[![Test-install and cache deps](https://github.com/JJ/p6-algorithm-evolutionary-simple/actions/workflows/test.yaml/badge.svg)](https://github.com/JJ/p6-algorithm-evolutionary-simple/actions/workflows/test.yaml) NAME ==== Algorithm::Evolutionary::Simple - A simple evolutionary algorithm SYNOPSIS ======== use Algorithm::Evolutionary::Simple; # From resources/examples/max-ones.p6 my UInt :$length = 64; my UInt :$population-size = 256; my @initial-population = initialize( size => $population-size, genome-length => $length ); my %fitness-of; my $population = evaluate( population => @initial-population, fitness-of => %fitness-of, evaluator => &max-ones ); my $result = 0; while $population.sort(*.value).reverse.[0].value < $length { $population = generation( population => $population, fitness-of => %fitness-of, evaluator => &max-ones, population-size => $population-size) ; $result += $population-size; info(to-json( { best => best-fitness($population) } )); } say $result DESCRIPTION =========== `Algorithm::Evolutionary::Simple` is a module for writing simple and quasi -canonical evolutionary algorithms in Raku. It uses binary representation, integer fitness (which is needed for the kind of data structure we are using) and a single fitness function. It is intended mainly for demo purposes, although it's been actually used in research. In the future, more versions will be available. It uses a fitness cache for storing and not reevaluating already seen chromosomes, so be mindful of memory bloat. EXAMPLES ======== Go to [`resources/examples`](resources/examples) for examples. For instance , run `max-ones.p6` or `p-peaks.p6` there. You'll need to run zef install --deps-only . To install needed modules in that directory. METHODS ======= initialize( UInt :$size, UInt :$genome-length --> Array ) is export ------------------------------------------------------------------- Creates the initial population of binary chromosomes with the indicated length; returns an array. random-chromosome( UInt $length --> List ) ------------------------------------------ Generates a random chromosome of indicated length. Returns a `Seq` of `Bool`s max-ones( @chromosome --> Int ) ------------------------------- Returns the number of trues (or ones) in the chromosome. leading-ones( @chromosome --> Int ) ----------------------------------- Returns the number of ones from the beginning of the chromosome. royal-road( @chromosome ) ------------------------- That's a bumpy road, returns 1 for each block of 4 which has the same true or false value. multi evaluate( :@population, :%fitness-of, :$evaluator, :$auto-t = False --> Mix ) is export --------------------------------------------------------------------------------------------- Evaluates the chromosomes, storing values in the fitness cache. If `auto-t` is set to 'True', uses autothreading for faster operation (if needed). In absence of that parameter, defaults to sequential. sub evaluate-nocache( :@population, :$evaluator --> Mix ) --------------------------------------------------------- Evaluates the population, returning a Mix, but does not use a cache. Intended mainly for concurrent operation. get-pool-roulette-wheel( Mix $population, UInt $need = $population.elems ) is export ------------------------------------------------------------------------------------ Returns `$need` elements with probability proportional to its *weight*, which is fitness in this case. mutation( @chromosome is copy --> Array ) ----------------------------------------- Returns the chromosome with a random bit flipped. crossover ( @chromosome1 is copy, @chromosome2 is copy ) returns List --------------------------------------------------------------------- Returns two chromosomes, with parts of it crossed over. Generally you will want to do crossover first, then mutation. produce-offspring( @pool, $size = @pool.elems --> Seq ) is export ----------------------------------------------------------------- Produces offspring from an array that contains the reproductive pool; it returns a `Seq`. produce-offspring-no-mutation( @pool, $size = @pool.elems --> Seq ) is export ----------------------------------------------------------------------------- Produces offspring from an array that contains the reproductive pool without using mutation; it returns a `Seq`. best-fitness( $population ) --------------------------- Returns the fitness of the first element. Mainly useful to check if the algorithm is finished. multi sub generation( :@population, :%fitness-of, :$evaluator, :$population-size = $population.elems, Bool :$auto-t --> Mix ) ----------------------------------------------------------------------------------------------------------------------------- Single generation of an evolutionary algorithm. The initial `Mix` has to be evaluated before entering here using the `evaluate` function. Will use auto-threading if `$auto-t` is `True`. multi sub generation( :@population, :%fitness-of, :$evaluator, :$population-size = $population.elems, Bool :$no-mutation --> Mix ) ---------------------------------------------------------------------------------------------------------------------------------- Single generation of an evolutionary algorithm. The initial `Mix` has to be evaluated before entering here using the `evaluate` function. Will not use mutation if that variable is set to `True` sub generations-without-change( $generations, $population ) ----------------------------------------------------------- Returns `False` if the number of generations in `$generations` has not been reached without changing; it returns `True` otherwise. mix( $population1, $population2, $size --> Mix ) is export ----------------------------------------------------------- Mixes the two populations, returning a single one of the indicated size and with type Mix. sub pack-individual( @individual --> Int ) ------------------------------------------ Packs the individual in a single `Int`. The invidual must be binary, and the maximum length is 64. sub unpack-individual( Int $packed, UInt $bits --> Array(Seq)) -------------------------------------------------------------- Unpacks the individual that has been packed previously using `pack-individual` sub pack-population( @population --> Buf) ------------------------------------------ Packs a population, producing a buffer which can be sent to a channel or stored in a compact form. sub unpack-population( Buf $buffer, UInt $bits --> Array ) ---------------------------------------------------------- Unpacks the population that has been packed using `pack-population` multi sub frequencies( $population) ----------------------------------- `$population` can be an array or a Mix, in which case the keys are extracted. This returns the per-bit (or gene) frequency of one (or True) for the population. multi sub frequencies-best( $population, $proportion = 2) --------------------------------------------------------- `$population` is a Mix, in which case the keys are extracted. This returns the per-bit (or gene) frequency of one (or True) for the population of the best part of the population; the size of the population will be divided by the $proportion variable. sub generate-by-frequencies( $population-size, @frequencies ) ------------------------------------------------------------- Generates a population of that size with every gene according to the indicated frequency. sub crossover-frequencies( @frequencies, @frequencies-prime --> Array ) ----------------------------------------------------------------------- Generates a new array with random elements of the two arrays that are used as arguments. SEE ALSO ======== There is a very interesting implementation of an evolutionary algorithm in [Algorithm::Genetic](https://github.com/samgwise/p6-algorithm-genetic). Check it out. This is also kind of a port of [Algorithm::Evolutionary::Simple to Perl6](https ://metacpan.org/release/Algorithm-Evolutionary-Simple), which has a few more goodies, but it's not simply a port, since most of the code is completely different. AUTHOR ====== JJ Merelo COPYRIGHT AND LICENSE ===================== Copyright 2018, 2019, 2022 JJ Merelo This library is free software; you can redistribute it and/or modify it under the Artistic License 2.0.