readme

人智算法基本程序

					 Perceptron Program

Welcome to the wonderful world of the Perceptron C++ directory.  This
program demonstrates the perceptron learning algorithm with a very
simple perceptron network, a logic threshold unit. It is meant to be a
companion to chapter 5 of the "Artificial Intelligence" textbook. This
code was written by Jon Monsarrat, [e-mail].

Four sections describe the program:

1. The problem
2. Building the program
3. Running the program
4. The solution, and how to read the source code

----------------------------------------------------------------------
1. The problem

The book goes into quite some detail about the problem, but here it is
in a nutshell.

We have some function (let's pretend it's unknown or takes a long time
to compute). We want a perceptron network to simulate the function.
Unfortunately, a perceptron network must be configured with a bunch
of weights that tell the network how to work. This program solves the
problem by learning the weights.
----------------------------------------------------------------------
2. Building the program

The source code in this directory is complete and does not depend upon
the support directory or any other source code from the textbook. It
only uses standard and libraries which should come any C++ compiler.

You can build this program in UNIX by simply typing

      make

You may need to edit the Makefile to change the name of the compiler,
if you do not have a compiler named "CC".

You will need to recompile the program to try out new functions. This
is because function are written in C++ and must be themselves compiled.
----------------------------------------------------------------------
3. Running the program

Running the program is very simple. Just type

   perceptron -or

To try to learn the two bit "or" function. You can also type

   perceptron -and     for "and"
   perceptron -xor     for "exclusive or"

These three functions come with the perceptron program, but you are
welcome to make up your own. You'll need to read the code and
recompile the program to do this.
----------------------------------------------------------------------
4. The solution, and how to read the source code

This solution doesn't have much to do with the Lisp version given
in the textbook. It's the exact same algorithm, but I chose to
represent functions in a more C++-like way.

   In the Lisp program, a function is a list of input-output
   combinations. You never need to generate all combinations of inputs
   because all combinations of inputs are listed explicitly in the function.
   It is possible to have a function where only some of the input
   values are valid.

   In the C++ program, a function is a real C++ function. To get
   all combinations of inputs, they must be generated. The perceptron
   program does this. It is not possible to have a partially defined
   function which does not return values for some inputs.
   
The classes are:

   Perceptron    -- Holds the entire perceptron, and runs the learning algorithm
   Function      -- Holds a single function, which is a pointer to a real C++
                    function.

There are only 2 source files in this directory:

   Perceptron.C    The Perceptron class
   Function.C      The Function class

----------------------------------------------------------------------