Welcome Guest
Log In | Register )
You last visited December 8, 2016, 12:55 am
All times shown are
Eastern Time (GMT-5:00)

Neural Net Lottery Picker

677 replies. Last post 12 days ago by dr san.

Page 41 of 46
511
PrintE-mailLink
Avatar

United States
Member #98539
October 9, 2010
57 Posts
Offline
Posted: August 16, 2016, 2:57 pm - IP Logged

This is a question about learning rate:

 

One of the parameters in the Neural Net Lottery Picker program is called Learning Rate (also known as alpha).  The setting in this parameter is controlled by moving a slider back and forth.   

When moving the slider as far to the left as you can, you get the value 1E-05.  When moving the slider as far to the right as you can you get the value 1.

 

What I want to know is which one of these two learning rates is the fastest learning rate: 1E-05 on one side or 1 on the other side?

 

Thanks.

 

Kosh

    Avatar

    United States
    Member #98539
    October 9, 2010
    57 Posts
    Offline
    Posted: August 16, 2016, 5:49 pm - IP Logged

    All of the screen shots / screen caps of the various settings in the Neural Net Lottery Picker (NNLP v. 2.1) are now missing from this forum.  Did anyone save them?

     

    I would like to have copies.

     

    Thanks.

     

     

    Kosh

      Avatar
      Krakow
      Poland
      Member #86302
      February 2, 2010
      859 Posts
      Offline
      Posted: August 17, 2016, 6:13 am - IP Logged

      All of the screen shots / screen caps of the various settings in the Neural Net Lottery Picker (NNLP v. 2.1) are now missing from this forum.  Did anyone save them?

       

      I would like to have copies.

       

      Thanks.

       

       

      Kosh

      Kosh

      And what your bet would be? Think a little. Anyway if you go with the slider way far to the right you run a high chance of overfitting.

      If you do the way Millions Wanted suggest i.e. keep it at zero level then there's a good chance no learning process will take place.

        Avatar
        Krakow
        Poland
        Member #86302
        February 2, 2010
        859 Posts
        Offline
        Posted: August 17, 2016, 8:51 am - IP Logged

        Here's my tip. NNLP may give you a nice chance of hitting the first 2, max 3 numbers and that's it. For the remaining ones use other methods.

        You gotta find the right settings for your game. Look at what MillionsWanted has written as this is a good starting point. You have to find the good value of Learning Rate which in my view is not 0, the threshold value and the numbers of games for training.

          MillionsWanted's avatar - 24Qa6LT

          Norway
          Member #9517
          December 10, 2004
          1272 Posts
          Online
          Posted: August 17, 2016, 11:09 am - IP Logged

          Here's my tip. NNLP may give you a nice chance of hitting the first 2, max 3 numbers and that's it. For the remaining ones use other methods.

          You gotta find the right settings for your game. Look at what MillionsWanted has written as this is a good starting point. You have to find the good value of Learning Rate which in my view is not 0, the threshold value and the numbers of games for training.

          It's not zero, but 0.00009 actually. The typical setting for LR is supposed to be 0.3.

          I'll try LR 0.3 on the draw next week of VikingLotto and see if it makes any difference.

            Avatar
            Lyon
            France
            Member #136722
            December 17, 2012
            44 Posts
            Offline
            Posted: August 17, 2016, 1:32 pm - IP Logged

            a new trial for tonight loto 5/49 (results in one hour) shop closed ... but let's see

            02-40-01-28-43-;7,58841504124247;0,511536395753787;0,419165492748023;19:19:02.3769919;Learning Rate:9,9E-05| Noise:0,25|Threshold Error:0,1|Hidden Neurons:200|Test Data Range:29;)
            34-02-14-11-39-;7,38246306943065;0,504547016688567;0,378664273304976;19:22:25.1130542;Learning Rate:9,9E-05| Noise:0,25|Threshold Error:0,1|Hidden Neurons:200|Test Data Range:29;)
            02-08-35-38-48-;7,2462563739919;0,499870892779322;0,400708444742224;19:24:15.7876040;Learning Rate:9,9E-05| Noise:0,25|Threshold Error:0,1|Hidden Neurons:200|Test Data Range:29;)

            Boxing

              Avatar
              Lyon
              France
              Member #136722
              December 17, 2012
              44 Posts
              Offline
              Posted: August 17, 2016, 4:24 pm - IP Logged

              and the result is 12-21-24-35-40

              02-40-01-28-43

              34-02-14-11-39

              02-08-35-38-48

              only one , next time .... but not so bad ....

                lottolot's avatar - Sphere animated2.gif
                Tanhauser Gates
                Holy See (Vatican City State)
                Member #139281
                February 18, 2013
                173 Posts
                Offline
                Posted: August 18, 2016, 6:45 am - IP Logged

                Hello jlg68:

                Another idea:

                You can combine(wheel),the numbers of your three bets(about 12 numbers in this example),and play some bets.

                With a little of luck and filtering,you will get two or more numbers in one winning bet.

                I have never played a bet of 5 numbers with the same last digit(8) in

                3 numbers.Perhaps two numbers is the maximun for me.

                See Ya!Best regards.

                  MillionsWanted's avatar - 24Qa6LT

                  Norway
                  Member #9517
                  December 10, 2004
                  1272 Posts
                  Online
                  Posted: August 18, 2016, 9:27 am - IP Logged

                  I used my settings on page 18 for VikingLotto in Next week, with the exception of Learning Rate which I tested on 0.3 and 0.15.

                  The lower the Total Error, the stranger the output became. I had to stop the training and will go back to the "old" 9E-05(0.00009).

                    Avatar
                    Krakow
                    Poland
                    Member #86302
                    February 2, 2010
                    859 Posts
                    Offline
                    Posted: August 18, 2016, 10:09 am - IP Logged

                    I used my settings on page 18 for VikingLotto in Next week, with the exception of Learning Rate which I tested on 0.3 and 0.15.

                    The lower the Total Error, the stranger the output became. I had to stop the training and will go back to the "old" 9E-05(0.00009).

                    MW, try following settings :

                    05-08-17-24-37-44-;12,3929044970288;0,612815274382113;0,219650652835587;16:03:07.5431528;Learning Rate:0,0192| Noise:0,24|Threshold Error:0,1|Hidden Neurons:200|Test Data Range:33;1-)
                    05-07-18-25-36-45-;12,2329451409627;0,608847523852977;0,210849514552098;16:03:08.9650306;Learning Rate:0,0192| Noise:0,24|Threshold Error:0,1|Hidden Neurons:200|Test Data Range:33;2-)
                    04-09-19-26-38-43-;12,1434674087449;0,606616733095172;0,207347696357373;16:03:34.5432073;Learning Rate:0,0192| Noise:0,24|Threshold Error:0,1|Hidden Neurons:200|Test Data Range:33;1-)
                    05-08-17-23-37-44-;14,5761832110259;0,66460704295564;0,232426332580341;16:04:11.5589127;Learning Rate:0,017067| Noise:0,24|Threshold Error:0,1|Hidden Neurons:200|Test Data Range:33;1-)
                    05-08-18-23-36-44-;12,336940638186;0,611430033614343;0,214044260444525;16:04:37.8245994;Learning Rate:0,017067| Noise:0,24|Threshold Error:0,1|Hidden Neurons:200|Test Data Range:33;1-)
                    05-09-17-24-37-43-;12,158807610587;0,606999765627905;0,218647854143968;16:04:39.2464729;Learning Rate:0,017067| Noise:0,24|Threshold Error:0,1|Hidden Neurons:200|Test Data Range:33;1-)
                    04-10-16-22-38-45-;11,9047850774904;0,600625559691107;0,239013818590761;16:04:52.0902559;Learning Rate:0,017067| Noise:0,24|Threshold Error:0,1|Hidden Neurons:200|Test Data Range:33;0-)
                    05-08-17-24-37-44-;11,8299026484079;0,59873357884505;0,209510490414856;16:05:25.1996982;Learning Rate:0,014935| Noise:0,24|Threshold Error:0,1|Hidden Neurons:200|Test Data Range:33;1-)
                    05-07-18-25-36-45-;11,484387815926;0,58992520881698;0,19387289802022;16:05:28.0590839;Learning Rate:0,014935| Noise:0,24|Threshold Error:0,1|Hidden Neurons:200|Test Data Range:33;2-)
                    05-06-19-23-35-43-;11,3966185205304;0,587666637116409;0,194271855338321;16:06:10.7623016;Learning Rate:0,014935| Noise:0,24|Threshold Error:0,1|Hidden Neurons:200|Test Data Range:33;2-)
                    05-08-17-24-37-44-;12,0383319854959;0,603985048618275;0,188781262932607;16:06:32.5592197;Learning Rate:0,012803| Noise:0,24|Threshold Error:0,1|Hidden Neurons:200|Test Data Range:33;1-)
                    05-09-18-23-36-43-;11,7533751674402;0,596793836983782;0,199210897903377;16:06:39.6998607;Learning Rate:0,012803| Noise:0,24|Threshold Error:0,1|Hidden Neurons:200|Test Data Range:33;1-)
                    05-07-19-25-35-45-;11,397915151742;0,587700066559981;0,206960887888588;16:06:46.8405009;Learning Rate:0,012803| Noise:0,24|Threshold Error:0,1|Hidden Neurons:200|Test Data Range:33;3-)

                    The actual winning numbers were : 5, 7, 19, 41, 46, 49

                      Avatar
                      Lyon
                      France
                      Member #136722
                      December 17, 2012
                      44 Posts
                      Offline
                      Posted: August 18, 2016, 12:09 pm - IP Logged

                      @lottolot

                      Thanks for your advice

                      Yes of course but the goal is to try to have the best straight line with the minimum  error.

                      @adam custom

                      I get your setting.

                      I will give them a trial, but I see

                      Test Data Range:33;2-

                      Please can you enlighten what is the setting data range to get the ;- (0 1 2) at the end

                      Thanks

                        Avatar
                        Krakow
                        Poland
                        Member #86302
                        February 2, 2010
                        859 Posts
                        Offline
                        Posted: August 18, 2016, 12:19 pm - IP Logged

                        @lottolot

                        Thanks for your advice

                        Yes of course but the goal is to try to have the best straight line with the minimum  error.

                        @adam custom

                        I get your setting.

                        I will give them a trial, but I see

                        Test Data Range:33;2-

                        Please can you enlighten what is the setting data range to get the ;- (0 1 2) at the end

                        Thanks

                        I go no idea what it stands for. The training range was 34 draws and I was skipping 1 draw to check the sets against the winning line. Same as if I used the training range of 33 draws in live play.

                        Just as I said earlier if everything gets in place you should be able to the first 2- 3 numbers correct. Sometimes you will see one more but it is by pure chance and you shouldn't count on it, especially as coming directly after the 3 good ones.

                        Still if you can have first 3 correct numbers every now and then your chances go up considerably.

                          Avatar
                          Lyon
                          France
                          Member #136722
                          December 17, 2012
                          44 Posts
                          Offline
                          Posted: August 18, 2016, 12:25 pm - IP Logged

                          I see now it's the  Hit with skipped training

                          Thanks

                            Avatar
                            Krakow
                            Poland
                            Member #86302
                            February 2, 2010
                            859 Posts
                            Offline
                            Posted: August 19, 2016, 6:18 am - IP Logged

                            A word of warning. It's better not to use the skip function for backtesting. The program seems to be not skipping the most recent draw and that's why you can see it hits the first 2 or 3 numbers pretty often. In order to go around this problem, it's better to remove the most recent line and then generate the sets comparing them with the winning line. Then you know for sure what happens.

                              Avatar
                              Lyon
                              France
                              Member #136722
                              December 17, 2012
                              44 Posts
                              Offline
                              Posted: August 20, 2016, 1:15 pm - IP Logged

                              an other try with the same settings

                              30-46-29-01-04-
                              01-33-41-21-19-
                              17-01-35-40-20-

                              39-36-10-18-09-

                              results around 9

                                 
                                Page 41 of 46