Feedforward Neural Network Initial Settings
Dialog Box
Overview
A dialog box used to view and edit the basic parameters of new feedforward
neural networks. Accessed through the Preferences
Panel.
Preferences
- Hidden Unit Coef: The coefficient that is used to multiply the
value that is used to calculate the automatically calculated number of hidden
units in a network. A value between 0 and 1 will decrease the number of hidden
units and a value above 1 will increase it.
- Learning Rate: The learning rate of the neural network. This affects
the number of iterations needed for the network to converge as well as the
probability of it falling into a local minimum in error space.
- Momentum: The momentum of the neural network. This affects the
number of iterations needed for the network to converge as well as the probability
of it falling into a local minimum in error space.
- Min Initial Weight: The minimum values of weights between network
units when these weights are initially randomly generated.
- Max Initial Weight: The maximum values of weights between network
units when these weights are initially randomly generated.
Buttons
Warnings
- These parameters are used when new neural networks are generated during
training. They do not affect existing trained neural networks used during
classification.
- All values entered must be 0 or greater.
Screen Shot