Overview
Neural Network
This section shows the structure of neural networks, can be seen from the structure of FIG network has three hidden layers, the number of neurons were 9, 8, 7
Algorithms
This section shows the training algorithm used by the network, it can be seen
Division the Data : The network uses the randomly divided data set into training set, validation set, test set
Training : The network using Levenberg-Marquardt algorithm for training
The Performance : The network error algorithm using the mean square error
Calculations : Save the network mex format
Progress
Epoch : The network allows the maximum number of iterations of 1000, five times the actual iteration
Time : 3 seconds long when the network training
Performance : maximum error of the network is 0.475, 0.001 target error, the actual error is 0.000520, can be viewed in detail in Plots in Performance
Gradient : maximum gradient of the network is 1.91, the threshold value of the gradient , the actual gradient of 0.033. Plots can be viewed in detail in the Training State in
MU : The minimum damping factor of the network in the Levenberg-Marquardt algorithm used is 0.001, the threshold value is , actual value , Mu larger value means better convergence effect. Plots can be viewed in detail in the Training State in
Checks Validation : generalization inspection standards of the network, the actual value of 0 indicates an error in the training process continues to reduce, if not six consecutive training error is reduced, the end of the training mission. Plots can be viewed in detail in the Training State in
Plots
The Performance : This error transform network training process visualization
Training State visualization of the network training process the gradient, Mu and generalization conversion factor information such as:
Regression : The network training set, validation set, the ability of regression test suite of visualization
The Interval Plot : scale the abscissa of FIG.