Allele

Base Allele class

sgx.allele.base

class sgx.allele.base.Allele

Abstract class for Allele.

An allele must be Hashable (ie. non modifiable)

abstract describe() → str

Pretty describes the current allele.

property mode

Returns the most frequent allele.

property possible_values

Possible values of the allele. None if not reasonably applicable (eg. a float)

abstract sample(sample_type: Optional[str] = 'sample') → Hashable

Sample.

Parameters

sample_type – ‘sample’ (default): random value according to the current probability distribution ‘uniform’: random value according to a uniform probability distribution (ie. completely random) ‘mode’: most common value according to the current probability distribution

abstract update(winner: Hashable, loser: Hashable) → None

Updates the winner and the loser Genotype, so as to modify the new Learning Rate.

Parameters

  • winner – The genotype of the better solution.

  • loser – The genotype of the worse solution.

Boolean Allele class

sgx.allele.boolean

class sgx.allele.boolean.Boolean(learning_rate: float = 0.001)
describe() → str

Pretty describes the current boolean allele.

is_valid(value: Hashable) → bool

Checks if the allele is boolean.

run_paranoia_checks() → bool

Returns True, if all tests are passed.

sample(sample_type: Optional[str] = 'sample') → Hashable

Sample.

Parameters

sample_type – The type of sample (sample (default), uniform, mode).

static sigmoid(x: float, k: Optional[float] = 1) → float

Logistic function with given logistic growth (k). See https://en.wikipedia.org/wiki/Logistic_function

Parameters

  • x – A float number.

  • k – Logistic Growth.

Returns

The result probability of the sigmoid function.

update(winner: Hashable, loser: Hashable) → None

Updates the winner and the loser Genotype, so as to modify the new Learning Rate.

Parameters

  • winner – The genotype of the better solution.

  • loser – The genotype of the worse solution.

Categorical Allele class

sgx.allele.categorical

class sgx.allele.categorical.Categorical(alternatives: Sequence[Hashable], weights: Optional[Union[Sequence[float], dict]] = None, learning_rate: Optional[float] = None)
describe() → str

Pretty describes the current categorical allele.

is_valid(value: Hashable) → bool

Checks if the allele is categorical.

property possible_values

Possible values of the allele. None if not reasonably applicable (eg. a float)

run_paranoia_checks() → bool

Returns True, if all tests are passed.

sample(sample_type: Optional[str] = 'sample') → Hashable

Sample.

Parameters

sample_type – The type of sample (sample (default), uniform, mode).

update(winner: Hashable, loser: Hashable) → None

Updates the winner and the loser Genotype, so as to modify the new Learning Rate.

Parameters

  • winner – The genotype of the better solution.

  • loser – The genotype of the worse solution.

Fitness

Base fitness class

sgx.fitness.base

class sgx.fitness.base.Fitness

Fitness of a phenotype, handle multiple formats (eg. scalar, tuple).

The class also redefines the relational operator in order to handle different types of optimization (eg. maximization, minimization) and to provide limited support to more complex scenarios (eg. multi-objective optimization)

Equalities (‘==’ and ‘!=’) are based on is_distinguishable.

Single angular-bracket operators (‘>’, ‘<’, ‘>=’, and ‘<=’) are based on is_fitter and may be randomized (the result may not be reproducible).

Double angular-bracket operators (‘>>’ and ‘<<’) are based on is_dominant and the result is stable. By default is_dominant is defined as is_fitter.

When subclassing, one should only redefine is_fitter, and optionally is_distinguishable and is_dominant; is_dominant must be changed if is_fitter is randomized (the result is uncertain).

Additional sanity checks should be added to check_comparable. Subclasses may redefine the decorate method to change the values appearance.

__eq__(other: sgx.fitness.base.Fitness) → bool

Returns True if one fitness is equal (==) to the other.

__ge__(other: sgx.fitness.base.Fitness) → bool

Returns True if one fitness is greater or equal (>=) to the other.

__gt__(other: sgx.fitness.base.Fitness) → bool

Returns True if one fitness is greater (>) than the other.

__le__(other: sgx.fitness.base.Fitness) → bool

Returns True if one fitness is less or equal (<=) to the other.

__lshift__(other: sgx.fitness.base.Fitness) → bool

Returns True if one fitness does not dominates (<<) the other.

__lt__(other: sgx.fitness.base.Fitness) → bool

Returns True if one fitness is less (<) than the other.

__ne__(other: sgx.fitness.base.Fitness) → bool

Returns True if one fitness is not equal (!=) to the other.

__rshift__(other: sgx.fitness.base.Fitness) → bool

Returns True if one fitness dominates (>>) the other.

check_comparable(other: sgx.fitness.base.Fitness)

Checks if the fitness is able to be compared.

decorate() → str

Represent the individual fitness value with a nice string.

is_distinguishable(other: sgx.fitness.base.Fitness) → bool

Check whether some differences from the other Fitness may be perceived.

is_dominant(other: sgx.fitness.base.Fitness) → bool

Check whether dominates the other (result is certain).

is_fitter(other: sgx.fitness.base.Fitness) → bool

Check whether fitter than the other (result may be accidental).

is_valid(fitness: sgx.fitness.base.Fitness) → bool

Returns True if the fitness is able to be compared, otherwise Raises an Assertion Error.

run_paranoia_checks() → bool

Returns True if checks are successful.

sgx.fitness.base.reversed(fitness_class: sgx.fitness.base.Fitness) → sgx.fitness.base.Fitness

Reverse fitness class turning a maximization problem into a minimization one.

Fitness Function class

sgx.fitness.function

Multi-Objective class

sgx.fitness.multi_objective

class sgx.fitness.multi_objective.Lexicase(value: Sequence, fitness_type: Type[sgx.fitness.base.Fitness] = <class 'sgx.fitness.simple.Scalar'>, **kwargs)

Pseudo-MO through Lexicase selection (DOI:10.1109/TEVC.2014.2362729).

is_fitter(other: sgx.fitness.multi_objective.Lexicase) → bool

Check whether fitter than the other.

class sgx.fitness.multi_objective.MultiObjective(value: Sequence, fitness_type: Type[sgx.fitness.base.Fitness] = <class 'sgx.fitness.simple.Scalar'>, **kwargs)

Abstract class for handling Multi-Objective problems.

is_dominant(other: sgx.fitness.multi_objective.Lexicase) → bool

Check whether dominates the other (result is certain).

abstract is_fitter(other: sgx.fitness.multi_objective.Lexicase) → bool

Check whether fitter than the other.

Simple class

sgx.fitness.simple

class sgx.fitness.simple.Approximate(argument, rel_tol: float = 1e-09, abs_tol: float = 0)

A single, floating-point value with approximate equality – Larger is better.

check_comparable(other: sgx.fitness.simple.Approximate)

Checks if the fitness is able to be compared.

decorate() → str

Represent the individual fitness value with a nice string.

is_distinguishable(other: sgx.fitness.base.Fitness) → bool

Check whether some differences from the other Fitness may be perceived.

is_fitter(other: sgx.fitness.base.Fitness) → bool

Check whether fitter than the other.

class sgx.fitness.simple.Integer

A single numeric value – Larger is better.

class sgx.fitness.simple.Scalar(x=0, /)

A single numeric value – Larger is better.

class sgx.fitness.simple.Vector(value: Sequence, fitness_type: Type[sgx.fitness.base.Fitness] = <class 'sgx.fitness.simple.Scalar'>, **kwargs)

A generic vector of Fitness values.

fitness_type is the subtype, **kwargs are passed to fitness init

Examples

f1 = sgx.fitness.Vector([23, 10], fitness_type=Approximate, abs_tol=.1) f2 = sgx.fitness.Vector([23, 10], fitness_type=Approximate, abs_tol=.001)

f1 > sgx.fitness.Vector([23, 9.99], fitness_type=Approximate, abs_tol=.1) is False f2 > sgx.fitness.Vector([23, 9.99], fitness_type=Approximate, abs_tol=.001) is True

check_comparable(other: sgx.fitness.simple.Vector)

Checks if the fitness is able to be compared.

static compare_vectors(v1: Sequence[sgx.fitness.base.Fitness], v2: Sequence[sgx.fitness.base.Fitness]) → int

Compare Fitness values in v1 and v2.

Parameters

  • v1 – The first fitness vector.

  • v2 – The second fitness vector.

Returns

-1 if v1 < v2; +1 if v1 > v2; 0 if v1 == v2

decorate() → str

Represent the individual fitness value with a nice string.

is_distinguishable(other: sgx.fitness.simple.Vector) → bool

Check whether some differences from the other Fitness may be perceived.

is_fitter(other: sgx.fitness.base.Fitness) → bool

Check whether fitter than the other.

Utils

Jupyter Support

sgx.utils.jupyter_support

sgx.utils.jupyter_support.is_notebook() → bool

Check if running inside a notebooks

Credits: https://stackoverflow.com/questions/15411967/

Logging

sgx.utils.logging

sgx.utils.logging.log_cpu(level: int = 20, msg: str = '', *args, **kwargs) → None

Like log(), but including cpu time.

Random class

sgx.utils.random

Archive class

sgx.archive

Base class

sgx.base

class sgx.base.Genome(*args)

A tuple of Alleles, each one specifying a set of alternative genes.

is_valid(genotype: sgx.base.Genotype) → bool

Check an object against a specification.

The function may be used to check a value against a parameter definition, a node against a section definition).

Returns

True if the object is valid, False otherwise.

run_paranoia_checks() → bool

Check the internal consistency of a “paranoid” object.

The function should be overridden by the sub-classes to implement the required, specific checks. It always returns True, but throws an exception whenever an inconsistency is detected.

Notez bien: Sanity checks may be computationally intensive, paranoia checks are not supposed to be used in production environments (i.e., when -O is used for compiling). Their typical usage is: assert foo.run_paranoia_checks()

Returns

True (always)

Raises

AssertionError if some internal data structure is incoherent –

class sgx.base.Genotype(*args)

A tuple containing the organism’s actual genes (their values).

run_paranoia_checks() → bool

Check the internal consistency of a “paranoid” object.

The function should be overridden by the sub-classes to implement the required, specific checks. It always returns True, but throws an exception whenever an inconsistency is detected.

Notez bien: Sanity checks may be computationally intensive, paranoia checks are not supposed to be used in production environments (i.e., when -O is used for compiling). Their typical usage is: assert foo.run_paranoia_checks()

Returns

True (always)

Raises

AssertionError if some internal data structure is incoherent –

class sgx.base.Paranoid

Abstract class: Paranoid classes do implement run_paranoia_checks().

run_paranoia_checks() → bool

Check the internal consistency of a “paranoid” object.

The function should be overridden by the sub-classes to implement the required, specific checks. It always returns True, but throws an exception whenever an inconsistency is detected.

Notez bien: Sanity checks may be computationally intensive, paranoia checks are not supposed to be used in production environments (i.e., when -O is used for compiling). Their typical usage is: assert foo.run_paranoia_checks()

Returns

True (always)

Raises

AssertionError if some internal data structure is incoherent –

class sgx.base.Pedantic

Abstract class: Pedantic classes do implement is_valid().

abstract is_valid(obj: Any) → bool

Check an object against a specification.

The function may be used to check a value against a parameter definition, a node against a section definition).

Returns

True if the object is valid, False otherwise.

Species class

sgx.species

class sgx.species.Species(genome: Sequence[Any], fitness_function: sgx.fitness.function.FitnessFunction, mutation_rate: Optional[float] = None)

Missing

evaluate(genotype: sgx.base.Genotype) → sgx.fitness.base.Fitness

Evaluates a genotype according to a specific Fitness Function.

Parameters

genotype – The Genotype which is going to be evaluated.

Returns

The fitness calculated from a Fitness Function.

sample(sample_type: Optional[str] = 'sample') → sgx.base.Genotype

Sampling the genotypes with a specific method.

Parameters

sample_type – The type of sampling is going to be used.

Returns

The candidate’s Genotype after sampling.

update(winner: sgx.base.Genotype, loser: sgx.base.Genotype)

Updates the winner and the loser Genotype, so as to modify the new Learning Rate.

Parameters

  • winner – The genotype of the better solution.

  • loser – The genotype of the worse solution.

SGX Algorithm

Simple Algorithm class

sgx.algorithms.simple

Modular Design Rationale

…?