VariableTransformer

class pybads.variable_transformer.VariableTransformer(D, lower_bounds: ndarray = None, upper_bounds: ndarray = None, plausible_lower_bounds: ndarray = None, plausible_upper_bounds: ndarray = None, apply_log_t=None, fixed_values: ndarray = None)

A class enabling linear or non-linear transformation of the bounds (plausible_lower_bounds and plausible_upper_bounds) and map them to an hypercube [-1, 1]^D

Parameters:

Dint

The dimension of the space.

lower_boundsnp.ndarray, optional

The lower bounds of the space. lower_bounds and upper_bounds define a set of strict lower and upper bounds for each variable, given in the original space. By default None, which is -inf.

upper_boundsnp.ndarray, optional

The upper bounds of the space. lower_bounds and upper_bounds define a set of strict lower and upper bounds for each variable, given in the original space. By default None, which is inf.

plausible_lower_boundsnp.ndarray, optional

The plausible lower bounds such that lower_bounds <= plausible_lower_bounds < plausible_upper_bounds <= upper_bounds. plausible_lower_bounds and plausible_upper_bounds represent a “plausible” range for each variable, given in the original space, and need to be finite. By default None, which is lower_bounds.

plausible_upper_boundsnp.ndarray, optional

The plausible upper bounds such that lower_bounds <= plausible_lower_bounds < plausible_upper_bounds <= upper_bounds. plausible_lower_bounds and plausible_upper_bounds represent a “plausible” range for each variable, given in the original space, and need to be finite. By default None, which is upper_bounds.

apply_log_tnp.ndarray, optional

A boolean array of shape (1, D) that indicates the variables to which the non-linear log transformation applies; a scalar applies to every variable. By default None, in which case the log transformation is applied if the bounds are all positive and the plausible box spans at least one order of magnitude (pub/plb >= 10), as it is to a variable whose entry is NaN.

fixed_valuesnp.ndarray, optional

The values of the fixed variables of a problem, when the transform is of its other variables: one value per variable of the original space, of shape (1, D_orig) or (D_orig,), NaN at each of the D variables that the bounds describe, in their order, and the value elsewhere. inverse_transf then returns points of the D_orig variables, with the fixed ones at their values, and __call__ takes such points and transforms their other coordinates. By default None, which, as a row of NaN, fixes no variable: the original space has the D variables of the bounds.

Each bound is an array of ``D`` elements, of shape ``(1, D)`` or

``(D,)``, or a scalar or an array of one element, which stands for the

same bound in each dimension; bounds of integers are taken as floats.

Attributes:

D_origint

The number of variables of the original space: D and the fixed variables.

fixed_valuesnp.ndarray or None

fixed_values as a row of shape (1, D_orig), or None when it fixes no variable.

Raises:

ValueError

When a bound is not a number or an array of numbers, or is neither a scalar nor an array of one or D elements, when the plausible bounds are not finite or the bounds are out of the order above, or when the transform cannot be inverted at the bounds.

ValueError

When fixed_values is not a row of numbers, NaN at exactly D of them and finite at the others.

__call__(input: ndarray)

Performs direct transform of original variables input into the hypercube space.

Parameters:

inputnp.ndarray

A N x D array, where N is the number of input data and D is the number of dimensions; N x D_orig with fixed_values, whose fixed coordinates are left out.

Returns:

unp.ndarray

The variables transformed.

inverse_transf(input: ndarray)

Performs inverse transform of the transformed variables input in the hypercube into the original space.

Parameters:

inputnp.ndarray

The transformed variables that will be mapped in the original space.

Returns:

xnp.ndarray

The original variables retrieved by the inverse transform; with fixed_values, all D_orig of them, the fixed ones at their values.