AI-Genetic-Pro
view release on metacpan or search on metacpan
NAME
AI::Genetic::Pro - Efficient genetic algorithms for professional
purpose with support for multiprocessing.
SYNOPSIS
use AI::Genetic::Pro;
sub fitness {
my ($ga, $chromosome) = @_;
return oct('0b' . $ga->as_string($chromosome));
}
sub terminate {
my ($ga) = @_;
my $result = oct('0b' . $ga->as_string($ga->getFittest));
return $result == 4294967295 ? 1 : 0;
}
my $ga = AI::Genetic::Pro->new(
-fitness => \&fitness, # fitness function
-terminate => \&terminate, # terminate function
-type => 'bitvector', # type of chromosomes
-population => 1000, # population
-crossover => 0.9, # probab. of crossover
-mutation => 0.01, # probab. of mutation
-parents => 2, # number of parents
-selection => [ 'Roulette' ], # selection strategy
-strategy => [ 'Points', 2 ], # crossover strategy
-cache => 0, # cache results
-history => 1, # remember best results
-preserve => 3, # remember the bests
-variable_length => 1, # turn variable length ON
-mce => 1, # optional MCE support
-workers => 3, # number of workers (MCE)
);
# init population of 32-bit vectors
$ga->init(32);
# evolve 10 generations
$ga->evolve(10);
# best score
print "SCORE: ", $ga->as_value($ga->getFittest), ".\n";
# save evolution path as a chart
$ga->chart(-filename => 'evolution.png');
# save state of GA
$ga->save('genetic.sga');
# load state of GA
$ga->load('genetic.sga');
DESCRIPTION
This module provides efficient implementation of a genetic algorithm
for professional purpose with support for multiprocessing. It was
designed to operate as fast as possible even on very large populations
and big individuals/chromosomes. AI::Genetic::Pro was inspired by
AI::Genetic, so it is in most cases compatible (there are some
changes). Additionally AI::Genetic::Pro isn't a pure Perl solution, so
it doesn't have limitations of its ancestor (such as slow-down in the
case of big populations ( >10000 ) or vectors with more than 33
fields).
If You are looking for a pure Perl solution, consider AI::Genetic.
Speed
To increase speed XS code is used, however with portability in mind.
This distribution was tested on Windows and Linux platforms (and
should work on any other).
Multicore support is available through Many-Core Engine (MCE). You
can gain the most speed up for big populations or time/CPU consuming
fitness functions, however for small populations and/or simple
fitness function better choice will be single-process version.
You can get even more speed up if you turn on use of native arrays
(parameter: native) instead of packing chromosomes into single
scalar. However you have to remember about expensive memory use in
that case.
Memory
This module was designed to use as little memory as possible. A
population of size 10000 consisting of 92-bit vectors uses only ~24MB
(AI::Genetic would use about 78MB). However - if you use MCE - there
will be bigger memory consumption. This is consequence of necessity
of synchronization between many processes.
Advanced options
To provide more flexibility AI::Genetic::Pro supports many
statistical distributions, such as uniform, natural, chi_square and
others. This feature can be used in selection and/or crossover. See
the documentation below.
METHODS
$ga->new( %options )
Constructor. It accepts options in hash-value style. See options and
an example below.
-fitness
This defines a fitness function. It expects a reference to a
subroutine.
-terminate
This defines a terminate function. It expects a reference to a
subroutine.
-type
This defines the type of chromosomes. Currently, AI::Genetic::Pro
supports four types:
bitvector
Individuals/chromosomes of this type have genes that are bits.
Each gene can be in one of two possible states, on or off.
listvector
Each gene of a "listvector" individual/chromosome can assume one
string value from a specified list of possible string values.
rangevector
Each gene of a "rangevector" individual/chromosome can assume one
integer value from a range of possible integer values. Note that
only integers are supported. The user can always transform any
desired fractional values by multiplying and dividing by an
appropriate power of 10.
combination
Each gene of a "combination" individual/chromosome can assume one
string value from a specified list of possible string values. All
genes are unique.
-population
This defines the size of the population, i.e. how many chromosomes
simultaneously exist at each generation.
-crossover
This defines the crossover rate. The fairest results are achieved
with crossover rate ~0.95.
-mutation
-variable_length
This defines whether variable-length chromosomes are turned on
(default off) and a which types of mutation are allowed. See below.
level 0
Feature is inactive (default). Example:
-variable_length => 0
# chromosomes (i.e. bitvectors)
0 1 0 0 1 1 0 1 1 1 0 1 0 1
0 0 1 1 0 1 1 1 1 0 0 1 1 0
0 1 1 1 0 1 0 0 1 1 0 1 1 1
0 1 0 0 1 1 0 1 1 1 1 0 1 0
# ...and so on
level 1
Feature is active, but chromosomes can varies only on the right
side, Example:
-variable_length => 1
# chromosomes (i.e. bitvectors)
0 1 0 0 1 1 0 1 1 1
0 0 1 1 0 1 1 1 1
0 1 1 1 0 1 0 0 1 1 0 1 1 1
0 1 0 0 1 1 0 1 1 1
# ...and so on
level 2
Feature is active and chromosomes can varies on the left side and
on the right side; unwanted values/genes on the left side are
replaced with undef, ie.
-variable_length => 2
# chromosomes (i.e. bitvectors)
x x x 0 1 1 0 1 1 1
x x x x 0 1 1 1 1
x 1 1 1 0 1 0 0 1 1 0 1 1 1
0 1 0 0 1 1 0 1 1 1
# where 'x' means 'undef'
# ...and so on
In this situation returned chromosomes in an array context
($ga->as_array($chromosome)) can have undef values on the left
side (only). In a scalar context each undefined value is replaced
with a single space. If You don't want to see any undef or space,
just use as_array_def_only and as_string_def_only instead of
as_array and as_string.
-parents
This defines how many parents should be used in a crossover.
-selection
This defines how individuals/chromosomes are selected to crossover.
It expects an array reference listed below:
-selection => [ $type, @params ]
where type is one of:
RouletteBasic
Each individual/chromosome can be selected with probability
proportional to its fitness.
Roulette
First the best individuals/chromosomes are selected. From this
collection parents are selected with probability poportional to
their fitness.
RouletteDistribution
Each individual/chromosome has a portion of roulette wheel
proportional to its fitness. Selection is done with the specified
distribution. Supported distributions and parameters are listed
below.
-selection => [ 'RouletteDistribution', 'uniform' ]
Standard uniform distribution. No additional parameters are
needed.
-selection => [ 'RouletteDistribution', 'normal', $av, $sd ]
Normal distribution, where $av is average (default: size of
population /2) and $$sd is standard deviation (default: size of
population).
-selection => [ 'RouletteDistribution', 'beta', $aa, $bb ]
Beta distribution. The density of the beta is:
X^($aa - 1) * (1 - X)^($bb - 1) / B($aa , $bb) for 0 < X < 1.
$aa and $bb are set by default to number of parents.
Argument restrictions: Both $aa and $bb must not be less than
1.0E-37.
-selection => [ 'RouletteDistribution', 'binomial' ]
Binomial distribution. No additional parameters are needed.
-selection => [ 'RouletteDistribution', 'chi_square', $df ]
Chi-square distribution with $df degrees of freedom. $df by
default is set to size of population.
-selection => [ 'RouletteDistribution', 'exponential', $av ]
Exponential distribution, where $av is average . $av by default
is set to size of population.
-selection => [ 'RouletteDistribution', 'poisson', $mu ]
Poisson distribution, where $mu is mean. $mu by default is set
to size of population.
Distribution
Chromosomes/individuals are selected with specified distribution.
See below.
-selection => [ 'Distribution', 'uniform' ]
Standard uniform distribution. No additional parameters are
needed.
-selection => [ 'Distribution', 'normal', $av, $sd ]
Normal distribution, where $av is average (default: size of
population /2) and $$sd is standard deviation (default: size of
population).
-selection => [ 'Distribution', 'beta', $aa, $bb ]
Beta distribution. The density of the beta is:
X^($aa - 1) * (1 - X)^($bb - 1) / B($aa , $bb) for 0 < X < 1.
$aa and $bb are set by default to number of parents.
Argument restrictions: Both $aa and $bb must not be less than
1.0E-37.
-selection => [ 'Distribution', 'binomial' ]
Binomial distribution. No additional parameters are needed.
-selection => [ 'Distribution', 'chi_square', $df ]
Chi-square distribution with $df degrees of freedom. $df by
default is set to size of population.
-selection => [ 'Distribution', 'exponential', $av ]
Exponential distribution, where $av is average . $av by default
is set to size of population.
-selection => [ 'Distribution', 'poisson', $mu ]
Poisson distribution, where $mu is mean. $mu by default is set
to size of population.
-strategy
This defines the astrategy of crossover operation. It expects an
array reference listed below:
-strategy => [ $type, @params ]
where type is one of:
PointsSimple
Simple crossover in one or many points. The best
chromosomes/individuals are selected for the new generation. For
example:
-strategy => [ 'PointsSimple', $n ]
where $n is the number of points for crossing.
PointsBasic
Crossover in one or many points. In basic crossover selected
parents are crossed and one (randomly-chosen) child is moved to
the new generation. For example:
-strategy => [ 'PointsBasic', $n ]
where $n is the number of points for crossing.
Points
Crossover in one or many points. In normal crossover selected
parents are crossed and the best child is moved to the new
generation. For example:
-strategy => [ 'Points', $n ]
where $n is number of points for crossing.
PointsAdvenced
Crossover in one or many points. After crossover the best
chromosomes/individuals from all parents and chidren are selected
for the new generation. For example:
-strategy => [ 'PointsAdvanced', $n ]
where $n is the number of points for crossing.
Distribution
In distribution crossover parents are crossed in points selected
with the specified distribution. See below.
-strategy => [ 'Distribution', 'uniform' ]
Standard uniform distribution. No additional parameters are
needed.
-strategy => [ 'Distribution', 'normal', $av, $sd ]
Normal distribution, where $av is average (default: number of
parents/2) and $sd is standard deviation (default: number of
parents).
-strategy => [ 'Distribution', 'beta', $aa, $bb ]
Beta distribution. The density of the beta is:
X^($aa - 1) * (1 - X)^($bb - 1) / B($aa , $bb) for 0 < X < 1.
$aa and $bb are set by default to the number of parents.
Argument restrictions: Both $aa and $bb must not be less than
1.0E-37.
-strategy => [ 'Distribution', 'binomial' ]
Binomial distribution. No additional parameters are needed.
-strategy => [ 'Distribution', 'chi_square', $df ]
Chi-squared distribution with $df degrees of freedom. $df by
default is set to the number of parents.
-strategy => [ 'Distribution', 'exponential', $av ]
Exponential distribution, where $av is average . $av by default
is set to the number of parents.
-strategy => [ 'Distribution', 'poisson', $mu ]
Poisson distribution, where $mu is mean. $mu by default is set
to the number of parents.
PMX
PMX method defined by Goldberg and Lingle in 1985. Parameters:
none.
OX
OX method defined by Davis (?) in 1985. Parameters: none.
-cache
This defines whether a cache should be used. Allowed values are 1
or 0 (default: 0).
-history
This defines whether history should be collected. Allowed values
( run in 0.949 second using v1.01-cache-2.11-cpan-f03e8824b8d )