BoolIndividual : example Individual class using a binary string genotype

Every AbstractEvolution subclass needs to have the following fields: config::NamedTuple logger::CambrianLogger population::Array{<:Individual} for config, see getdefaultconfig

Evolution classes should implement the following methods: populate(e::Evolution) evaluate(e::Evolution) generation(e::Evolution) see the default functions for reference

Evolution(cfg::NamedTuple; logfile::String="logs/id.csv")

create a base Evolution class. This is provided as an example class and for tests. Intended usage of Cambrian is to subclass the AbstractEvolution type and define the evolutionary methods which define the intended algorithm.

FloatIndividual : example Individual class using a floating point genotype in [0, 1]

GAEvo implements a classic genetic algorithm. To do this, a new subtype of AbstractEvolution is created, GAEvo, and the following functions are defined for this type:

evaluate populate

the function generation can also be defined. Without specification, the default generation function (which is void) is applied.

every Individual subtype needs to implement: Individual(cfg::NamedTuple) Individual(json::String) mutate(ind::Individual) crossover(parents::Vararg{Individual})

and have the fields genes fitness::Array

BoolIndvidiual and FloatIndividual are provided as Cambrian defaults

OnePlusEvo implements a 1+λ evolutionary algorithm. To do this, a new subtype of AbstractEvolution is created, OnePlusEvo, and the following functions are defined for this type:

evaluate populate

the function generation can also be defined. Without specification, the default generation function (which is void) is applied.

default crossover method, using uniform crossover for N parents

update e.elites with the best individuals from the current population or existing elites, based on fitness

evaluate(e::AbstractEvolution)

Default evaluate function. Sets each element of the fitness array of each individual to -Inf.

function fitness_evaluate(e::AbstractEvolution; fitness::Function=null_evaluate)

sets the fitness of each individual to the Array of values returned by fitness

populate(e::GAEvo)

generation should be used for any additional record keeping necessary in an algorithm. It is called after populate and evaluate, so all individuals should have genes and fitness reflecting their current state. The default function calls elites_generation if e.elites is defined, otherwise does nothing.

convert Dict to named tuple

combine YAML file and kwargs, make sure ID is specified

create all members of the first generation

log a generation, including max, mean, and std of each fitness dimension

return the best individual from a population

mutate(parent::BoolIndividual, mrate::Float64). To use, define mutate(parent::BoolIndividual) = mutate(parent, mrate) with configured m_rate for a Boolean individual, this random flips the bits of the parent

sets the fitness value of each individual to rand()

return a random individual from a population

Call step!(e) e.config.n_gen times consecutively

save the population in gens/

The generic iteration of an evolution. Calls populate, evaluate, and generation. Also calls loggen and savegen based on the provided config values. Subclasses of AbstractEvolution should override the populate, evaluate, or generation functions rather than overriding this function.

return a winner individual from a tournament of size t_size from a population