HELP TRAININGOPTIONS                     Julian Clinton  Feb 1990


This file describes the options available on the Training Options Menu.

         CONTENTS - (Use <ENTER> g to access required sections)

 -- 'Current network'
 -- 'Current example set'
 -- 'Training cycles'
 -- 'Use current network'
 -- 'Use current example set'
 -- 'Use random selection'
 -- 'Call generator function'
 -- 'Update display'
 -- 'Show target result'


-- 'Current network' --------------------------------------------------         

The variable NN_CURRENT_NET is used to hold the name of the current
neural network. The current net is used as the default value whenever
the name of a network is needed as an input to some function e.g.
training or testing.


-- 'Current example set' ----------------------------------------------

The variable NN_CURRENT_EGS is used to hold the name of the current
network. The current example set is used as the default value whenever
the name of a set is needed as an input to some function e.g. training
or testing.


-- 'Training cycles' --------------------------------------------------

During training, the number of cycles through the example set is defined
by the variable NN_TRAINING_CYCLES. This is an integer.


-- 'Use current network' ----------------------------------------------

In some cases, say if a large number of networks have been defined or
you will only be interested in one network at a time, it is useful to
stop Poplog-Neural from asking which network should be used for a
particular operation. Setting this option to TRUE (held in the variable
NN_USE_CURR_NET), Poplog-Neural will always use the current network
without asking for confirmation.


-- 'Use current example set' ------------------------------------------

In the same way that you can tell Poplog-Neural not to ask which network
to use, you can also prevent Poplog-Neural from asking which example set
to use by setting this option to TRUE. The value is held in the variable
NN_USE_CURR_EGS.


-- 'Use random selection' ---------------------------------------------

During network training, it is often an advantage to present the
training examples in random order since this can help prevent networks
from getting stuck in local minima. When the variable NN_RANDOM_SELECT
is set TRUE, the examples in the example set are presented to the
network for training in random order.


-- 'Call generator function' ------------------------------------------

By setting this option TRUE, Poplog-Neural will call the generator
function in the selected example set before each training session. It
will not call the function if you are only testing the network. This
value is held in the variable NN_CALL_GENFN.


-- 'Update display' ---------------------------------------------------

When Poplog-Neural is running with the graphical user interface, setting
this variable to TRUE will ensure that any windows displaying part of
the network which has just had some operation performed on it will be
updated automatically. The value of this option is held in the variable
NN_UPDATE_WINS.


-- 'Show target result' -----------------------------------------------

Once an example set has been tested on a network, a set of results will
be available for browsing. If the variable NN_SHOW_TARG is set TRUE then
the target results (if there are any) will be displayed along with the
actual results.


--- Copyright Integral Solutions Ltd. 1990. All rights reserved. ---