readme.txt

标准的GP源代码,由Andy Singleton维护

*************************************************************************
GPQUICK
A simple Genetic Programming system in C++
Version 2, released 2/12/94
*************************************************************************

 Copyright Andy Singleton, 1993,1994
 This code is released for non-commercial use only
 For questions or upgrades contact:
    Andy Singleton
    Creation Mechanics, Inc.
    PO Box 248, Peterborough, NH 03458
    Internet: [e-mail]
    Compuserve: 73313,757
    Phone: [phone]

 Further documentation appears in the associated Byte Magazine article
 "Genetic Programming with C++", February, 1994

*************************************************************************
Packing List
*************************************************************************
Source Files:
     CHROME.H        // Core object definitions and GP code
     CHROME.CPP
     PRIMITIV.H              // Standard GP primitive functions
     PRIMITIV.CPP
     SELECTOR.H       // A utility GA fitness selector object
     SELECTOR.CPP
     PCH.H           // Standard headers for "precompiled headers"

     SSPROB.CPP      // A "Simple Symbolic Regression" sample problem
					 // Contains "main" and most user modified sections

     ANTPROB.CPP     // The "Artificial Ant" sample problem
     SANTAFE.TRL     // A trail of food required by the artificial ant

Other Files:
     README.TXT      // This file

The DOS package may also include
     GPUICK.EXE                  // DOS executable of SSPROB
     GPQUICK.PRJ                 // Borland project file

Optional Files (at McGraw Hill's option)
     BYTEGP.PS,TXT,WRI      // The associated magazine article
			    // It is a general introduction to GP,
			    // but it contains some helpful technical doc
*************************************************************************
Instructions
*************************************************************************

To Make:
     Compile CHROME.CPP, PRIMITIV.CPP, SELECTOR.CPP and SSPROB.CPP
	 Link with SSPROB as the main

     You can substitute ANTPROB or any compatible problem for SSPROB

The code is ANSI C++, but you may encounter a few portability issues.
It was originally compiled with 16 bit Borland C++ under DOS and MS Windows.
Configure for UNIX by uncommenting "#define UNIX" in file PCH.H
I will share alterations for other compilers and OS as users provide them.

Runtime Behavior:
     The sample problem, SSPROB, will attempt to evolve an expression
equivalent to John*(George-Paul)+2.5*Ringo.  Every 5 seconds or 1000
expression evaluations, it will list to the console the best expression
found so far. It will run for 80,000 evaluations, until it finds the
right answer, or (under DOS) until you hit a key.

	The sample problem ANTPROB will attempt to evolve an artificial ant
which can traverse the trail of food.  The ant moves across a 32x32 grid,
and it can sense food immediately ahead, move forward, or turn.  As the
problem runs, it will display the best ant by the amount of food found.
After the ant finds all of the food (89 pieces on the Santa Fe trail), or
after 25,000 generates, the problem terminates and prints the path of the
best ant to file "RESULT.TXT".  This is a replication of the problem
described by John Koza in "A-Life II" and "Genetic Programming".

Exercises:
	 You can evolve more interesting or useful behavior by editing the problems
	 provide to create your own GP problems.

Please: If you develop an interesting problem, send me a copy.

Technical Details:
     GPQUICK as provided here uses a steady state GA, tournament selection
three types of mutation, and subtree crossover.
    It uses tournament selection with a tournament size of six, and a
"kill tournament" size of 2.  It uses a node mutation (switching a node
for another node of the same arity), constant mutation (small adjustments
to constant values) and shrink mutation (promote a sub-subtree to replace a
subtree) 10% of the time each, a copy with reevaluation on new cases 10% of
the time, and subtree crossover 70% of the time.  The default population
size is 2000.
	GPQUICK also supports both crossover and mutation "annealing" operations,
which replace a parent instead of reproducing into a target site.  These can
make a small population look big, improve hill climbing, and reduce convergence.
    The GP representation is a linear array of 2 byte nodes, suggested as
"linear prefix jump table" by Mike Keith.  This version also contains the
FASTEVAL option for fast eval calls.
    The function argument/return values can be any floating point number.
Numeric constants must be integers from -128 to +127.  Unless you adjust the
constants with hill-climbing methods, the precise constant values are not 
important.
    GPQUICK features an elegant object architecture with function
(Function), program (Chrome), GA (Pop) and problem (Problem) classes.

*************************************************************************
Improvements in version 2
*************************************************************************
In atonement for past sins, I am offering an updated version of GPQUICK.
The new version of GPQUICK has the following improvements:

Several bug fixes
Unified UNIX/DOS/ANSI source
More extensible object structure
FitnessValue object for carrying problem specific fitness information
    (in response to popular demand from the GP mailing list)
Streamlined Problem files
Improved function constructor
Standard primitive library, including a working IF and IFLTE.
Annealing
2 more mutations
Artificial Ant problem

*************************************************************************
For More GP fun
*************************************************************************

DOODLE GARDEN, a GP toy and "Grow your own" screen saver, is software
which allows you to experiment interactively with a wide variety of
primitives and genetic operators.  Doodle Garden is an MS Windows
package which allows the user to evolve programs that draw pictures,
or "doodles" by mutating and crossing other doodles.  Doodle Garden
is available from Creation Mechanics for $39.00.


Creation Mechanics will provide a professional GP system called GP-GIM
(Genetic Programming - Genetic Induction Machine).  If you use the 
evaluation macros, GPQUICK problems should be source code compatible
with GP-GIM problems.

GP-GIM includes:
* Built in Problems for regression, classification, and selection from data
* Problems plug in at runtime as DLL's
* Rich set of primitive functions, with many add in function packs
* Numerous GA options, operators, and over 50 parameters
* Meta-GA for automated primitive selection, parameter selection and
  generalization
* Subroutine defining structures, constrained crossover, fixed roots
* Interactive user interface
* Runtime libraries for using and distributing the resulting code
* High speed implementation
* "Desktop Supercomputing" architecture linking a network of PC's or 
workstations into a parallel GP machine.  Rack mounted clusters are also
available.

For more information, contact Creation Mechanics at the phone number
and address listed on the top of this file.