Table of Contents
Back

Loading a training file

Before you can set the training parameters or train the loaded network, you must have a training file loaded.

To load the training file, select Load Training File from the Train menu and browse to find the required file.

The file can have any extension, but the convention for training files is *.trn, and it must be of the required format.

The only restriction is that the path cannot contain a ).



Back

Specification of Training Parameters

Learning rate:
A floating point number greater than or equal to zero.
If the learning rate is too high the weights will oscillate with a large amplitude. Conversely a small learning rate will cause the weights too converge slowly.

Momentum:
A floating point number between zero and one.
Momentum is introduced to accelerate learning when the learning rate is small, and prevent oscillation of the weight when the learning rate is high.

Epochs:
The number of learning cycles during which one or more training examples are propagated through the network.

Terminating Error:
A floating point number greater than or eqal to zero.
This is the error value at which you want the training of the network to cease.

LWF:
Check this box to enable Learning With Forgetting.
During learning with forgetting a decay coefficient is included which causes unnecessary connection weights to decrease to almost zero. This way only the significant connections within the network will be activated.

Training Mode:
You can choose to train in Batch or Pattern mode.
Batch mode:
The changes in weights are calculated after all or some of the training examples have been passed through the network. The changes are accumulated and applied only after the entire training set has been passed through the network.
Pattern Mode:
The changes in weights are calculated and applied after each training example has been propagated through the network. If this mode is used the training data must be shuffled to avoid the network adjusting the weights for one output.



Back

Viewing the training file

You can press the View Training File button to view your training file, or alternatively select View Training File from the Train menu.

Note: You need to associate files of type *.trn with a text editor in Windows Explorer in order to view the files through the FuzzyCOPE GUI.



Back

Training in MLP

Press the Train button to commence training, or select Back Propogation from the Train menu.

During training the plot of the Root Mean Square (RMS) error is shown along with the value of the RMS error and the number of epochs completed.

When the RMS error reaches the specified terminating error the training is completed.

The number of epochs specified is sometimes reached before the terminating error. In this case the number of epochs can be readjusted if required and the training started again.



Back

Saving the trained network

After training is completed the network can be saved.

To do this, select Save As from the Network menu.

Give the file a name that is different from the original network file.

The file can have any extension, but the convention is for network files to have the extension *.wgt.

The only restriction is that the path cannot contain a ).


This page is maintained by Melanie Middlemiss mmiddlemiss@infoscience.otago.ac.nz
Last modified on: 4/2/98.