Welcome Guest
Log In | Register )
You last visited January 21, 2017, 12:28 pm
All times shown are
Eastern Time (GMT-5:00)

Neural Networks- Applicable to Lottery?

Topic closed. 32 replies. Last post 13 years ago by Andrew.

Page 2 of 3
PrintE-mailLink

Do you think that Neural Nets can help?

Yes (Why?) [ 5 ]  [25.00%]
No (Why?) [ 5 ]  [25.00%]
Not Sure [ 3 ]  [15.00%]
What's a Neural Network? [ 7 ]  [35.00%]
Total Valid Votes [ 20 ]  
Discarded Votes [ 3 ]  
Avatar
New Member
Tulsa, Oklahoma
United States
Member #2939
December 3, 2003
2 Posts
Offline
Posted: March 7, 2004, 9:01 am - IP Logged

 

Thank You, hypersonic; for the info, on neural networks.  I agree, a form of pattern recognition; is needed for the lottery analysis.  And, your suggestion of training the synaptic self, is probably a needed addition; to help in working with the data.

    I work with, usually two forms of system analysis; due to one system cylcing - hot / cold tendency.  Observing similarity, between the systems, can provide a grouping, which may provide better results.  So, some form of varying the data, may allow a greater successs possibility.  jimm

    RJOh's avatar - chipmunk
    mid-Ohio
    United States
    Member #9
    March 24, 2001
    19901 Posts
    Offline
    Posted: March 7, 2004, 11:09 am - IP Logged

    My question is:

    If anyone can detect certain reoccurring patterns with data mining software, why won't states running lotteries use similar software to make sure their systems were truly random?  I would think if they found such patterns, they would introduce other factors randomly to make the process truly random and the only constants would be the field and combinations sizes.

    RJOh

     * you don't need to buy more tickets, just buy a winning ticket * 
       
                 Evil Looking       

      Avatar
      NYC
      United States
      Member #2988
      December 10, 2003
      237 Posts
      Offline
      Posted: March 7, 2004, 12:35 pm - IP Logged

      Can this help with wetware neural nets? (IE, your own brain skills)



      Program link



      (Remember, I am mainly interested in the Pick 3 games.)

      I've found it necessary to add this signature line: My posting is not an invitation to send me a private message.

        hypersoniq's avatar - 8ball
        Pennsylvania
        United States
        Member #1340
        April 6, 2003
        2450 Posts
        Offline
        Posted: March 7, 2004, 12:58 pm - IP Logged
        Quote: Originally posted by RJOh on March 07, 2004



        My question is:

        If anyone can detect certain reoccurring patterns with data mining software, why won't states running lotteries use similar software to make sure their systems were truly random?  I would think if they found such patterns, they would introduce other factors randomly to make the process truly random and the only constants would be the field and combinations sizes.

        RJOh




        Maybe you are giving the states too much credit... They have a seemingly endless revenue stream, money for nothing... If anyone could detect recurring patterns I don't see them catching on... If EVERYONE could, then that might set off some alarm bells.

        Playing more than one ticket per game is betting against yourself.

          hypersoniq's avatar - 8ball
          Pennsylvania
          United States
          Member #1340
          April 6, 2003
          2450 Posts
          Offline
          Posted: March 7, 2004, 1:15 pm - IP Logged
          Quote: Originally posted by Millennium on March 07, 2004


          Can this help with wetware neural nets? (IE, your own brain skills)

          Program link

          (Remember, I am mainly interested in the Pick 3 games.)




          I can't see it helping unless you document your reasoning at the time for each number selected, then go back and see what you were thinking when successful. Much of what we do with lottery numbers is usually scrapped when it doesn't work... I think better recordkeeping might help to yield some insights later on.

          Some common things that we should probably all do is document attempts at system creation better, this might help as some sort of ad-hoc protocol to follow, we ARE experimenting after all...

          A. Clearly define your goal at the beginning, including putting into words what it is you will be using as a system to pick numbers (or eliminate numbers)

          B. When you first set up the system, take copious notes of even seemingly insignificant details.

          C. When the system produces a pick, keep track of it (or them) and note the difference between that and the actual draw result, perhaps here some sort of common error can be discovered.

          D. Using notes from step B, formalize a data sheet with pertinent information to keep from each attempt with the system pre-draw and each result post-draw.

          E. Keep up with it, periodically review all data and jot down anything that seems interesting.

          I think if we run these lotto system experiments more like a Laboratory experiment, we may walk away with useable data rather than just frustration.

          Playing more than one ticket per game is betting against yourself.

            BobP's avatar - bobp avatar.png
            Dump Water Florida
            United States
            Member #380
            June 5, 2002
            3113 Posts
            Offline
            Posted: March 7, 2004, 7:53 pm - IP Logged

            You can have a little fun watching a Neural Net at work here . . .

            http://www.20q.net/

            BobP

              hypersoniq's avatar - 8ball
              Pennsylvania
              United States
              Member #1340
              April 6, 2003
              2450 Posts
              Offline
              Posted: March 8, 2004, 11:50 am - IP Logged

              Interesting stuff so far... Still not exactly sure of what is needed, but apparently it's not an Autoassociative NN... In those, if the pattern presented is recognised, it just echoes the pattern back.

              I have read some interesting intro articles on 2 closely related A.I. methodologies... Genetic Algorithms and Simulated Annealing (both built into JOONE BTW)... they seem to be better than back propagation because they avoid the local minima trap.

              This won't be an over-nighter... Combining weight matrices, error correction, hyperbolic tangents, sigmoids, hidden layers...

              My arsenal of digital tools...

              Visual Studio.NET (and the .NET port of JOONE)

              Sun Java 2 SDK (and the Java implementation of JOONE)

              Mathworks MatLab 6.5 R13 (with the Neural Network Toolbox)

              MS Office XP & Office 2000

              I would be surprised if I have playable results by this time next year...

              Anyone else joining in on the fun?

              Playing more than one ticket per game is betting against yourself.

                hypersoniq's avatar - 8ball
                Pennsylvania
                United States
                Member #1340
                April 6, 2003
                2450 Posts
                Offline
                Posted: March 8, 2004, 12:18 pm - IP Logged

                perhaps the best order of things is a neural net used for classification (probabilistic, based on bayesian classifiers) and then on to a predictive network... It would appear that choosing the wrong type of network for the task ensures failure, also, the problem itself needs to be clearly stated for any chance at a solution...

                Playing more than one ticket per game is betting against yourself.

                  LottoBuddy's avatar - tails

                  Canada
                  Member #3722
                  February 15, 2004
                  46 Posts
                  Offline
                  Posted: March 8, 2004, 11:09 pm - IP Logged

                  No system can predict a random lottery draw.  Read http://www.camh.net/egambling/issue2/research :

                  The human mind is not very good at dealing with randomness. Our minds are designed to find order, not to appreciate chaos. Ever notice how easy it is to find faces in clouds? We are wired to look for patterns and find connections, and when we find patterns we interpret them as real. Consequently, many people will see patterns in random numbers. When people see patterns in randomness (e.g., repeated numbers, apparent sequences or winning streaks) they may believe that the numbers arent truly random, and therefore, can be predicted.

                  Many gamblers have experienced a wave-like roller coaster effect of wins and losses and may believe that you just have to ride out the down slope of the wave to follow the wave back up. Much of this learning process takes place unconsciously. The problem is that betting based on these patterns sometimes appears to work in the short term, reinforcing the belief. But it will not work in the long term; these patterns are flukes.

                    Avatar
                    NYC
                    United States
                    Member #2988
                    December 10, 2003
                    237 Posts
                    Offline
                    Posted: March 9, 2004, 12:34 pm - IP Logged
                    Quote: Originally posted by LottoBuddy on March 09, 2004

                    No system can predict a random lottery draw.








                    And yet I have predicted winning numbers in the Pick 3. So have others here.

                    I've found it necessary to add this signature line: My posting is not an invitation to send me a private message.


                      China
                      Member #3032
                      December 16, 2003
                      1081 Posts
                      Offline
                      Posted: March 10, 2004, 6:56 am - IP Logged

                      No system can predict a random lottery draw so far

                       

                        hypersoniq's avatar - 8ball
                        Pennsylvania
                        United States
                        Member #1340
                        April 6, 2003
                        2450 Posts
                        Offline
                        Posted: March 10, 2004, 9:32 am - IP Logged

                        From the state of the poll results so far...

                        Many views, few votes, fewer replies...

                        Apparently nobody is interested in pursuing this in an open-source sort of way, so I'll not continue further along that track.

                        Final thoughts...

                        Attrasoft demo... no good because you can't see the code. There will be no turn-key solution. Interestingly tho, the layout took the 5 white balls as input and the powerballs as targets, regardless of how I set it up... it had an 18% "hit" rate, but when it was off, it was way off.

                        On development... I plan to keep accurate records every step of the way, including a capture of each weight during each training iteration, current NN freeware misses the ability to document each step, time to pry the lid off of the "black box". When I did my random synch experiments it generated massive amounts of data, simple code to poll the variables at each level of code execution that is significant.

                        I'm not bothering to work anything for the pick 3 or any other game, just going after the powerball (5 white balls)... If that can show promise, the 1/42 is easy to beat by throwing enough $$$ at it...

                        Playing more than one ticket per game is betting against yourself.

                          LottoBuddy's avatar - tails

                          Canada
                          Member #3722
                          February 15, 2004
                          46 Posts
                          Offline
                          Posted: March 11, 2004, 4:02 pm - IP Logged

                          hypersoniq, will your neural network be based on the assumption that there are biases in the Powerball draws?  Will you record the order that the balls came out like some sites do?  Or are you assuming that independently drawn random numbers have some kind of magical repeating pattern that can be predicted??

                            hypersoniq's avatar - 8ball
                            Pennsylvania
                            United States
                            Member #1340
                            April 6, 2003
                            2450 Posts
                            Offline
                            Posted: March 11, 2004, 4:47 pm - IP Logged

                            no magic at all, I'm leaning towards bias.... look at the 27 as powerball (or lack of it in over 149 drawings)...the number of repeats on separating pb data by wednesdays and saturdays, isolating 10,000,000 drawings from higher jackpots... how the number 20 follows the 35 as pb 5 to 10 drawings later...

                            I have abandoned Quick picks, in favor of one ticket of "personal lucky" numbers and one ticket based on spreadsheet analysis. $4 per week... IF I manage to get results from this new approach, it will only be one more line to add, and one of the others can go... I'm comfortable with $4 per week. No other system apparently works, but many generate scores of numbers and then blame you if you can't pick ones to play out of it...

                            The numbers are just a hobby, keeps the math skills sharp, paid off today... I got a new job as a machinist, been laid off since december... woohoo ;-)

                            My ultimate goal in any system is to generate one pick, win or lose... prefer to win but will be happy missing out on the lower tiers in exchange for buying far fewer tix...

                            As for data, I keep it both ways (drawn and sorted) and plan on training several networks with each set... trial and error...

                            As for randomness... The 5 white balls are not independent, they come from the same set. the results are not truly random, they are as "random as humanly possible", yet results remain as bounded integers... They will never draw a number lower than one and barring a matrix change, never draw one over 53. The powerball is independent of the white balls, so I will not add that info into the NN, perhaps a secondary NN just for the PB would be more fitting.

                            Will it work? who knows? nothing else has so far, but the process of investigation and experimentation is fun in its own right, and a really cheap hobby...

                            Playing more than one ticket per game is betting against yourself.

                              hypersoniq's avatar - 8ball
                              Pennsylvania
                              United States
                              Member #1340
                              April 6, 2003
                              2450 Posts
                              Offline
                              Posted: March 12, 2004, 6:03 pm - IP Logged

                              classification is one difficult task to get any info from.... I may end up doing this by hand, composing a weight matrix based on same-draw AND next draw announcer frequency.

                              Put simply, If a 5 drawn in the first position has been followed in the past by 6,7,9,12,9,3  I will give a heavier weight to a 9 because it happened twice...

                              Also I will look "out-of-position" because positional announcers are not reliable enough, but if, for instance, a 1 in the first position is often followed by a 41 in the 4th position, there is one relationship to work with....

                              Before I try to code a NN, maybe I'll try a protracted process of trying classification with my own snappin' synapses....

                              Playing more than one ticket per game is betting against yourself.