Welcome Guest
Log In | Register )
You last visited January 19, 2017, 6:21 pm
All times shown are
Eastern Time (GMT-5:00)

Nonlinear regression to predict lotteries

Topic closed. 119 replies. Last post 2 years ago by lottoswe.

Page 6 of 8
53
PrintE-mailLink
Avatar
bgonçalves
Brasil
Member #92564
June 9, 2010
2134 Posts
Offline
Posted: December 22, 2014, 11:42 am - IP Logged

Hello, rl, the calculation is also possible when the draws are registered in order of
  Sweepstakes?

    RJOh's avatar - chipmunk
    mid-Ohio
    United States
    Member #9
    March 24, 2001
    19900 Posts
    Offline
    Posted: December 22, 2014, 1:35 pm - IP Logged
    My previous post was aimed atRL- RANDOMLOGIC .bob790 could also indicate the basic data of % correct and economic ROI or overall performance of the model / system.You may indicate too.Show this data means nothing for the user indicates ...

    I assumed bob790 was trying to get a conversation started about nonlinear regression since he had nothing to sell and that was my reasoning for looking
    at his prediction results to get an idea of how well his theory was working.  I have a few ideas of my own and have been posting on the predictions
    board the combinations I have been playing based on my ideas.  If they prove to be worthwhile I will consider sharing them with anyone who is interested

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

      RL-RANDOMLOGIC's avatar - usafce

      United States
      Member #59354
      March 13, 2008
      4079 Posts
      Offline
      Posted: December 22, 2014, 6:50 pm - IP Logged

      dr san

      I am kind of excited about this new program, Setting 10 rogue values will produce a 5of5 on one line

      as compared to setting DMP's 10 digit filters to single values which produced 45 lines in last nights

      game.   Most of the digit filters will range around 0 to 3 for most games so in a heads up comparison

      the new program beats it  hand's down.  You can also see in the pic below that using the digit filters

      in DMP that there were 24 non-prize paying lines.  If I set the new program to generate using just 5

      rogue values it produces 563 lines and every line will at least break even.  It's a very nice reduction

      tool and I hope I can build a nice predictor for it.   Anyway, only time will tell.

      RL 

       

       

      dmp-5

       

      rogue sm5

      Working on my Ph.D.  "University of hard Knocks"

      I will consider the opinion that my winnings are a product of chance if you are willing to consider

      they are not.  Many great discoveries come while searching for something else

      USAF https://en.wikipedia.org/wiki/Prime_Base_Engineer_Emergency_Force

        US Flag Trump / 2016 & 2020  

        RL-RANDOMLOGIC's avatar - usafce

        United States
        Member #59354
        March 13, 2008
        4079 Posts
        Offline
        Posted: December 22, 2014, 6:59 pm - IP Logged

        Playing 7 rogue values, 36 lines

        rogue 7

        Working on my Ph.D.  "University of hard Knocks"

        I will consider the opinion that my winnings are a product of chance if you are willing to consider

        they are not.  Many great discoveries come while searching for something else

        USAF https://en.wikipedia.org/wiki/Prime_Base_Engineer_Emergency_Force

          US Flag Trump / 2016 & 2020  

          Avatar
          bgonçalves
          Brasil
          Member #92564
          June 9, 2010
          2134 Posts
          Offline
          Posted: December 22, 2014, 7:11 pm - IP Logged

          Hello, rl, perfect ok !! Thank you, yes, position by position separate bear fruit
          And also seeing the relationship with the next example column
          2 with the column 1, column 2 with Column3. etc., and delta values by position

            bootleg233's avatar - Lottery-034.jpg
            Tn
            United States
            Member #54963
            September 4, 2007
            1166 Posts
            Offline
            Posted: December 28, 2014, 6:41 am - IP Logged

            dr san

            I am kind of excited about this new program, Setting 10 rogue values will produce a 5of5 on one line

            as compared to setting DMP's 10 digit filters to single values which produced 45 lines in last nights

            game.   Most of the digit filters will range around 0 to 3 for most games so in a heads up comparison

            the new program beats it  hand's down.  You can also see in the pic below that using the digit filters

            in DMP that there were 24 non-prize paying lines.  If I set the new program to generate using just 5

            rogue values it produces 563 lines and every line will at least break even.  It's a very nice reduction

            tool and I hope I can build a nice predictor for it.   Anyway, only time will tell.

            RL 

             

             

            dmp-5

             

            rogue sm5

            Looks great RL be fun to fool with this see what develops!  Think we will get to try it on sometime?  Da Boot!!!

            WHEN IT FEELS THE WHOLE WORLD SUCKS!

            RELAX.........IT'S ONLY GRAVITY Big Smile

            I think I can I think I can!!!!

              Avatar
              Tahiti- Polynesia
              Tuvalu
              Member #34524
              March 4, 2006
              54 Posts
              Offline
              Posted: December 28, 2014, 9:14 pm - IP Logged

              With Euromillions 5/50, with numbers between 1 and 50, variations are large and the curb is chaotic. Prediction is more difficult. We use nonlinear regression.

              Or you can use a model to make for example 10 groups (you can use FFT for equalice numbers in groups). The model is not valid if you obtain one group with 30 numbers and others groups with 2 or 3 numbers. Here is a result example of good theorical model:

              • group 1: 12 numbers
              • group 2: 9 numbers
              • group 3: 8 numbers
              • group 4: 7 numbers
              • group 5: 5 numbers
              • group 6: 5 numbers
              • group 7: 4 numbers

              Then you can use LinReg to this groups; if the LinReg function show a group to predict, then you know the numbers of this group.

              Hi anubinareloaded,

              Yes, lotteries are chaotic. The more numbers, the more chaotic. Graphs are chaotic.  Now, you can use techniques to make the graph less chaotic as scientists do. Some use normalization. I had little success with that. Other use splines to smooth the graph. I got better results with this technique using a sorted historyfile of drawings. That's a little surprising as unsorted history is the logic way using nonlinear regression but using the sorted file make the history less random.

              Using splines, you insert intermediary numbers between two successive numbers. I use B-spline to insert 80 intermediary numbers between two successive numbers. If you have say 1000 numbers in your file, your new file will have 80000 numbers. That's huge but you'll get better predictions. Then, when predicting, you'll take the 80th number predicted as the prediction for number 1. Then you add the predict number to your original file and use spline to get the history file for number 2, etc...

              That's not an easy task but that's the way to go.

              Other thing. Nonlinear regression give the best results when you already have 2.3 or more numbers which popped in a former drawing say the last drawing or 2 or 3 drawing before. 

              Bob


                United Kingdom
                Member #162012
                December 19, 2014
                23 Posts
                Offline
                Posted: December 29, 2014, 3:40 pm - IP Logged

                Hi anubinareloaded,

                Yes, lotteries are chaotic. The more numbers, the more chaotic. Graphs are chaotic.  Now, you can use techniques to make the graph less chaotic as scientists do. Some use normalization. I had little success with that. Other use splines to smooth the graph. I got better results with this technique using a sorted historyfile of drawings. That's a little surprising as unsorted history is the logic way using nonlinear regression but using the sorted file make the history less random.

                Using splines, you insert intermediary numbers between two successive numbers. I use B-spline to insert 80 intermediary numbers between two successive numbers. If you have say 1000 numbers in your file, your new file will have 80000 numbers. That's huge but you'll get better predictions. Then, when predicting, you'll take the 80th number predicted as the prediction for number 1. Then you add the predict number to your original file and use spline to get the history file for number 2, etc...

                That's not an easy task but that's the way to go.

                Other thing. Nonlinear regression give the best results when you already have 2.3 or more numbers which popped in a former drawing say the last drawing or 2 or 3 drawing before. 

                Bob

                Buenas tardes bob, y felices fiestas!

                "using a sorted historyfile of drawings" -> I understand that you order the data 'in crescendo' by frequency or appearances, but... you translate the problem to the number of draw, ok? You translate the incognita from axis Y to axis X, then I suppose that you can resolve the X incognita with NonLinReg... What is the performance of this technique?

                "Using splines, you insert intermediary numbers between two successive numbers" -> Do you refer to 'Interpolation Thechnique'?

                "I use B-spline to insert 80 intermediary numbers between two successive numbers. If you have say 1000 numbers in your file, your new file will have 80000 numbers" -> Have you tried to insert the numbers that have not been winners in a drawing but having frequency information? I do not understand the competitive advantage of spline... Spline applied to NonLinReg?

                "That's huge but you'll get better predictions" -> I understand , slow to calculate about 4 hours a draw (with 7 cores)

                "Then, when predicting, you'll take the 80th number predicted as the prediction for number 1" -> Do you refer to 'First ball drawn' for number 1 o do you refer number 1 to 'ball identified by number 1'?

                "Nonlinear regression give the best results when you already have 2.3 or more numbers which popped in a former drawing say the last drawing or 2 or 3 drawing before" -> Then you use a Trend observed in the past drawing (2, 3..) ok?

                I am googliano, so I request patience to understand

                  Avatar
                  Tahiti- Polynesia
                  Tuvalu
                  Member #34524
                  March 4, 2006
                  54 Posts
                  Offline
                  Posted: December 29, 2014, 9:31 pm - IP Logged

                  Buenas tardes bob, y felices fiestas!

                  "using a sorted historyfile of drawings" -> I understand that you order the data 'in crescendo' by frequency or appearances, but... you translate the problem to the number of draw, ok? You translate the incognita from axis Y to axis X, then I suppose that you can resolve the X incognita with NonLinReg... What is the performance of this technique?

                  "Using splines, you insert intermediary numbers between two successive numbers" -> Do you refer to 'Interpolation Thechnique'?

                  "I use B-spline to insert 80 intermediary numbers between two successive numbers. If you have say 1000 numbers in your file, your new file will have 80000 numbers" -> Have you tried to insert the numbers that have not been winners in a drawing but having frequency information? I do not understand the competitive advantage of spline... Spline applied to NonLinReg?

                  "That's huge but you'll get better predictions" -> I understand , slow to calculate about 4 hours a draw (with 7 cores)

                  "Then, when predicting, you'll take the 80th number predicted as the prediction for number 1" -> Do you refer to 'First ball drawn' for number 1 o do you refer number 1 to 'ball identified by number 1'?

                  "Nonlinear regression give the best results when you already have 2.3 or more numbers which popped in a former drawing say the last drawing or 2 or 3 drawing before" -> Then you use a Trend observed in the past drawing (2, 3..) ok?

                  I am googliano, so I request patience to understand

                  Hi,

                  You asked :

                  1 -"using a sorted historyfile of drawings" -> I understand that you order the data 'in crescendo' by frequency or appearances, but... you translate the problem to the number of draw, ok? You translate the incognita from axis Y to axis X, then I suppose that you can resolve the X incognita with NonLinReg... What is the performance of this technique?

                  Let's take the last Euromillions drawing :

                  26/12/2014;26;45;27;49;17 Unsorted

                  26/12/2014;17;26;27;45;49 Sorted

                  I update my history files adding :

                  26
                  45
                  27
                  49
                  17

                  to the unsorted history file and

                  17
                  26
                  27
                  45
                  49

                  to the sorted history file.

                   

                  As I said, results are better when one previous drawing has already 2 or 3 numbers repeating in the new drawing. You cannot know when that will happen, but it will happen regularly. Just wait for it.

                   

                  2 -"Using splines, you insert intermediary numbers between two successive numbers" -> Do you refer to 'Interpolation Thechnique'?

                  Yes, that is interpolation with spline.

                  3 -"I use B-spline to insert 80 intermediary numbers between two successive numbers. If you have say 1000 numbers in your file, your new file will have 80000 numbers" -> Have you tried to insert the numbers that have not been winners in a drawing but having frequency information? I do not understand the competitive advantage of spline... Spline applied to NonLinReg?

                  Interpolation will give a smoother graph. You can also use the Kalman filter to get a processed file. Please,read articles about those techniques to have a better understanding.

                  About Spline applied to NonLinReg : Yes, you can use the technique you want to pre-process your time series, here, your history file. You just use nonlinear regression because your data is nonlinear.

                  4 -"That's huge but you'll get better predictions" -> I understand , slow to calculate about 4 hours a draw (with 7 cores)

                  Not so long to process. Using the entire file is not mandatory. In fact, I only use 40000 data, training from 1 to 40000 and predicting from 4001 to 40080. Only tests will show you the number of data to use. I just use half of my history file to train.

                  5 -"Then, when predicting, you'll take the 80th number predicted as the prediction for number 1" -> Do you refer to 'First ball drawn' for number 1 o do you refer number 1 to 'ball identified by number 1'?

                  To predict the first number, I train 1 to 4000 and predict from 40001 to 40080. I take the 40080th number as the prediction number. That will be the first number predicted. It can be say 3.5343743. I will round it to 4. Using splines will give you a history file with decimal numbers.

                  6 -"Nonlinear regression give the best results when you already have 2.3 or more numbers which popped in a former drawing say the last drawing or 2 or 3 drawing before" -> Then you use a Trend observed in the past drawing (2, 3..) ok?

                  No. I don't care about that. I just hope that will happen regularly. You need patience.

                  I don't look at statistics, I don't look at numbers. I just update my files, put them in one column, and apply nonlinear regression. The biggest challenge is to find the best parameters for prediction : dimension, delay and number of nearest numbers.Not an easy thing.

                  There are many more tips to say. I'm exploring all techniques used by scientists to get better results.

                   

                  Bob

                    Avatar
                    Tahiti- Polynesia
                    Tuvalu
                    Member #34524
                    March 4, 2006
                    54 Posts
                    Offline
                    Posted: December 29, 2014, 9:52 pm - IP Logged

                    Hi,

                    Finding the 5 numbers in Euromillions is very easy. We know how to do that : play all combinations. So, the challenge is to find just some numbers you can afford to play. My basic playing technique is to choose 10 numbers, combine them then filtering out the most improbable combinations. You can also use 15 numbers, using the last three drawings. One or two times a year, the 5 numbers have already popped up in these last 15 numbers. You'll often find repeaters in those last drawings, so that you'll be left with 14 or 13 numbers to play. You just need patience.


                      United Kingdom
                      Member #162012
                      December 19, 2014
                      23 Posts
                      Offline
                      Posted: January 5, 2015, 7:21 am - IP Logged

                      Hi,

                      You asked :

                      1 -"using a sorted historyfile of drawings" -> I understand that you order the data 'in crescendo' by frequency or appearances, but... you translate the problem to the number of draw, ok? You translate the incognita from axis Y to axis X, then I suppose that you can resolve the X incognita with NonLinReg... What is the performance of this technique?

                      Let's take the last Euromillions drawing :

                      26/12/2014;26;45;27;49;17 Unsorted

                      26/12/2014;17;26;27;45;49 Sorted

                      I update my history files adding :

                      26
                      45
                      27
                      49
                      17

                      to the unsorted history file and

                      17
                      26
                      27
                      45
                      49

                      to the sorted history file.

                       

                      As I said, results are better when one previous drawing has already 2 or 3 numbers repeating in the new drawing. You cannot know when that will happen, but it will happen regularly. Just wait for it.

                       

                      2 -"Using splines, you insert intermediary numbers between two successive numbers" -> Do you refer to 'Interpolation Thechnique'?

                      Yes, that is interpolation with spline.

                      3 -"I use B-spline to insert 80 intermediary numbers between two successive numbers. If you have say 1000 numbers in your file, your new file will have 80000 numbers" -> Have you tried to insert the numbers that have not been winners in a drawing but having frequency information? I do not understand the competitive advantage of spline... Spline applied to NonLinReg?

                      Interpolation will give a smoother graph. You can also use the Kalman filter to get a processed file. Please,read articles about those techniques to have a better understanding.

                      About Spline applied to NonLinReg : Yes, you can use the technique you want to pre-process your time series, here, your history file. You just use nonlinear regression because your data is nonlinear.

                      4 -"That's huge but you'll get better predictions" -> I understand , slow to calculate about 4 hours a draw (with 7 cores)

                      Not so long to process. Using the entire file is not mandatory. In fact, I only use 40000 data, training from 1 to 40000 and predicting from 4001 to 40080. Only tests will show you the number of data to use. I just use half of my history file to train.

                      5 -"Then, when predicting, you'll take the 80th number predicted as the prediction for number 1" -> Do you refer to 'First ball drawn' for number 1 o do you refer number 1 to 'ball identified by number 1'?

                      To predict the first number, I train 1 to 4000 and predict from 40001 to 40080. I take the 40080th number as the prediction number. That will be the first number predicted. It can be say 3.5343743. I will round it to 4. Using splines will give you a history file with decimal numbers.

                      6 -"Nonlinear regression give the best results when you already have 2.3 or more numbers which popped in a former drawing say the last drawing or 2 or 3 drawing before" -> Then you use a Trend observed in the past drawing (2, 3..) ok?

                      No. I don't care about that. I just hope that will happen regularly. You need patience.

                      I don't look at statistics, I don't look at numbers. I just update my files, put them in one column, and apply nonlinear regression. The biggest challenge is to find the best parameters for prediction : dimension, delay and number of nearest numbers.Not an easy thing.

                      There are many more tips to say. I'm exploring all techniques used by scientists to get better results.

                       

                      Bob

                      1 -"using a sorted historyfile of drawings -> I understand that you order the data 'in crescendo' by frequency or appearances, but... you translate the problem to the number of draw, ok? You translate the incognita from axis Y to axis X, then I suppose that you can resolve the X incognita with NonLinReg... What is the performance of this technique?

                      Let's take the last Euromillions drawing :

                      26/12/2014;26;45;27;49;17 Unsorted

                      26/12/2014;17;26;27;45;49 Sorted

                      I update my history files adding :

                      26
                      45
                      27
                      49
                      17

                      to the unsorted history file and

                      17
                      26
                      27
                      45
                      49

                      to the sorted history file.

                       

                      As I said, results are better when one previous drawing has already 2 or 3 numbers repeating in the new drawing. You cannot know when that will happen, but it will happen regularly. Just wait for it.

                      Do you think that ordering the balls by pre-assigned numeric identifier get some competitive advantage?

                        Avatar
                        Tahiti- Polynesia
                        Tuvalu
                        Member #34524
                        March 4, 2006
                        54 Posts
                        Offline
                        Posted: January 5, 2015, 10:09 pm - IP Logged

                        1 -"using a sorted historyfile of drawings -> I understand that you order the data 'in crescendo' by frequency or appearances, but... you translate the problem to the number of draw, ok? You translate the incognita from axis Y to axis X, then I suppose that you can resolve the X incognita with NonLinReg... What is the performance of this technique?

                        Let's take the last Euromillions drawing :

                        26/12/2014;26;45;27;49;17 Unsorted

                        26/12/2014;17;26;27;45;49 Sorted

                        I update my history files adding :

                        26
                        45
                        27
                        49
                        17

                        to the unsorted history file and

                        17
                        26
                        27
                        45
                        49

                        to the sorted history file.

                         

                        As I said, results are better when one previous drawing has already 2 or 3 numbers repeating in the new drawing. You cannot know when that will happen, but it will happen regularly. Just wait for it.

                        Do you think that ordering the balls by pre-assigned numeric identifier get some competitive advantage?

                        Hi,

                        I just choose the method which gave me the best results till now, and try to make it better. I can include others methods to see what happen. 

                        Numbers on balls are just references to each ball with his own specifications. : you can give each number a name, or anything you want as a, b,c.... With splines however, references must be numbers. 

                         Difficult for me to explain my whole method and how I see lottery. 

                        Bob


                          United Kingdom
                          Member #162012
                          December 19, 2014
                          23 Posts
                          Offline
                          Posted: January 6, 2015, 4:01 am - IP Logged

                          Hi,

                          I just choose the method which gave me the best results till now, and try to make it better. I can include others methods to see what happen. 

                          Numbers on balls are just references to each ball with his own specifications. : you can give each number a name, or anything you want as a, b,c.... With splines however, references must be numbers. 

                           Difficult for me to explain my whole method and how I see lottery. 

                          Bob

                          OK, thank you.

                          Do you have data (%) on successes with this system?

                          My performance data on 100 sweepstakes EuroMillions (main 5/50 balls) in a selection pool of 15 numbers:

                          1 winner number : 57% - 87%
                          2 Winning numbers: 44% - 65%
                          3 winning numbers: 23% - 42%
                          4 winning numbers: 4% - 17%
                          5 winning numbers: 0% - 6%

                            SergeM's avatar - slow icon.png
                            Economy class
                            Belgium
                            Member #123700
                            February 27, 2012
                            4035 Posts
                            Offline
                            Posted: January 6, 2015, 6:28 am - IP Logged

                            Compare to N:

                            2,5X
                            2XX
                            1,5XXXXX
                            1XXXXXXX
                            0,5XXXXXXXXX
                            XXXXXXXX
                            0,5XXXXX
                            1XXX
                            1,5XXX
                            2X
                            2,5X

                            First column approx.

                            http://www.lotterypost.com/thread/284274/3920689

                            http://www.lotterypost.com/thread/284274/3920693

                            ...

                            The above you understand, if you had enough mathematics at school or in self study.

                              Avatar
                              Tahiti- Polynesia
                              Tuvalu
                              Member #34524
                              March 4, 2006
                              54 Posts
                              Offline
                              Posted: January 6, 2015, 5:15 pm - IP Logged

                              Hi,

                              Are those statistics calculated on real tickets you bought, or they calculated on a theorical basis? How do you get them?

                              I don't calculate the statistics of my results. If "Sweepstakes" mean "ready made tickets", then I'm sure that my results are far better than random tickets sold by Euromillions. 

                              5/15 happen 1, or more rarely, 2 times a year. Drawings happen 2 times a week, that's 104 times a year. So, success percentage is closer to 1 or 2%. 6% is huge.

                               

                              Bob