Welcome Guest
Log In | Register )
You last visited December 4, 2016, 7:21 pm
All times shown are
Eastern Time (GMT-5:00)

Artificial Neural Network for CUDA

Topic closed. 10 replies. Last post 3 months ago by lottologix.

Page 1 of 1
51
PrintE-mailLink
MillionsWanted's avatar - 24Qa6LT

Norway
Member #9517
December 10, 2004
1272 Posts
Online
Posted: August 22, 2016, 2:50 pm - IP Logged

MBP for CUDA.

If you got a NVIDIA GTX card or a NVIDIA Tesla card, you might be interested in testing it on your lotteries.

There's a tutorial on the website.

I don't think it can replace NNLP which is very flexible and allow more than one line to be predicted, but it can be used to experiment with.

I haven't tested it myself yet, so I can't help you with any settings.

    adulane62's avatar - file php?avatar=16228.gif
    From Denver, Rocky Mountain Empire,
    United States
    Member #49750
    February 13, 2007
    439 Posts
    Offline
    Posted: August 25, 2016, 11:44 pm - IP Logged

    It won't import game like perceptelon or NNLP. Used all forms like space, dash, etc. between numbers. Still won't work. What gives?

    Go Broncos!  White Bounce

      MillionsWanted's avatar - 24Qa6LT

      Norway
      Member #9517
      December 10, 2004
      1272 Posts
      Online
      Posted: August 26, 2016, 3:30 am - IP Logged

      It won't import game like perceptelon or NNLP. Used all forms like space, dash, etc. between numbers. Still won't work. What gives?

      I've been trying both .txt files and Excel .csv files. It's a minor mystery. Not giving up yet.

        MillionsWanted's avatar - 24Qa6LT

        Norway
        Member #9517
        December 10, 2004
        1272 Posts
        Online
        Posted: August 26, 2016, 4:20 am - IP Logged

        It won't import game like perceptelon or NNLP. Used all forms like space, dash, etc. between numbers. Still won't work. What gives?

        If your lottery got 6 numbers the topology must be for example 3 input and 3 output before you start training.

          MillionsWanted's avatar - 24Qa6LT

          Norway
          Member #9517
          December 10, 2004
          1272 Posts
          Online
          Posted: August 26, 2016, 5:18 am - IP Logged

          For some reason, the MBP does not show a prediction based on the training. It only shows the result of the training.

            adulane62's avatar - file php?avatar=16228.gif
            From Denver, Rocky Mountain Empire,
            United States
            Member #49750
            February 13, 2007
            439 Posts
            Offline
            Posted: August 26, 2016, 11:19 am - IP Logged

            If your lottery got 6 numbers the topology must be for example 3 input and 3 output before you start training.  

             

                                                                           Thinking of...

            MW, how do I input 3 numbers in the first place, if it won't run any file? Thanks!

            Go Broncos!  White Bounce

              MillionsWanted's avatar - 24Qa6LT

              Norway
              Member #9517
              December 10, 2004
              1272 Posts
              Online
              Posted: August 26, 2016, 11:40 am - IP Logged

              If your lottery got 6 numbers the topology must be for example 3 input and 3 output before you start training.  

               

                                                                             Thinking of...

              MW, how do I input 3 numbers in the first place, if it won't run any file? Thanks!

              Did you create the file in Excel or Notepad? I had no problems with a .CSV file made in Excel.

              You need to create the topology before you start the training. The input and output must match the number of columns in your file.

              As I wrote earlier, it doesn't seem to have a predictions feature, so perhaps it's a waste of time.

                adulane62's avatar - file php?avatar=16228.gif
                From Denver, Rocky Mountain Empire,
                United States
                Member #49750
                February 13, 2007
                439 Posts
                Offline
                Posted: August 26, 2016, 12:08 pm - IP Logged

                I give up! This thing is too complicated. Thanks for your help anyway!

                Go Broncos!  White Bounce

                  Avatar

                  United States
                  Member #139779
                  March 3, 2013
                  115 Posts
                  Offline
                  Posted: September 4, 2016, 11:30 pm - IP Logged

                  MBP for CUDA.

                  If you got a NVIDIA GTX card or a NVIDIA Tesla card, you might be interested in testing it on your lotteries.

                  There's a tutorial on the website.

                  I don't think it can replace NNLP which is very flexible and allow more than one line to be predicted, but it can be used to experiment with.

                  I haven't tested it myself yet, so I can't help you with any settings.

                  The deep learning stuff is on NVIDIAs web site.  The Microsoft CNTK looks very promising, however you are looking at some very expensive hardware, like 4 GTX 1080s costing hundreds of dollars. Then there's the programming aspect of it, even though it can be done with just a command line interface. You would be looking at every aspect of the ANN not to mention the learning curve for that software. Mabe if we can see something promising with NNLP or the Predictor from Ward Systems which is genetically optimized, it may be worth looking further. So far I am experimenting with sums and category's of box numbers that come out X number of times in a year with NNLP. Since these ideas have a high repetition rate, and the Neural Net is looking for patterns repeating in the numbers NNLP may be able to spot a trend that just the straight statistics may not. Atleast make a better guess than I would make on my own.

                  I think NNlp is a feed foward NN not back propagation. I may experiment with the MBP software, thanks for the link. :)

                    MillionsWanted's avatar - 24Qa6LT

                    Norway
                    Member #9517
                    December 10, 2004
                    1272 Posts
                    Online
                    Posted: September 5, 2016, 9:05 am - IP Logged

                    The deep learning stuff is on NVIDIAs web site.  The Microsoft CNTK looks very promising, however you are looking at some very expensive hardware, like 4 GTX 1080s costing hundreds of dollars. Then there's the programming aspect of it, even though it can be done with just a command line interface. You would be looking at every aspect of the ANN not to mention the learning curve for that software. Mabe if we can see something promising with NNLP or the Predictor from Ward Systems which is genetically optimized, it may be worth looking further. So far I am experimenting with sums and category's of box numbers that come out X number of times in a year with NNLP. Since these ideas have a high repetition rate, and the Neural Net is looking for patterns repeating in the numbers NNLP may be able to spot a trend that just the straight statistics may not. Atleast make a better guess than I would make on my own.

                    I think NNlp is a feed foward NN not back propagation. I may experiment with the MBP software, thanks for the link. :)

                    You don't need the most expensive GTX cards for accelerating with CUDA. All the NVIDIA cards down to GT got CUDA cores.

                    Unfortunately the BMP software does not predict anything, it just learns.

                    I am considering NeuroSolutions ANN which can use the cores of both the AMD Radeon and NVIDIA GTX graphics cards.
                    Makes it easier to switch between the brands. I already own the Predictor XL add-in for Excel, but it's rather slow.

                      Avatar

                      United States
                      Member #139779
                      March 3, 2013
                      115 Posts
                      Offline
                      Posted: September 5, 2016, 1:06 pm - IP Logged

                      You don't need the most expensive GTX cards for accelerating with CUDA. All the NVIDIA cards down to GT got CUDA cores.

                      Unfortunately the BMP software does not predict anything, it just learns.

                      I am considering NeuroSolutions ANN which can use the cores of both the AMD Radeon and NVIDIA GTX graphics cards.
                      Makes it easier to switch between the brands. I already own the Predictor XL add-in for Excel, but it's rather slow.

                      I agree, Predictor XL along with the Analyzer XL suite is more for stock forecasting. If you follow the examples it does a decent job of predicting the next days closing price, not perfect though. So far in predicting sums for pick 3 NNLP  2.1 seems to work ok it tends to predict sums 9-18 which could be nothing better than a good guess due to the high repetition of sums 1-27. At first here I have only been looking for general sums coming out within the next 10 games in back testing using 5 predictions with the lowest error rate. It seems NNLP never quite gets a really low error rate very quickly. I am going to try Predictor XL next because it does give results with lower error rates. I am finding that NNLP seems to work better with these single number lists, this may hold true for Predictor XL. For folks using NNlp for pick 3 and other games it may work out better to arrange the game history from lowest number to highest number and only use the single position number lists in the draws tab of NNLP. Its a copy and paste from excel. Today I am trying to see if I can get a good prediction for next draw for sums in back testing. Going to look for the 1 or 2 sums predicted with lowest error rates in the shortest amount of time, see how that works out.

                      NeuroSolutions looks good I wonder how much the course costs, the website says you get a 200 dollar credit towards the software if you take the course. Neurosolutions 7 priced at 495 but I would want to take the course if not too pricey, reading the book now.