Kohonen Learning

Top  Previous  Next

 

Use this module to train Kohonen networks.   Start training from the Run Menu.  Training continues until the number of epochs specified in the Kohonen Architecture module is reached.

 

The module allows you to view graphics and statistics for the training patterns as learning progresses.

 

Training Graphics

NeuroShell 2 allows you to view either a pie or bar chart of the network's output category distribution after training is completed.

 

_bm42Distribution by Category - Bar

This module creates a var chart that displays the number of patterns that appear in each of the output categories for the Kohonen network.  It will take two training epochs before the graph displays correctly.

 

_bm56Distribution by Category - Pie

this module is the same as the bar chart, except the categories are displayed in the "pie" format. It will take two training epochs before the graph displays correctly.

 

Statistics

By clicking on the appropriate box, you can display a variety of statistics for the Training Patterns as learning progresses.

 

Training Patterns

Learning events completed: This box displays the number of learning events or patterns that have been propagated through the network.

 

Learning epochs left to go: This box displays the number of epochs that remain before learning is completed.  An epoch is the entire set of training patterns.  You set the number of epochs required to train the network in the Kohonen Architecture module.

 

Current learning rate: This box displays the learning rate that is being used to train the network during this epoch.  NeuroShell 2 automatically reduces learning rate as learning progresses.

 

Current neighborhood size: This box displays the neighborhood size that is being used to train the network during this epoch.  NeuroShell 2 automatically reduces neighborhood size as learning progresses.

 

Unused output categories: This box displays the number of output categories which the network has not used when classifying training patterns.  You set the maximum number of output categories when enter the number of neurons in the output layer in the Kohonen Architecture module.

 

Note:  If you change the number of input or output neurons, you must retrain the network.  You cannot continue training the original network.

 

Use the Run Menu's Set Random Number Seed option if you want to either reproduce a training sequence or change a training sequence.  If you are trying to reproduce a training sequence, type in the same random number seed each time before you start training.  If you are trying to change the training sequence (and perhaps change your results), type in a different random number seed.  Random number seeds range from 0 to 32767.  The default is 1.