神经网络的C++实现

There are two parts to this directory 1) Training by backprop 2) The GRID1 user interface for testing Both source and exe are provided 1) TRAINING BY BACKPROP. Usage: bprop TrainingFile ParmFile WeightFile Parm file format: Temperature); ETA - learning rate ALPHA - momentum MAXITER - max iteration ERRTOL - error tolerence for convergence NumLayers - number of layers N(input) - Number of neurons in input layer N(hidden1) - Number of neurons in 1st hidden layer N(hidden2) - Number of neurons in 2nd hidden layer (if present) . . . N(hiddenK) - Number of neurons in Kth hidden layer (if present) N(output) - Number of neurons in output hidden layer Training file format: NumPatterns - Number of patterns in training set I0 I1 ... In D - hex byte input for each input / desire value for pattern 1 I0 I1 ... In D - hex byte input for each input / desire value for pattern 2 . . . . . . . . . . . . I0 I1 ... In D - hex byte input for each input / desire value for pattern P Example usage may be found in the command files provided to train networks on the sample training data. The Samples likewise provide examples of parm and training files. Note that after training a .GBL file must be created so that GRID1 will be able to access the trained weights and network parameters. (For the sample data these files are provided with a .gbx extension. After training rename the .GBX file giving it a .GBL extension. 2) GRID1 To run the program it is only necessary to type GRID1 <ENTER> on the command line.