ParameterTransformer

class pyvbmc.parameter_transformer.ParameterTransformer(D: int, lb_orig: ndarray = None, ub_orig: ndarray = None, plb_orig: ndarray = None, pub_orig: ndarray = None, scale: ndarray = None, rotation_matrix: ndarray = None, transform_type='logit')

A class used to enable transforming of variables from unconstrained to constrained space and vice versa.

Parameters:

Dint

The dimension of the space.

lb_orignp.ndarray, optional

The lower bounds of the space. lb_orig and ub_orig define a set of strict lower and upper bounds for each parameter, given in the original space. Each parameter is bounded on both sides or on neither. By default None.

ub_orignp.ndarray, optional

The upper bounds of the space. lb_orig and ub_orig define a set of strict lower and upper bounds for each parameter, given in the original space. Each parameter is bounded on both sides or on neither. By default None.

plb_orignp.ndarray, optional

The plausible lower bounds such that lb_orig <= plb_orig < pub_orig <= ub_orig. plb_orig and pub_orig represent a “plausible” range for each parameter, given in the original space. By default None, which takes the hard bounds.

pub_orignp.ndarray, optional

The plausible upper bounds such that lb_orig <= plb_orig < pub_orig <= ub_orig. plb_orig and pub_orig represent a “plausible” range for each parameter, given in the original space. By default None, which takes the hard bounds.

scalenp.ndarray, optional

Per-coordinate scale applied after transforming and rotating the parameters, of shape (D,) and finite and positive throughout. By default None.

rotation_matrixnp.ndarray, optional

A finite orthogonal matrix of shape (D, D) applied after the coordinate-wise transformation. Reflections are accepted. By default None.

transform_typestr, optional

A string indicating the type of transform for bounded variables: one of [“logit”, (“norminv” || “probit”), “student4”]. Default “logit”.

__call__(x: ndarray)

Performs direct transform of original variables x into unconstrained variables u.

Parameters:

xnp.ndarray

A N x D array, where N is the number of input data and D is the number of dimensions

Returns:

unp.ndarray

The variables transformed to unconstrained variables.

inverse(u: ndarray)

Performs inverse transform of unconstrained variables u into variables x in the original space

Parameters:

unp.ndarray

The unconstrained variables that will be transformed.

Returns:

xnp.ndarray

The original variables which result of the transformation.

log_abs_det_jacobian(u: ndarray)

The log absolute determinant of the Jacobian of the inverse transform at u, \(\log |\det J_{g^{-1}}(u)|\), where \(g\) maps the original space to the transformed space.

Parameters:

unp.ndarray

The points where the log determinant of the Jacobian should be evaluated (in transformed space).

Returns:

pnp.ndarray

The log absolute determinant of the Jacobian.