United States Member #121769 January 16, 2012 13 Posts Offline

Posted: January 16, 2012, 7:41 pm - IP Logged

Hello,

I've written a simple program to take the numbers for pick 3 games and try to predict the next number. It reads the number and then breaks that number down into the first, second, and third number. It uses an artificial neural network to read the first, second, or third number and then outputs a number for its respective position. It does not combine stats like if 2 is the first and 4 happens nine times in the second with 2 while 5 happens 15 times.

What I do have it do is to use the learning rate as the observed probability minus the expected probability. I derived the expected probability from what I noticed that the counts of the sums from 0 to 27 is a bell curve and that 96% of all possible winning combinations would occur from a sum of 4 to 23. Within these sums I've noticed that the numbers that compose these sums is also another curve. It is not as near as normal as the sums curve, but it is normal when compared to the observed curve.

I am curious to know whether you all think this is a normal output. I am getting probabilities of predicting the right number for the first, second, and third anywhere from <1% to >7%. It varies because of the random weights used to initialize the network.

Does anyone here have any interest in this program or experience with AI programming for lottery predictions? I would think to run it about 1000 times and see if it comes up with anything over 10%, which would be working in my experience.

What ideas do you all think I should add to increase outcomes?

San Diego, CA United States Member #61467 May 24, 2008 28146 Posts Offline

Posted: January 16, 2012, 7:47 pm - IP Logged

Quote: Originally posted by integralsum on January 16, 2012

Hello,

I've written a simple program to take the numbers for pick 3 games and try to predict the next number. It reads the number and then breaks that number down into the first, second, and third number. It uses an artificial neural network to read the first, second, or third number and then outputs a number for its respective position. It does not combine stats like if 2 is the first and 4 happens nine times in the second with 2 while 5 happens 15 times.

What I do have it do is to use the learning rate as the observed probability minus the expected probability. I derived the expected probability from what I noticed that the counts of the sums from 0 to 27 is a bell curve and that 96% of all possible winning combinations would occur from a sum of 4 to 23. Within these sums I've noticed that the numbers that compose these sums is also another curve. It is not as near as normal as the sums curve, but it is normal when compared to the observed curve.

I am curious to know whether you all think this is a normal output. I am getting probabilities of predicting the right number for the first, second, and third anywhere from <1% to >7%. It varies because of the random weights used to initialize the network.

Does anyone here have any interest in this program or experience with AI programming for lottery predictions? I would think to run it about 1000 times and see if it comes up with anything over 10%, which would be working in my experience.

What ideas do you all think I should add to increase outcomes?

I haven't read all of your post, but what you need to do is to start posting your predictions on the Prediction Board. Todd went to a LOT of work to design that and it works great. It is the proof of how well your system/program/method works. When people start making huge claims (not talking about you!), then when they are asked to prove it by posting on the Prediction Board, they come up with a lot of excuses...I don't know how...I don't have the time...etc. But if you do that, you and the rest of us will be able to see if your program works well enough to make money.

United States Member #121769 January 16, 2012 13 Posts Offline

Posted: January 16, 2012, 8:00 pm - IP Logged

Quote: Originally posted by Jack-C on January 16, 2012

I haven't read all of your post, but what you need to do is to start posting your predictions on the Prediction Board. Todd went to a LOT of work to design that and it works great. It is the proof of how well your system/program/method works. When people start making huge claims (not talking about you!), then when they are asked to prove it by posting on the Prediction Board, they come up with a lot of excuses...I don't know how...I don't have the time...etc. But if you do that, you and the rest of us will be able to see if your program works well enough to make money.

Good luck.

Thank you for your input.

I am not saying that it is winning. In fact, that is what my post is about: it doesn't win more than standard odds. I am getting a prediction rate for one number between <1% and >7% but that is still far below the 10% that one would think would be correct.

I would be happy to post predictions for people, but it took me about two days (10+ hours) to read and write all of the numbers into a text document that I could read. If people would like to volunteer and provide a list of numbers for their state, I would be more than happy to accomodate them with a predicting number.

----------

Another problem is that I do not update the bias like I should for the network. This means that sometimes it won't output any number but a wildcard 'x'.

San Diego, CA United States Member #61467 May 24, 2008 28146 Posts Offline

Posted: January 16, 2012, 8:18 pm - IP Logged

Quote: Originally posted by integralsum on January 16, 2012

Thank you for your input.

I am not saying that it is winning. In fact, that is what my post is about: it doesn't win more than standard odds. I am getting a prediction rate for one number between <1% and >7% but that is still far below the 10% that one would think would be correct.

I would be happy to post predictions for people, but it took me about two days (10+ hours) to read and write all of the numbers into a text document that I could read. If people would like to volunteer and provide a list of numbers for their state, I would be more than happy to accomodate them with a predicting number.

----------

Another problem is that I do not update the bias like I should for the network. This means that sometimes it won't output any number but a wildcard 'x'.

OK. Here are the draws for the last 12 months for GA Midday:

United States Member #121769 January 16, 2012 13 Posts Offline

Posted: January 17, 2012, 3:17 am - IP Logged

Edit: In regards to Jack-C's post.

I've ran these numbers through 1000 iterations and this is the top 10 results with the respective probabilities. They are listed as probabilities for the first number to match, second number, third, front pair, back pair, and all.

United States Member #121769 January 16, 2012 13 Posts Offline

Posted: January 17, 2012, 3:28 am - IP Logged

If there are any others out there who would like me to generate some numbers for them, I will. I just need you to take your numbers and put them in a .txt file with each number on its own line.

Like this:

123

456

789

000

Do not double space. This keeps the time it takes for me to a minimum of just running the data. Jack-C was kind enough to provide 310 numbers from GA's midday Pick 3 lotto. I've 2090 for another lotto. 300 data points seems about right and I can iterate my program 1000 times in about 20 minutes with 300. 2000 has not shown better results yet and takes about 3 hours to complete 1000 cycles.

Sometimes I am seeing results, and sometimes it actually can predict the winning number. If you would like me to make a pick 4/5 generator too, please show your interest. I am also working on a lotto 5+1 generator, mostly concentrating on predicting the 5 rather than the +1 at the moment.

San Diego, CA United States Member #61467 May 24, 2008 28146 Posts Offline

Posted: January 17, 2012, 4:57 pm - IP Logged

Quote: Originally posted by integralsum on January 17, 2012

Edit: In regards to Jack-C's post.

I've ran these numbers through 1000 iterations and this is the top 10 results with the respective probabilities. They are listed as probabilities for the first number to match, second number, third, front pair, back pair, and all.

United States Member #121673 January 14, 2012 26 Posts Offline

Posted: January 18, 2012, 12:18 am - IP Logged

Quote: Originally posted by integralsum on January 16, 2012

Hello,

I've written a simple program to take the numbers for pick 3 games and try to predict the next number. It reads the number and then breaks that number down into the first, second, and third number. It uses an artificial neural network to read the first, second, or third number and then outputs a number for its respective position. It does not combine stats like if 2 is the first and 4 happens nine times in the second with 2 while 5 happens 15 times.

What I do have it do is to use the learning rate as the observed probability minus the expected probability. I derived the expected probability from what I noticed that the counts of the sums from 0 to 27 is a bell curve and that 96% of all possible winning combinations would occur from a sum of 4 to 23. Within these sums I've noticed that the numbers that compose these sums is also another curve. It is not as near as normal as the sums curve, but it is normal when compared to the observed curve.

I am curious to know whether you all think this is a normal output. I am getting probabilities of predicting the right number for the first, second, and third anywhere from <1% to >7%. It varies because of the random weights used to initialize the network.

Does anyone here have any interest in this program or experience with AI programming for lottery predictions? I would think to run it about 1000 times and see if it comes up with anything over 10%, which would be working in my experience.

What ideas do you all think I should add to increase outcomes?

Why they say that your chances of getting one digit right on one straight position is 1/10 a person can try 1000 times or more and always fail or might get it before the tenth try.

If you tried to get 2 staight by posiiton digits right the chances of getting both of them right are said to be 1/100, but again a person might take a lot longer or might never get it right or might get it before the one hundreth try.

If a person tries to get 3 straight by position digits the chances of getting them right might be 1/1000, but just as a person might get them right before that, it might take longer or might never get it.

Failure compounds the more factors that a person tries to get right.

For example:

Single or Double.

Low Sum

Medium Sum

High Sum

Low Last Digit of the Sum

Medium Last Digit of the Sum

High Last Digit of the Sum

Low Root of the Sum

Medium Root of the Sum

High Root of the Sum

Even

Or

Odd

Low

Or

High

In

Or

Out

Low- Low to High Width

Medium- Low to High Width

High- Low to High Width

Etc.

-------------------

It might be easier to try to get 5 digits for each of the 3 straight positions, then make a filtered wheel.

But even like so as 5 X 5 x 5 = 125 the so called random chances of getting the number right among the 125 might be 1/8 and lower chances than that if the 125 are filtered-reduced to fewer.

But maybe not impossible to do.

------------------------

It would be nice to be able to predict 1 straight number and win.

United States Member #121769 January 16, 2012 13 Posts Offline

Posted: January 18, 2012, 1:55 am - IP Logged

In regards to Laterne's post:

Thanks for your info. I've not considered all these different combinations as the AI is supposed to work all that out by itself. What I've done is figure out the ideal probabilities for the different combinations of numbers. I am surprised I've not seen this on this forum anywhere else.

What I have is this

Sums : Permutations

0,27 : 1

1,26 : 3

2,25 : 6

3,24 : 10

4,23 : 15

5,22 : 21

6,21 : 28

7,20 : 36

8,19 : 45

9,18 : 55

10,17 : 63

11,16 : 69

12,15 : 73

13,14 : 75

So with these, it would behoove a player to use numbers that add up to 13 or 14 or between 10 and 17. If you used numbers that add between 10 and 17, 56% of most of the combinations will come from that range.

United States Member #121673 January 14, 2012 26 Posts Offline

Posted: January 18, 2012, 5:24 am - IP Logged

Quote: Originally posted by integralsum on January 18, 2012

In regards to Laterne's post:

Thanks for your info. I've not considered all these different combinations as the AI is supposed to work all that out by itself. What I've done is figure out the ideal probabilities for the different combinations of numbers. I am surprised I've not seen this on this forum anywhere else.

What I have is this

Sums : Permutations

0,27 : 1

1,26 : 3

2,25 : 6

3,24 : 10

4,23 : 15

5,22 : 21

6,21 : 28

7,20 : 36

8,19 : 45

9,18 : 55

10,17 : 63

11,16 : 69

12,15 : 73

13,14 : 75

So with these, it would behoove a player to use numbers that add up to 13 or 14 or between 10 and 17. If you used numbers that add between 10 and 17, 56% of most of the combinations will come from that range.

Dump Water Florida United States Member #380 June 5, 2002 3134 Posts Offline

Posted: January 18, 2012, 6:46 am - IP Logged

Sum is not a valid reason to exclude any pick-3 number group because the hits are being ranked based on population not sum, 0-1-2 hits just as often as any other number.

United States Member #121769 January 16, 2012 13 Posts Offline

Posted: January 18, 2012, 10:15 pm - IP Logged

Quote: Originally posted by BobP on January 18, 2012

Sum is not a valid reason to exclude any pick-3 number group because the hits are being ranked based on population not sum, 0-1-2 hits just as often as any other number.

BobP

I understand what you are saying, BobP; however, I am under the impression that it would be more beneficial to pick a number that falls within a group with a higher count of permutations than to try and pick the winner with absolute perfection. I am willing to lose a few rounds in order to win, than just aim to win only to lose in the end.

I will keep toying with this idea and see where it leads me.