64.exe

Perseplotron version 1.5 ANN2 (32 bit version): http://intelbet.somee.com/tools/perceplotron_32.rar or http://intelbet.somee.com/tools/perceplotron_32.exe

Good luck!

Hello stoopendaal,

Many thanks for this fantastic tool you give us.

I found the F+ strategy by accident, after a lot of trials,unfortunatly not as good as I wished for the loto but just keno 70/20 in France.

what could be the aim (advantage)  to get more hidden neurons at your point of view .?

Hi jlg69,

"what could be the aim (advantage)  to get more hidden neurons at your point of view .?"  There a different rules-of-thumb to determine the number of neurons in the hidden layer to optimize the neural network for your data (in our case to get better predictions )

Here some rules you can test to find the best number of neurons in the hidden layer:

• 1

In sum, for most problems, one could probably get decent performance (even without a second optimization step) by setting the hidden layer configuration using just two rules: (i) number of hidden layers equals one; and (ii) the number of neurons in that layer is the mean of the neurons in the input and output layers.

The ANN in perceplotron has 2 input neurons (and 1 hidden layer) and 1 output neuron. So the number of neurons in the hidden layer

should be (2+1) / 2  = 2 (when round up)  according this rule.

• 2

One additional rule of thumb for supervised learning networks, the upperbound on the number of hidden neurons that won't result in over-fitting is:

Ni = number of input neurons.
No = number of output neurons.
Ns = number of samples in training data set.
alpha = an arbitrary scaling factor usually 2-10.

Others recommend setting alpha to a value between 5 and 10, but I find a value of 2 will often work without overfitting. As explained by this excellent NN Design text, you want to limit the number of free parameters in your model (its degree or number of nonzero weights) to a small portion of the degrees of freedom in your data. The degrees of freedom in your data is the number samples * degrees of freedom (dimensions) in each sample or Ns∗(Ni+No) (assuming they're all independent). So alpha is a way to indicate how general you want your model to be, or how much you want to prevent overfitting.

For an automated procedure you'd start with an alpha of 2 (twice as many degrees of freedom in your training data as your model) and work your way up to 10 if the error for training data is significantly smaller than for the cross-validation data set.

For example you train the ANN with the (latest) 80 draw results (= samples in training data set)

For this training set the number of neurons in your hidden layer should be :  80 / (2*(2+1))  = 80/6 = 13.333 = 13  (for alpha = 2 in this example)

source:  http://stats.stackexchange.com/questions/181/how-to-choose-the-number-of-hidden-layers-and-nodes-in-a-feedforward-neural-netw

Good luck!

Hi stoopendal

I have done some tests with your new version, and I may say sorry but it does not work.

I don't get the same set than the previous one even with hidden neurones with 2 and keep repeating almost the same set.

with the F+ I can just have a 2/4 and more, if I go over up to 20 it does not change anything

it seems that only 1 and 2 are working and 1 gives better backup.

By exemple :

Old version

F 7 , Use Lat 89 , HN 2

gives : 46 7 32 57

New one (same settings, same file)

It gives 7 27 57 6 HN 2

And 7 27 62 46 HN1  over it does not change anything 7 27 57 4  HN3 5 7 10 ....20

but more the F + keep giving the same numbers all the time without changing the NN output.

Sorry

6 Next  draws with latest = 83 by tomorrow mid. (previous version of Percep-Lotron)

Good night.

Thanks. Testing the new function now.