|
Training tab (Training Options dialog box) |
|
Use this tab to specify the options for training local neural network.
Training Options
Indicates the maximum number of times that this algorithm will try to minimize the approximation error if previous tries are unsuccessful. Type in this box the maximum number of times you want to allow this algorithm to try approximating the underlying function.
 | The maximum number of tries is limited to 32,767 |
Specifies the maximum number of epochs that make up a try. Type in this box the maximum number of epochs you want to allow for each try.
| The maximum number of epochs is limited to 32,767. |
Specifies the normalized mean square error (NMSE) value at which network training stops. Type in this box the NMSE to which you want to approximate the underlying function, or leave the error as zero to let network training approximate the function as much as possible.
| This error is interpreted in normalized form. |
Check this box if you want to specify a maximum period of time that a local neural network can train for at each prediction point, and type in the box immediately below this check box the number of milliseconds (in the range [50, 30000] -- 1/20th of a second to 30 seconds) you want to allow this neural network to train for at each prediction point.
| It is recommended you specify a time out if you want to carry tests within a reasonable amount of time; otherwise, the local neural network will train until it minimizes the error to the specified Stop Error (NMSE) or trains for the specified number of Maximum Tries and Max. Try Epochs without reaching a new low, which can be a lengthily process. |
Shock Options
Check this box to shock network weights after every unsuccessful try.
| This option is not available when Maximum Tries is set to one. |
Specifies the amount by which to alter the network weights. Type in this box a floating-point number greater than zero and smaller than or equal to one.
Refer to Shock Factor (Network Weights Shock dialog box) for information on the amount by which network weights will change when shocked.
Sponsored
Try Predictive Systems Lab
Need Help?
Contact support