home *** CD-ROM | disk | FTP | other *** search
-
- This package contains all the functions necessary to generate a
- Neural network which you can train and use in your programs. The
- Neural network has an input layer, two hidden layers, and an ouput layer.
- The network is feedforward and fully connected. You may specify an size for
- each layer at run-time. This code is Public Domain and anyone can use it in
- any type of program.
-
- This is version 2 of the C++ Neural network code. The only modifications
- made were changing from <stdio.h> I/O to <iostream.h> I/O. This allows
- the code to be compiled under GCC2. Also I removed the dependency on the
- File class from libg++. The code still has a dependency on a String class
- but it can be removed by changing the prototypes of the following three
- functions:
- Neural_network (String& filename, ...);
- read_weights (String& filename, ...);
- save_weights (String& filename, ...);
-
- Just change String& to char * in Neural_network.h and Neural_network.cc and
- recompile. I only made them String& because GCC2 seems to have trouble
- passing a String as a char * reliably. The variables 'filename' are used
- as char *'s in each function.
-
- The following files are included in this package:
- C++ files
- Neural_network.cc --> C++ Neural_network class code
- Neural_network.h --> C++ Neural_network header file
- xor_dbd.cc --> Example of how to solve the XOR problem using the
- delta bar delta rule.
- xor_bp.cc --> Example of how to solve the XOR problem using
- straight back propagation.
- weights.xor --> Initialized weights that are read in by the XOR programs
- because the gcc,gpp rand () function generates a new
- sequence of random numbers everytime the program is
- run. This file makes sure every XOR training session
- starts at the exact same place so you can see how
- changing learning parameters affects the rate of
- convergence.
-
- Document files
- Readme_V2.1st --> This file
- Neural_network.doc --> Explanation of each Neural network function.
-
- I would like to here from anyone using this code on its performance or any
- improvements or options that should be added.
-
- You can email me at
-
- anstey@sun.soe.clarkson.edu
-
-
-