-
Hidden Layer Size
-
Prediction Type [Hidden Anticipation (HA) or Auto-Association (AA, Input Prediction)]
-
Prediction [on/off]
-
Error Anticipation [on/off]
-
Momentum Values [decimals 0 through 1]
The networkRuns() function allows you to execute multiple runs with different testing parameters in a single command. Since it is a nested for loop, all you need to do is edit the content of lists which are cycled through. Example: You can test the effect of the hidden layer size on a network with HA on by setting hiddens = [10,12,13] to test hiddenlayer sizes of 10, 12, and 13.
def networkRuns():
-
hiddens = [4,6,8,10]
-
modes = ['HA', 'AA']
-
momentums = [0.0,.3,.6,.9]
-
EA = ['on', 'off']
# EDIT PARAMETERS HERE:
-
hiddens = [8]
-
modes = ['HA']
-
momentums = [0]
-
EA = ['on']
-
moviedir = 'blah'
After editing the "Edit Parameters" simply type
-
>>networkRuns()
-
in the python shell.
In running a given Network configuration, the program will save Gnuplottable files for TSS Error, Prediction Deltas, Output Deltas, HiddenUnit Deltas, and Prediction & Output Error for a given pattern. These are plottable through the gplotXOR.py file in the autoassoc directory.
The program will also save files for PCA in order to use the movie function.