Welcome Guest
Log In | Register )
You last visited December 8, 2016, 10:44 pm
All times shown are
Eastern Time (GMT-5:00)

A.N.N.'s : SharkyNN

Topic closed. 12 replies. Last post 7 years ago by Delta Draw.

Page 1 of 1
PrintE-mailLink

United States
Member #81843
October 31, 2009
856 Posts
Offline
Posted: December 28, 2009, 12:46 am - IP Logged

SharkyNN is a free program to give the user a visual seat-of-the-pants insight to training and testing an Artificial Neural Network.

I like the XOR feature for a visual fit and a 3% error stop can be quickly had with few points of input. As opposed to a few years ago, there are many more affordable ANN products for the casual user and user friendliness is greatly improved. ANN's can exceed the human ability in the volume of data processed, but the simplicity and speed of organic decisions has yet to be paralleled through simple five senses observation.

ANN's may better serve the user in looking for patterns amidst many data inputs and variables too complex for human evaluation or, in seeking those patterns not easily revealed in much data.

DD

    Avatar
    komotini
    Greece
    Member #72381
    March 16, 2009
    28 Posts
    Offline
    Posted: December 28, 2009, 11:57 am - IP Logged

    Can you give an example about using this program. How can we  take advantage of this program for prediction of numbers. Thanks.

      PlraX's avatar - large flag_of_dominican_republic.gif
      Deeping in Ramdoness
      Dominican Republic
      Member #73824
      April 25, 2009
      238 Posts
      Offline
      Posted: December 28, 2009, 12:35 pm - IP Logged

      http://sharky-neural-network.software.informer.com/

      this is the URL FOR THAT software

       

      how we can use this software FOR LOTTERY PRUPOSES? its to complicated of use how we can use

       i will get MY COUNTRY jackpot..

      THE SMARTER PEOPLE ...are succesfull, person


        United States
        Member #81843
        October 31, 2009
        856 Posts
        Offline
        Posted: December 29, 2009, 3:13 am - IP Logged

        Yes I can.

        Q1:It it EDUCATIONAL software.

           What did you learn?

        Q2: Number prediction?: Impossible. It has as much abiity to predict numbers as a pencil does.

        Welcome

        DD

          RJOh's avatar - chipmunk
          mid-Ohio
          United States
          Member #9
          March 24, 2001
          19830 Posts
          Offline
          Posted: December 29, 2009, 1:13 pm - IP Logged

          Yes I can.

          Q1:It it EDUCATIONAL software.

             What did you learn?

          Q2: Number prediction?: Impossible. It has as much abiity to predict numbers as a pencil does.

          Welcome

          DD

          ANN's may better serve the user in looking for patterns amidst many data inputs and variables too complex for human evaluation or, in seeking those patterns not easily revealed in much data.

          DD

          I gather from your last post a human evaluating lottery data would probably do better using a pencil than this software.

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

            Avatar

            United States
            Member #83701
            December 13, 2009
            225 Posts
            Offline
            Posted: December 30, 2009, 11:05 am - IP Logged

            SharkyNN is a free program to give the user a visual seat-of-the-pants insight to training and testing an Artificial Neural Network.

            I like the XOR feature for a visual fit and a 3% error stop can be quickly had with few points of input. As opposed to a few years ago, there are many more affordable ANN products for the casual user and user friendliness is greatly improved. ANN's can exceed the human ability in the volume of data processed, but the simplicity and speed of organic decisions has yet to be paralleled through simple five senses observation.

            ANN's may better serve the user in looking for patterns amidst many data inputs and variables too complex for human evaluation or, in seeking those patterns not easily revealed in much data.

            DD

            Neural Networks is really about information representation, the connections of neurons itself is just expressed as a matrice and it's simple vector calculations to run the net.   With lottery numbers, there would be the issue of how you represent each number, would you use a set of six neurons and represent the numbers in binary in which case the neural net essentially makes six 1 or 0 predictions for each number, would you represent each number individually with a neuron thereby avoiding any interdependence between the numbers ie.: 49 neurons to represent a single number drawn, this requires that each of these 49 neurons has suppresive connections to the other 48 or the net will be postulating multiple numbers for each number to be predicted.   Then there's how to represent historical data, neural nets are spatial but historical data is time sequenced also historical data grows so do you grow the net or window the data, growing the net means the learning process is always on the steep end of the curve for certain portions of data.   Most neural nets made by men are limited in the number of layers and in propagation passes, I believe that three layer neural nets are common to allow for abstraction of concepts in the intermediate layer in hopes of accounting for undesired effects of the information representation strategy used.   Back in the 80's I did come up with a scheme to port temporal heuristics such as traditional computer program flow controls and logic into a net but this results in multiple propagations and even oscillations, the idea was that this would serve as a framework for the learning process to start from in hopes that the temporal structure would begin to collapse into more traditional neural nets.     People often think of neural nets as being fast because of the so called "organic" or "natural" nature but that's nothing of the sort, it's because of it's spatial representation and single propagation pass, an analogy would be the analog computers of the 40's and 50's, they would solve complex equations in an instant but take years to configure.   The analog computers used in soviet fighter jets still outperform modern digital computers in the limited roles that they serve.

            To work with neural networks, you just need a linear algebra library so that you can conveniently perform matrice operations.


              United States
              Member #81843
              October 31, 2009
              856 Posts
              Offline
              Posted: December 30, 2009, 12:55 pm - IP Logged

              WoW! right-on JW!

               

              I have been trying to expand my noodle network about this stuff but it is to be honest, evasive. SharkyNN really is well designed software, very clean and with the exception of no manual is very nice. Just another stepping stone for me. Every year I learn a little more about the whole topic. How would you classify NN's and GP (Genetic Programming)?

              This is an honest question, not a test.

              DD


                United States
                Member #81843
                October 31, 2009
                856 Posts
                Offline
                Posted: December 30, 2009, 1:02 pm - IP Logged

                ANN's may better serve the user in looking for patterns amidst many data inputs and variables too complex for human evaluation or, in seeking those patterns not easily revealed in much data.

                DD

                I gather from your last post a human evaluating lottery data would probably do better using a pencil than this software.

                Yep, it's just a pretty thing to look at. The old neural noodlework is hard to beat until there is a Skynet.

                Just when I was getting mine configured, I get a braincloud! 

                DD

                 

                Did you hear the one about the constipated mathematician?

                  Avatar

                  United States
                  Member #83701
                  December 13, 2009
                  225 Posts
                  Offline
                  Posted: December 30, 2009, 2:15 pm - IP Logged

                  WoW! right-on JW!

                   

                  I have been trying to expand my noodle network about this stuff but it is to be honest, evasive. SharkyNN really is well designed software, very clean and with the exception of no manual is very nice. Just another stepping stone for me. Every year I learn a little more about the whole topic. How would you classify NN's and GP (Genetic Programming)?

                  This is an honest question, not a test.

                  DD

                  Genetic programming is really more a matter of automated programming using the concepts of mutation and natural selection, or as is the case unnatural selection.   Basically a mechanism is created for creating diversity, this is usually mutation but it could also be code swapping, regardless a definition of the granularity at which you want the code modified needs to be made, usually in blocks of code but it can extend all the way down to individual commands, there is also the concept of insertion or deletion of code or code blocks and of "switches" where code is inherited but switched off then there is a criteria for survival, the selection process.   Basically, although there is no guiding intent behind the automated programming, the sheer parallelism and speed of a machine running through the possible combinations and evaluating them by the survival criteria offsets the fact that most of the variations will be quite useless.   Also, since there isn't a rationale behind the diversity, out of the box solutions are much more likely to be evaluated and hence incredibly ingenious solutions can be found.   The only similarity behind neural network and genetic programming is that neural networking has the concept of feedback learning where all the connections are strengthened or weakened depending on how far off the result is from what was intended and this represents a form of automated progression just as the mutation and natural selection is a form of automated progression in genetic programming.   I avoid using the term evolving because people assume evolving is towards the better and more complex when it's completely independent of what's better or what's complex, most plants have more complex genome then people and neanderthals may have had more intelligence and certainly more speed and strength, saying a form has evolved means nothing more then that it has changed.


                    United States
                    Member #81843
                    October 31, 2009
                    856 Posts
                    Offline
                    Posted: December 30, 2009, 5:20 pm - IP Logged

                    Awesome JW!

                    Thank you, I was wondering much about the two and not quite sure about what was similar and what was different. I needed more input, and an insight from someone who has a scope of the topic. Now, what do you think about NN's that have some kind of GP feature you can tune to get a better fit? I may have to reread your posts but I think I am keeping up with you. It's not like one can strike up a conversion like this the average publican or pedestrian.

                    DD

                      Avatar

                      United States
                      Member #83701
                      December 13, 2009
                      225 Posts
                      Offline
                      Posted: December 30, 2009, 8:09 pm - IP Logged

                      Awesome JW!

                      Thank you, I was wondering much about the two and not quite sure about what was similar and what was different. I needed more input, and an insight from someone who has a scope of the topic. Now, what do you think about NN's that have some kind of GP feature you can tune to get a better fit? I may have to reread your posts but I think I am keeping up with you. It's not like one can strike up a conversion like this the average publican or pedestrian.

                      DD

                      That's actually a brilliant idea.   With a neural network, the mutation can be applied on a very fine grained basis yet the probability of still having a functional net is very high while with normal GP approaches you need to keep the granularity large in order to assure some functionality.   Should work great.

                        Avatar

                        United States
                        Member #83701
                        December 13, 2009
                        225 Posts
                        Offline
                        Posted: January 1, 2010, 10:39 am - IP Logged

                        The thing to remember is that a neural network and genetic programming doesn't involve theorizing in their search for solutions hence it's a far broader search and more apt to develop superstitions rather than find actual relationships if the data is sufficiently random.   A neural net intended to recognize tanks hiding in foliage wound up basing it's judgement on whether or not the photograph was well lit, it turned out that of the photos used to train the net, the ones without a tank were taken on any old day but the ones where they intentionally hid a tank were only taken on clear sunny days hence the neural net developed the superstition that tanks are only found hiding on sunny clear days.


                          United States
                          Member #81843
                          October 31, 2009
                          856 Posts
                          Offline
                          Posted: January 3, 2010, 4:14 am - IP Logged

                          Been digging around on the net again on AI. It's an annual thing. Looking for content I can follow along and got many repeat lessons. A NN would be great to train and look at data but the feed-forward thing is another animal. Maybe conditional logic is not the way to go.

                          Many likely use a IF-THEN and a noodle network (organic) trying to come to the answer to the THEN in selecting number sets or digits to play. There are many trends to follow in different ways and interactions of trends within trend data. Trends are the IF used for the THEN.

                           

                          Training computers to look at tanks like we do is not the answer. Our visible spectrum is not as discerning to information present as opposed to a CCD- but thermal imaging, or a different view would provide much enhanced perception beyond the human eye's range. The tank ID NN could possibly perform better with information not normally taken into consideration. Maybe we need a different perspective from or with IF-THEN.

                           

                          I heard something interesting on the radio about the sense of smell. It apparently bears a stronger weight than was thought on our combined perceptions and reactions. It was a study of vegans and meat-eaters. The vegans did not react (brainwave) like one would think when smelling meat and seeing a picture of a meat meal. In fact, the results for each consumer were 180 degrees out of what one would think.

                           

                          K

                          DD