The New Network dialog is used for the creation and growth of a new neural network from the values in the Grid. During optimization a new network is created every time the period of cycles or seconds completes. The number of hidden nodes in each layer that is checked grows from minimum to maximum. Each network is checked by training it for a short period. The network that produces the lowest error is selected for further training and validating.
Frameworks can be created without connections if the Network Editor is to be used.
 Growth rate
A network is produced when the cycles or seconds elapses until the optimum network is found.
|
|
|
 Connect layers
If checked all connections are created automatically. If not checked the optimum number of nodes will be created but not connected together. This is so that the network editor can be used to produce complex configurations.
|
|
|
 Set freezes
Opens the Freezes dialog. Set the node types to freeze, the initial range, the direction of freezes and the freeze decrement conditions.
|
|
|
 Set trimming
Opens the trimming dialog to set up weight pruning, unimportant input node deleting and highest error example row excluding.
|
|
|
 Set cloning
Sets hidden node cloning. If learning is not progressing the hidden nodes are tested. A node with a high net input that is contributing to the learning process is cloned.
|
|
|
 Input layer
The number of nodes in the input layer is determined by the number of input columns in the grid.
|
|
|
 Create hidden layers
Must be checked to create or grow any hidden layers.
|
|
|
 Grow hidden layers
If the layer number is checked the layer is grown from the minimum to the maximum number of nodes.
|
|
|
 Output layer
The number of nodes in the output layer is determined by the number of output columns in the grid.
|
|
|
 OK button
Press to accept all the settings.
|
|
|
 Cancel button
Press to reject all the settings.
|
|
|
