Base Variables implementation

class Variable(label, **kwargs)

Bases: ABC

Variable is an Abstract class defining what a variable is in a given search space.

Parameters:

• label (str) – Name of the variable.

• kwargs (dict) – Kwargs will be the different addons you want to add to a Variable. Known addons are: * neighbor : VarNeighborhood

label

abstract random(size=None)

abstract isconstant()

Variables functionnalities can be extended with Addons.

Base Variables are the low level bricks to compose a search space in DRAGON.

class IntVar(label, lower, upper, sampler=<built-in method randint of numpy.random.mtrand.RandomState object>, **kwargs)

Bases: Variable

IntVar defines Variable discribing Integer variables. The user must specify a lower and an upper bounds for the Integer variable.

Parameters:

• label (str) – Name of the variable.

• lower (int) – Lower bound of the variable

• upper (int) – Upper bound of the variable

• sampler (Callable, default=np.random.randint) – Function that takes lower bound, upper bound and a size as parameters and defines how the random values for the variable should be sampled.

up_bound

Lower bound of the variable

: int

Upper bound of the variable

Type:

int

Examples

>>> from dragon.search_space.base_variables import IntVar
>>> a = IntVar("test", 0, 5)
>>> print(a)
IntVar(test, [0;5])
>>> a.random()
1

random(size=None)

Parameters:

size (int, default=None) – Number of draws.

Returns:

out – Return an int if size`=1, a :code:`list[int] else.

Return type:

int or list[int]

isconstant()

Returns:

out – Return True, if this Variable is a constant (lower`==:code:`upper), False otherwise.

Return type:

boolean

class FloatVar(label, lower, upper, sampler=<built-in method uniform of numpy.random.mtrand.RandomState object>, tolerance=1e-14, **kwargs)

Bases: Variable

FloatVar defines Variable discribing Float variables.

Parameters:

• label (str) – Name of the variable.

• lower ({int,float}) – Lower bound of the variable

• upper ({int,float}) – Upper bound of the variable

• sampler (Callable, default=np.random.uniform) – Function that takes lower bound, upper bound and a size as parameters.

up_bound

Lower bound of the variable

Type:

{int,float}

low_bound

Upper bound of the variable

Type:

{int,float}

Examples

>>> from dragon.search_space.base_variables import FloatVar
>>> a = FloatVar("test", 0, 5.0)
>>> print(a)
FloatVar(test, [0;5.0])
>>> a.random()
2.2011985711663056

random(size=None)

Parameters:

size (int, default=None) – Number of draws.

Returns:

out – Return a float if size`=1, a :code:`list[float] else.

Return type:

float or list[float]

isconstant()

Returns:

out – Return True, if this Variable is a constant (lower`==:code:`upper), False otherwise.

Return type:

boolean

class CatVar(Variable)

Bases: Variable

CatVar defines Variable discribing categorical variables.

Parameters:

• label (str) – Name of the variable.

• features (list) – List of all choices.

• weights (list[float]) – Weights associated to each elements of features. The sum of all positive elements of this list, must be equal to 1.

features

weights

Examples

>>> from dragon.search_space.base_variables import CatVar, IntVar
>>> a = CatVar("test", ['a', 1, 2.56, IntVar("int", 100 , 200)])
>>> print(a)
CatVar(test, ['a', 1, 2.56, IntVar(int, [100;200])])
>>> a.random(10)
['a', 180, 2.56, 'a', 'a', 2.56, 185, 2.56, 105, 1]

random(size=1)

Parameters:

size (int, default=1) – Number of draws.

Returns:

out – Return a feature if size`=1, a :code:`list[features] else. Features can be Variable. When seleted, it will return a random point from this Variable.

Return type:

float or list[float]

isconstant()

Returns:

out – Return True, if this Variable is a constant (len(feature)==1), False otherwise.

Return type:

boolean

class Constant(label, value, **kwargs)

Bases: Variable

Constant is a Variable discribing a constant of any type.

Parameters:

• label (str) – Name of the variable.

• value (object) – Constant value

label

Name of the variable.

Type:

str

value

Constant value

Type:

object

Examples

>>> from dragon.search_space.base_variables import Constant
>>> a = Constant("test", 5)
>>> print(a)
Constant(test, 5)
>>> a.random()
5

random(size=None)

Parameters:

size (int, default=None) – Number of draws.

Returns:

out – Return an int if size`=1, a :code:`list[self.value] else.

Return type:

int or list[int]

isconstant()

Returns:

out – Return True

Return type:

boolean