When you have the program SNNS running, you can work with it. You must see the snns-manager display. You can manage all the application of the simulator. First of all, you have to load the ANN. To do this click on "FILE" button, and load the encoder network: answer yes to the question of loading the configuration file. It will appear the display of the network, something like this:
Then select the file of patterns for training the encoder network. Just press on PAT button and select encoder.
Now that we have the ANN and the patterns
loaded, we can train it. For it we have to get the CONTROL display
from the snns-manager. Press on control button. Set the number
of cycles (sweep) needed to
train the network. At the beginning you can try with this cases:
Learning-Function Learning-Parameters Cycles Std.-Backpropagation 2.0 750 Backpropagation with Momentum 0.8 0.6 0.1 300 Quickprop 0.2 1.75 0.0001 75 Rprop 0.2 75You can open the GRAPH window to see how the error goes down at each cycle (sweep). To start training the network first we have to stablish the initial weights. To do this press on INIT button. Then press on the ALL button to train all the patterns in each sweep (block adaptive method).
You can select other examples given by the simulator and see different networks than backpropagation one.
You can repeat this process for different aleatory initial weights. You will see that the final error will be different, just because you started the training process in different points of the error surface.