TransformationDist

class liesel_ptm.TransformationDist(coef, bspline, parametric_distribution=None, reference_distribution=None, validate_args=False, allow_nan_stats=True, name='TransformationDist', centered=False, scaled=False, batched=True, **parametric_distribution_kwargs)

Bases: Distribution

Transformation distribution using a spline and optional parametric component.

This combines a spline-based monotonically increasing transformation with a parametric distribution, using a reference distribution (default: standard normal) for mapping and likelihoods.

Parameters:

• coef (Any) – Coefficients for the spline basis.

• bspline (PTMSpline | OnionSpline) – Spline object providing transformation and its inverse/derivative.

• parametric_distribution (type[Distribution] | None) – Parametric distribution class to include in the model. (default: None)

• reference_distribution (Distribution | None) – Reference distribution used for transformations; defaults to Normal(0, 1). (default: None)

• validate_args (bool) – Whether to validate input arguments. (default: False)

• allow_nan_stats (bool) – Whether to allow NaN statistics. (default: True)

• name (str) – Name of the distribution. (default: 'TransformationDist')

• centered (bool) – If True, the transformation is centered such that any side-effect the spline transformation might have on the location of the distribution is negated. (default: False)

• scaled (bool) – If True, the transformation is scaled such that any side-effect the spline transformation might have on the scale of the distribution is negated. (default: False)

• batched (bool) – If True, allow for batched computations (might be slightly less efficient in the scalar case). (default: True)

• **parametric_distribution_kwargs – Additional keyword arguments passed to the parametric distribution.

coef

Coefficients for the spline basis.

bspline

Spline object used for transformations.

reference_distribution

Reference distribution used for mapping and likelihoods.

parametric_distribution

Instantiated parametric distribution if provided, else None.

parametric_distribution_kwargs

Keyword arguments used to construct the parametric distribution.

centered

Indicates whether centering is applied.

scaled

Indicates whether scaling is applied.

Methods

cdf_spline	CDF under the spline-only transformation.
inverse_transformation	Inverse transformation combining parametric and spline parts.
inverse_transformation_parametric	Inverse of the parametric transformation.
inverse_transformation_spline	Inverse of the spline transformation.
log_prob_spline	Log probability under the spline-only transformation.
prob_spline	Probability under the spline-only transformation.
quantile_spline	Quantile function using only the spline transformation.
transformation_and_logdet	Apply parametric then spline transforms; return value and logdet.
transformation_and_logdet_parametric	Apply the parametric transformation and its log determinant.
transformation_and_logdet_spline	Apply spline transformation with centering/scaling and compute logdet.
transformation_spline_mean	Expected value under the spline transformation.
transformation_spline_variance	Variance under the spline transformation.

Attributes