Loading ...
Sorry, an error occurred while loading the content.
 

How to Predict Out of Range Output

Expand Messages
  • patrick_mcdonnell9
    I have downloaded a trial version of EasyNN-plus and I am having a hard time getting the type of response that I am looking for. I am trying to use the NN to
    Message 1 of 2 , Mar 1, 2005
      I have downloaded a trial version of EasyNN-plus and I am having a
      hard time getting the type of response that I am looking for.

      I am trying to use the NN to 'learn' a function based on observed
      responses with the hope that I can then interrogate the NN to find a
      region of the input space with expected good response.
      Unfortunately, I cant find a way to have the NN predict an output
      value outside the range of what it has already observed in the
      training set. I assume this is due to the scaling that is employed,
      but if I cant predict responses that are higher/lower that the
      observed max/min in the training set it pretty much defeats my
      purpose.

      Does anyone know a way around this? If I know where in the input
      space to find min / max response I would interrogate that region for
      the training set.

      Thanks

      Patrick
    • steve@easynn.com
      ... Neural networks are useless at extrapolating. EasyNN-plus makes no attempt to predict outputs outside the training range. Some other applications try to
      Message 2 of 2 , Mar 1, 2005
        On Tue, 01 Mar 2005 16:08:33 -0000, you wrote:

        >
        >Does anyone know a way around this? If I know where in the input
        >space to find min / max response I would interrogate that region for
        >the training set.
        >
        >Thanks
        >
        >Patrick
        >

        Neural networks are useless at extrapolating. EasyNN-plus makes no
        attempt to predict outputs outside the training range. Some other
        applications try to extrapolate but are rarely successful.

        There are some possibilities. The obvious one is to increase the
        training range. This is sometimes possible using Tools > Extend
        Learning... but can produce unexpected results. If the observed
        responses can be ordered you may find that Action > Forecast... helps.


        --
        Steve Wolstenholme Neural Planner Software

        EasyNN-plus. The easy way to build neural networks.
        http://www.easynn.com
      Your message has been successfully submitted and would be delivered to recipients shortly.