This repository implements a conventional algorithm and demonstrates its own capabilities with sample programs of varying complexity. "test.c" is a minimal demonstration of the genetic algorithm; individuals are strings of numeric characters whose fitness is simply the value of the number they represent. The executable "main-Ant" finds solutions to the artificial ant problem (as described by John R. Koza in "Genetic Programming"), in which finite state machines ("ants") are tasked with following a trail of "breadcrumbs" while only being able to detect the state of the tile directly in front of them. Here is an example world:

 -----------------------------------------
|                 X X                     |
| X             X         X       X       |
| X                       X             X |
| X X X X                               X |
|       X       X               X         |
|       X       X           X X X         |
|       X                                 |
|       X       X                       X |
|   X X X       X             X       X X |
|   X                   X                 |
|   X                                     |
|   X X X                                 |
|       X       X     X                   |
|               X     X                   |
|       X       X       X                 |
|       X                             X X |
| X   X X                               X |
| X             X                       X |
| X             X                       X |
| X X   X     X       X X X X X X X X X X |
 -----------------------------------------

A well trained ant can be seen successfully exploring this map in this YouTube video

The finite state machines controlling each ant is stored as a 2D array, indexed by the ant's current state and whether or not there is a breadcrumb in the tile directly in front of it. The values in a given cell determine what action the ant should take and what state it should change to.

Two ants can produce two offspring by selecting a random contiguous region of the matrix and swapping those values between the two ants.

Simply running make will compile the executables "test" and "main-Ant".

If make displays compilation errors, it is likely because your compiler provides additional compilation warnings. Such errors can be suppressed by changing the value of the variable "STRICT"; this can either be done my modifying the makefile or by running make STRICT=100 instead.

By default the makefile compiles in "debug" mode; this may be changed by replacing the line CFLAGS += $(OPTS_DEBUG) in the makefile with CFLAGS += $(OPTS_OPTIMIZED), doing so can decrease the runtimes by a factor of four or more.

If the preprocessor flag "DEBUG" was defined during compilation, the genetic algorithm will print a message at the end of each generation. Otherwise, it will have no output.

"main-Ant" is run with no arguments.

Runtimes typically vary between two seconds and two minutes; most of the variation is caused by the speed on your computer and whether the makefile was configured to use debugging flags or optimization flags.

"test" is run with no arguments. When complete, it will print the most fit string it found to the console.

Runtimes typically vary between two seconds and two minutes; most of the variation is caused by the speed on your computer and whether the makefile was configured to use debugging flags or optimization flags.