Examples of the data used in my project:

The Letter A in .nn format:

00000000
00111000
01000100
01000100
01111100
01000100
01000100
01000100

A in .dat format (high contrast, real values from 0 - 1). 20 x 20.

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 1 1 1 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 1 1 1 1 1 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0


This 20 x 20 matrix is actually used as a 400 item vector in NevProp.

.net files contain 11 or more of these vectors (there are 11 stimuli), along with data identifying the correct output. I have not included a .net file, as they can be quite large.

Results are displayed as follows:

######################### Welcome to NevProp #########################
NevProp started on Sun Jun 1 22:49:08 1997

Parameters read from file "set7.net"...

TRAINING set patterns: 44
TESTING set patterns: 11

InputUnits: 400 -- HiddenUnits: 120 -- OutputUnits: 11

SEED for initial random weights=61972; Using lrand48(),srand48().

IBMorDOS 0 UseQuickProp 1 EpochWiseUpdate 1
BestByCindex 0 MinEpochs 200 BeyondBestEpoch 2
WtRange 0.01 HyperErr 1 SigmoidPrimeOffset 0.1
Epsilon 0.1 SplitEpsilon 1 Momentum 0.1
Decay -0.001 ScoreThreshold 0.1
MaxFactor 1.75 ModeSwitchThreshold 0
*--------------------------------------------------------------------*
Epoch 0:
TRAINING: 0.00 %correct ; RMSErr=0.49890
TESTING: 0.00 %correct ; RMSErr=0.49888
*--------------------------------------------------------------------*
Epoch 5: Did QuickProp on 100.00 %, GradDesc on 0.00 % of weights.
TRAINING: 0.00 %correct ; RMSErr=0.29042
TESTING: 0.00 %correct ; RMSErr=0.29055
*--------------------------------------------------------------------*
Epoch 10: Did QuickProp on 100.00 %, GradDesc on 0.00 % of weights.
TRAINING: 0.00 %correct ; RMSErr=0.29018
TESTING: 0.00 %correct ; RMSErr=0.29041

... and so on.

I then entered the Testing %correct value into Excel and created the fine graphs you'll find on the Charts & Tables page.


Back to Report | Back to AI