Welcome Guest
Log In | Register )
You last visited December 4, 2016, 1:02 am
All times shown are
Eastern Time (GMT-5:00)

Lottery Prediction by Recurrence Analysis

Topic closed. 170 replies. Last post 11 years ago by Raven62.

Page 8 of 12
PrintE-mailLink
NemeSys's avatar - Nemesis
Harrogate
United Kingdom
Member #16589
June 1, 2005
107 Posts
Offline
Posted: September 4, 2005, 3:32 pm - IP Logged

The author of VRA himself! We're honoured. I'm still looking for related software to augment the analysis I have done so far. This list looks promising:

http://usti.net/home/search/od/dir.cgi?d=/Science/Math/Statistics/Software/Time_Series_Analysis/

"God is subtle but He is not malicious." - Albert Einstein

"In my experience there's no such thing as luck." - Alec Guinness, Star Wars Episode IV

    FrankieH's avatar - cat anm.gif

    United States
    Member #21258
    September 5, 2005
    267 Posts
    Offline
    Posted: September 5, 2005, 8:03 pm - IP Logged

    Hello Everyone,

    I'm new to L.P. and i'm happy to be here. I've just downloaded the Visual Recurrence Analysis and after following instructions, i found that it's taking a long time to generate numbers. I'm still waiting. Maybe it's not working. Is that normal? Any advice will be welcomed. Help a girl out. Smile

    Thanks in advance.

    FrankieH

     

      NemeSys's avatar - Nemesis
      Harrogate
      United Kingdom
      Member #16589
      June 1, 2005
      107 Posts
      Offline
      Posted: September 6, 2005, 2:46 am - IP Logged

      Hello and welcome FrankieH.

      Have you loaded a data file in first? Without that VRA won't do anything. Go to my web site and you'll see some screen shots of VRA taking you through the initial steps. Let me know if you're still stuck:

      http://zarnia.250free.com/

      For those already familiar with my site, it seems I made a mistake interpreting the chessboard pattern. The curve that generates it isn't a sinusoid but a square wave (which can be neatly expressed as a Fourier series: I'll put it up on the site).

      I think I'll take a look at the EuroMillions results (of which are there are still less than a hundred as of this post)

      "God is subtle but He is not malicious." - Albert Einstein

      "In my experience there's no such thing as luck." - Alec Guinness, Star Wars Episode IV

        retxx's avatar - mrthumbs
        BOSTON
        United States
        Member #48
        September 9, 2001
        3584 Posts
        Offline
        Posted: September 6, 2005, 11:22 am - IP Logged

        can I load all the pick daily 4 Mass. numbers since 1979 into vra and if so how is it done.Thanks also will it analyze for play.?Thanks

          NemeSys's avatar - Nemesis
          Harrogate
          United Kingdom
          Member #16589
          June 1, 2005
          107 Posts
          Offline
          Posted: September 6, 2005, 5:27 pm - IP Logged

          VRA will take any ASCII file of numerical data. Click on 'File', then 'Open signal'. Select 'all files' from 'Files of type' then click on 'Open', then you're ready to start experimenting.

          "God is subtle but He is not malicious." - Albert Einstein

          "In my experience there's no such thing as luck." - Alec Guinness, Star Wars Episode IV

            retxx's avatar - mrthumbs
            BOSTON
            United States
            Member #48
            September 9, 2001
            3584 Posts
            Offline
            Posted: September 6, 2005, 10:42 pm - IP Logged

            another question; do I enter just the 4 digit numbers or do I need the date for any reason?thanks

              NemeSys's avatar - Nemesis
              Harrogate
              United Kingdom
              Member #16589
              June 1, 2005
              107 Posts
              Offline
              Posted: September 7, 2005, 3:44 am - IP Logged

              Delete all non-result values from your file (such as the date), otherwise the program will treat them as part of the data. Also, remove any non-numerical entries (i.e. alphabetic characters and symbols).

              "God is subtle but He is not malicious." - Albert Einstein

              "In my experience there's no such thing as luck." - Alec Guinness, Star Wars Episode IV

                FrankieH's avatar - cat anm.gif

                United States
                Member #21258
                September 5, 2005
                267 Posts
                Offline
                Posted: September 7, 2005, 8:57 am - IP Logged

                Hi NemeSys,

                 

                Thank you very much for your advice, i did it and it worked. I think i might have a handle on comprehending the software too (albeit, not in a analytical fashion!) Wink Thanks again for your help.

                FrankieH 

                 

                  mybolade's avatar - praying hands.jpg
                  Detroit, MI
                  United States
                  Member #8705
                  November 15, 2004
                  1396 Posts
                  Offline
                  Posted: September 7, 2005, 4:49 pm - IP Logged

                  FrankieH,

                   

                  I am interested in the software if it words for pic 4.  Have you had any positive results? I do not play the lottery, only pic 3 and pic 4.  I have followed the forum from the beginning and appreciate NemeSys and others for your help.  However, you have been talking about the Lotto.

                  I am glad to see someone is working with pic 4. 

                  Bolade' 

                    NemeSys's avatar - Nemesis
                    Harrogate
                    United Kingdom
                    Member #16589
                    June 1, 2005
                    107 Posts
                    Offline
                    Posted: September 7, 2005, 7:16 pm - IP Logged

                    Glad to be of help, FrankieH and Bolade (and thank you to everyone else for making this a well-viewed thread).

                    Provided you have a complete set of results you can analyse any game with VRA.

                    I have begun looking into the EuroMillions game, and although the set of results is small it already looks promising. The best values for dimension seem to be 21 or 22, and 7 for delay. In fact, I've noticed the optimal value of delay appears to correspond to the number of balls per draw. For EuroMillions there are 5 main balls and 2 'lucky stars', hence del=7. For the UK main draw there are 6 balls, so del=6 (the seventh bonus ball is irrelevant). For the Thunderball, 5 main balls and the sixth Thunderball, so again del=6, and so on. This just leaves dimension to be pinned down (and there's no reason a best value for del doesn't exist).

                    I will put up some VRA graphs of lottery results on my site, they make interesting viewing.

                    "God is subtle but He is not malicious." - Albert Einstein

                    "In my experience there's no such thing as luck." - Alec Guinness, Star Wars Episode IV

                      Avatar
                      New Member
                      MA
                      United States
                      Member #21195
                      September 3, 2005
                      6 Posts
                      Offline
                      Posted: September 7, 2005, 9:29 pm - IP Logged

                      I decided to give it try, fully expecting to see an unambiguous confirmation that there is nothing to predict when it comes to random numbers.  The data for this experiment was the "UK six main balls" dataset that I downloaded from NemeSys' website. Below are the prediction results for the 100 balls, along with the model parameters. As you can see, the correlation between the predicted and the actual results is enormous. That visual result is also confirmed by a very low value of NMSE.

                      I am very suspicios about this unexpectedly successful prediction, and all I can think of is that either the source data set is not recorded correctly, or that the machine that generates those balls is really messed up. Something is missing here, because otherwise I would expect all of you guys (including me) sailing your private yachts instead of hanging aroung here.

                      Eugene Kononov, author of Visual Recurrence Analysis.

                      P.S. I took another look at it, and realized that one explanation for the high visual correlation is that the results for each draw are sorted from the lowest to the largest ball. That makes the prediction easy for VRA (since each draw result is a set of points whose regression line is always up), and the visual examination may be misleading. What needs to be done for a fair test is that the generated predictions should be evaluated in terms of how much money they make over the test period. If the test period is large enough (say 200 draws), that should converge the performance of the predictions to some stable value. I don't know much about lottery in general (or the payoffs and odds of this particular UK 6 balls lottery), so I'll leave this analysis to you, guys.

                      prediction results

                        NemeSys's avatar - Nemesis
                        Harrogate
                        United Kingdom
                        Member #16589
                        June 1, 2005
                        107 Posts
                        Offline
                        Posted: September 7, 2005, 10:26 pm - IP Logged

                        You had partly anticipated my reply. Also, the results are presented as a continuous string of numbers, when in reality each group of numbers is separated by two or three days. As some people have noted perhaps a fairer test would be to analyse the results in draw order rather than by ascending order. However, the predictions I'm getting don't really justify re-ordering the data (though I could always be wrong). The whole point of ascending numerical order is that it 'helps' VRA in making predictions.

                        A long-term trial would indeed settle the issue (though 200 draws from the UK main lottery means waiting for two years).

                        "God is subtle but He is not malicious." - Albert Einstein

                        "In my experience there's no such thing as luck." - Alec Guinness, Star Wars Episode IV

                          Avatar
                          New Member
                          MA
                          United States
                          Member #21195
                          September 3, 2005
                          6 Posts
                          Offline
                          Posted: September 7, 2005, 10:52 pm - IP Logged

                          Another thing that didn't immediately occur to me is that for the 6-ball loto, each prediction should be a set of 6 numbers. The only way to do this in VRA is to use the "multi-step" model type, which is using the prediction for time t to make a prediction for time t+1. This type of forecast is by nature inferior to a "one-step" type, where the actual result at time t is used to make a prediction for time t+1. I hope I am explaining it well. The analogy is forecasting the wheather: it's obvious that a forecast 1 day ahead will be more accurate than a forecast 100 days ahead. The only way to make a one-step prediction in VRA for a 6-ball loterry is to use muliti-step prediction for the prediction time window of length 6, record the result, then move the prediction window (both the predict from and predict to) 6 points forward, record the result, and so on, for each prediction of 6 numbers. That is a rather painful procedure (compared to just generating all the predictions in one shot using the multi-step as shown in the graph above, which is also valid, but probably less accurate), but that's all I have in VRA at the moment. I really meant VRA for the analysis of the natural and continuous series, such as heart beat, stock market, audio waves, etc.  The lotto is indeed unnatural in the sense that I need to predict multiple results and treat it as a single prediction.

                           

                           

                           

                            Avatar
                            New Member
                            MA
                            United States
                            Member #21195
                            September 3, 2005
                            6 Posts
                            Offline
                            Posted: September 7, 2005, 10:59 pm - IP Logged

                            You had partly anticipated my reply. Also, the results are presented as a continuous string of numbers, when in reality each group of numbers is separated by two or three days. As some people have noted perhaps a fairer test would be to analyse the results in draw order rather than by ascending order. However, the predictions I'm getting don't really justify re-ordering the data (though I could always be wrong). The whole point of ascending numerical order is that it 'helps' VRA in making predictions.

                            A long-term trial would indeed settle the issue (though 200 draws from the UK main lottery means waiting for two years).

                            What I actually meant was that the analysis of the 200 predictions can be made using the past data. Since VRA pretends not to see the data in the test interval (if it is set to be after the training interval), the fair test of the prediction quality can be evaluated using the actual draws that have occured in the past.  This is sometimes is refered as an "out-of-sample" prediction and is considered a standard method of evaluating the prediction models. That saves you 2 years of waiting before making the conclusion about the model effectiveness. Big Smile

                              NemeSys's avatar - Nemesis
                              Harrogate
                              United Kingdom
                              Member #16589
                              June 1, 2005
                              107 Posts
                              Offline
                              Posted: September 8, 2005, 3:47 am - IP Logged

                              Of course! I should have realised that! I use "post-dicting" to build tables of dimension and delay. From these I select those values which give two or more of the actual numbers. It's still laborious, though.

                              Your screen capture shows a Kernel regression predictor, a Tricube kernel, and Euclidean distance. I use the Nearest Neighbour predictor with Euclidean measure (the blue and red curves match exactly then, but that doesn't guarantee highly accurate predictions). As you know, the Nearest Neighbour option is multi-step only, but that isn't a problem since I often go for predicting all the draw numbers in one go (unless I look at the results column by column).

                              It's a great piece of software, by the way, and so far seems to have the edge over more conventional lottery predictor programs.

                              "God is subtle but He is not malicious." - Albert Einstein

                              "In my experience there's no such thing as luck." - Alec Guinness, Star Wars Episode IV

                                 
                                Page 8 of 12