leaspy.io.realizations.realization module

class Realization(name: str, shape: Tuple[int, ...], variable_type: str)

Contains the realization of a given parameter.

Parameters

Attributes

namestr: Variable name
shapetuple of int: Shape of variable (multiple dimensions allowed)
variable_typestr: 'individual' or 'population' variable?
tensor_realizationstorch.Tensor: Actual realizations, whose shape is given by shape

Methods

`from_tensor`(name, shape, variable_type, ...)	Create realization from variable infos and torch tensor object
`initialize`(n_individuals, model, *[, ...])	Initialize realization from a given model.
`set_autograd`()	Set autograd for tensor of realizations
`set_tensor_realizations_element`(element, dim)	Manually change the value (in-place) of tensor_realizations at dimension dim.
`unset_autograd`()	Unset autograd for tensor of realizations

classmethod from_tensor(name: str, shape: Tuple[int, ...], variable_type: str, tensor_realization: FloatTensor)

Create realization from variable infos and torch tensor object

Parameters

namestr: Variable name
shapetuple of int: Shape of variable (multiple dimensions allowed)
variable_typestr: 'individual' or 'population' variable?
tensor_realizationtorch.Tensor: Actual realizations, whose shape is given by shape

Returns

initialize(n_individuals: int, model: AbstractModel, *, individual_variable_init_at_mean: bool = False)

Initialize realization from a given model.

Parameters

n_individualsint > 0: Number of individuals
modelAbstractModel: The model you want realizations for.
individual_variable_init_at_meanbool (default False): If True: individual variable will be initialized at its mean (from model parameters) Otherwise: individual variable will be a random draw from a Gaussian distribution with loc and scale parameter from model parameters.

Raises

set_autograd()

Set autograd for tensor of realizations

TODO remove? only in legacy code

Raises