*********** Sonar.nns is a NeuNet Pro Sample File ***********

This file contains special properties that allow NeuNet Pro to recognize it
as authorized NeuNet sample data.  Anyone using the unlicensed version
of NeuNet Pro is welcome to experiment with this sample data.
Please do not modify this file, or it will lose its status as authorized sample data.

For further information about Neunet Pro and additional sample data,
please visit the NeuNet Pro website at http://www.cormactech.com/neunet

All of this data has been collected from publicly available sources.
CorMac Technologies Inc. does not guarantee the accuracy of the data.
This data is intended solely for experimental purposes.


    ***************  More about the Sonar Data *******************


   SONAR.NNS is a sample file that contains 207 underwater echos.
   These echos resulted from bouncing a PING sound off either a rock or a metal cylinder.
   The purpose is to predict whether the echo comes from rock or metal.

   Each pattern consists of 30 input numbers representing the volume of sound returned within
   each of 30 frequency bands.  All 207 patterns have been shuffled to random order.

   Trained human experts were able to make correct predictions on 88% to 97% of the pings.

   NeuNet Pro in SFAM mode seems able to quickly learn 100% of the training set,
   but performance on the test set is only 70% to 75% accurate.  NEUNET's ability to
   quickly lock-in on the training set shows there is a clear pattern to the data.
   The lack-luster performance on the test set is probably due to insufficient training patterns.
   If there were more training patterns, NEU-NET might be able to match or exceed the human
   experts.

   The researchers placed the target object at many different angles, so that there are
   several sub-classes of the two main classes.  If there were more patterns, NEU-NET
   would be able to get a better look at the various sub-catagories.  There may be sub-classes
   in the testing set that are not even seen in the training set.

   Perhaps more (or fewer ?) frequency bands would be helpful.
   P.S.  Try increasing the vigilence to .70.  Performance on the Testing Set is 85% to 90% !