General Analyser


This  HyperNext Creator project implements a neural network analyser that is easy to set up. 
The user just needs to specify the number of inputs and outputs, along with their data format and the Analyser will automatically create the neural network. 
The data format is very simple and the system automatically normalises and preprocesses the data. 
Once set up the user can then either use the default learning paramaters or else specify and experiment with their own.




Important
In order for the project to run, the BP1 plugin must be activated in the preferences located on the Edit menu bar.

Quick Start

The best place to start is with the XOR problem. 
Simply load the XOR project from the Start Screen and once loaded it will take you to the Data Screen. 
The Data Screen allows data files to be readin, shuffled and the training/testing ratio set after which the system can be trained using the Training Screen. 
Once trained the system can be queried and the neural network's analysis studied.




Help


Many of the screens have help available that can be accessed by simply clicking a button.




Data Format


The expected format of a data file is very simple. 
Each line in the file represents one sample for the system to learn or query. 
For example, 100 lines equals 100 samples.

Each line or sample is made up of the inputs and outpiut values separated by commas. 
If a value is unknown or missing the system will simply treat it as a "don't know".



Example,

The XOR problem has four samples(lines) made up from 2 inputs followed by a single output as shown below.
         0,0,0
         0,1,1
         1,0,1
         1,1,0





Training and Testing


There are several example problem domains of which the most important are "AutoPrices"and "Heart" that both have realworld data sets.

 
When training the neural network it is recommend that first time users should experiment with the smaller  datasets as the larger ones can be quite time consuming. The setup screen allows a dataset file to be loaded, shuffled and then divided into training and testing sections.For instance, a 40% value indicates that 40% of the loaded file will be used for training and the remaining 60% used for testing.

 The neural network can also be tested on the training data.



The neural network matrix or memory can be saved at any time and later recalled to receive further training or to answer a query.
Analyser Project itself



The project can be freely modified and shows various aspects of using HyperNext Creator and the BP1 neural network plugin. 

The code is not optimised and could be greatly improved, especially in coping with manual data input.



Note
  The project as set up can cope with dataset files have both Macintosh or UNIX line endings.

  

When training the neural network the Escape key can be used to abort the training but on very slow machines there can be a substantial delay between pressing it and the training aborting.

