Introduction - If you have any usage issues, please Google them yourself
This program is a demonstration program of BP algorithm, and the levenberg-marquardt algorithm is of practical value.
I. network training
The default state of the program is the sample training state, and how to train the network in the sample training state is described.
1. System accuracy:
Define the system target accuracy and define the accuracy of network training error according to the requirements. The error formula is the sum of the mean variance of the output layer node and the actual network output of the training network.
Maximum training frequency:
The default is 10000 times. According to the need, if the maximum training frequency network fails to reach the target accuracy, the program exits.
3. The step:
The default is 0.01, and the manual setting is generally not required because of the variable step length algorithm.
4. Number of input layer:
The number of nodes of the input layer neuron of the artificial neural network.
5. Number of hidden layers:
The number of nodes of the underlying layer of neurons in the artificial neural network.
6. Number of output layer:
Number of neurons in the output layer of the artificial neural network.
7. Training algorithm:
The levenberg-marquardt algorithm is strongly recommended, which is stable after testing.
8. Activation function:
Different network activation functions perform different performance, and can be selected according to the actual situation.
9. Processing of sample data:
Because the program does not have the normalization function, the sample data used for training must first be normalized to be trained.