|
Layer Definition tab (Network Definition dialog box) |
|
Use this tab to specify the details of each layer of a neural network.
Select from this list the network layer you want to modify.
General
Specifies the number of neurons for the currently selected layer. Type the numbers of neurons you want for this layer.
 | You cannot specify a number of neurons for the output layer because this number is indicated by the number of outputs for the underlying model. |
Refer to Data and System Models for additional information on model outputs.
Specifies the activation function for all neurons in the currently selected layer. Select one of the listed functions.
| If selecting a function other than Identity, Logistic, Arctangent, or Hyperbolic Tangent from this list, you must make sure that the derivative and the extent of that function are properly defined. |
Refer to Derivative - F'(x) on the User Function Edit dialog box for additional information on this subject.
Displays the Activation Function Scale dialog box (global neural networks) or the Local Network Output Scaling dialog box (local neural networks) to specify scaling on the outputs of this layer. Click this button if you want to scale the outputs of this layer to match the extent of the data in the underlying model.
| This option is only available when the currently selected layer is the output layer. |
Local neural networks scale their output to the extent of the neighborhood at hand by default.
Learning
Indicates the fraction of the partial derivative of the network error function to use for adjusting the neural connection weights. Type a value greater than zero and less than or equal to 1.
 | You can modify network learning rates anytime after network definition. |
Refer to the Learning Rates tab (Training Options dialog box) for additional information on this subject.
| Refer to neural weights for additional information in neural connection weights. |
Refer to backpropagation for additional information on learning rates.
Indicates a step by which to adjust the learning rate after every waiting period. Setting this box to a value less than 1 will diminish the learning rate by the specified factor after some waiting period of training.
| A value of 1 in this box indicates that the learning rates are not to be automatically modified during training. |
Note that the learning step value is not the same as the step size value, used elsewhere in the literature to denote what we call the learning rate value.
Refer to Step Waiting Period (General tab of the Training Options dialog box) for additional information on the step waiting period.
Click this button to copy the currently specified learning rate and learning step to all previous layers.
| This command is available when the currently selected layer is other than the first hidden layer. |
Click this button to copy the currently specified learning rate and learning step to all following layers.
| This command is available when the currently selected layer is other than the output layer. |
Momentum Base
Indicates the momentum rate is applied on the previous total learning rate change.
Indicates the momentum rate is applied on the previous partial derivative only.
| These options have no effect when the momentum rate for the neurons in this layer is set to zero. |
Refer to backpropagation for additional information on momentum base.
Momentum
Indicates the fraction of momentum to use for adjusting the neural connection weights. Type a value greater than or equal to zero and less than or equal to 1.
| A value of zero in this box indicates that momentum is not used in modifying connection weights during training. |
Refer to backpropagation for additional information on momentum rates.
Indicates a step by which to adjust the momentum rate after every waiting period. Setting this box to a value less than 1 will diminish the momentum rate by the specified factor after some waiting period of training.
| A value of 1 in this box indicates that momentum rates are not automatically modified during training. |
Refer to Step Waiting Period (General tab (Training Options dialog box)) for additional information on the step waiting period.
Click this button to copy the currently specified momentum rate and learning step to all previous layers.
| This command is available when the currently selected layer is other than the first hidden layer. |
Click this button to copy the currently specified momentum rate and learning step to all following layers.
| This command is available when the currently selected layer is other than the output layer. |
Sponsored
Try Predictive Systems Lab
Need Help?
Contact support