=============================================================================
	README file for the example files laser.xxx
=============================================================================


Description:	extended hierarchical Elman network
============	for the task to predict the intensity of a NH3-laser

The task of this partially recurrent network is to predict the light
intensity of an NH3-laser in its chaotic state.  This data was taken
from the Santa Fe Time Series Analysis and Prediction Competition,
time series A. The data is described in more detail there.

Here it suffices to say that the data values were originally integeger
values in the range [0..255] which were scaled linearly by division by
255, to fit them to the range [0..1] of the standard logistic
activation function used in this network.

The single input unit is assigned the laser intensity at the current
time t and the network predicts the laser intensity at time t+1 at its
output unit.

See the user manual for a more detailed description of extended
hierarchical Elman networks and their usage.


Network-Files:	laser.net
==============

This network file contains a trained elman network for the task to
predict the predict the intensity of a NH3-laser as stated above.
The network consists of the following layers
	1 input unit
	8 hidden units in hidden layer 1
	8 context units in context layer 1
	8 hidden units in hidden layer 2
	8 context units in context layer 2
	1 output unit

All feedforward-layers (input, hidden 1, hidden 2, output) are fully
connected, each hidden unit has a fixed connection to its context unit,
each context unit is connected to every hidden unit with the
corresponding layer number, 
each context unit has a fixed recursive connection to itself.

The configuration file for this network is laser.cfg (one 2D display
only).


Pattern-Files:	laser_999.pat
==============	laser_1000.pat

These files contain 999 resp. 1000 training patterns with one input
and one output unit each. They are the scaled data of the laser
intensity prediction time series described above.  The pattern files
differ only in the number of patterns they contain, indicated in the
name of the pattern file.


Hints:
======

The easiest way to create hierarchical Elman networks is with the BIGNET
panel from the info panel. All network parameters can then be
specified in a special Elman network creation panel called
with the respective button in the BIGNET panel.

If you want to train your own Elman network from scratch,
note to set the proper initialization function and initialization
parameters.

Remember to set the update function to JE_Order or JE_Special,
depending on your task (see the SNNS user manual for more details).

You may choose between four different learning functions. The network
shown here was trained for 10000 cycles with JE_BP (Backpropagation
for Jordan and Elman networks) with a learning rate of 0.2.

The behaviour of this network can very nicely be visualized with the
network analyzer tool which can be called from the info panel with the
GUI button as ANALYZER. Then proceed as follows:
Press ON and LINE (so that both buttons are highlighted) from the
buttons at the right.
Press SETUP and choose T-Y graph from the network analyzer setup panel.
Choose the following values for axis, min, max, unit, grid:
	t	0.0,  1000, - ,  10
	y	0.0,  1.0,  18,  10

This specifies the display area to be a time series with y values in
the range [0, 1] and the outputs of neuron 18 for y (the single output
unit of the hierarchical elman network).

Choose m-test: 100 in this network analyzer setup panel to test 100
patterns in a multiple inputs test sequence (You may also choose to test
more or less input patterns.

Finally, press the button M-TEST to test the trained network for the
number of input patterns specified. The predicted time series is then
displayed in the network analyzer tool.


=============================================================================
	End of README file
=============================================================================