PTMLocScale

class liesel_ptm.PTMLocScale(knots, y, normalization_tau2, transition_width=0.3, scale_after_transformation=False, scaling_factor=None, shape_param_prior=ShapePrior.RANDOM_WALK)

Bases: object

Penalized transformation model for location and scale.

Parameters:

• knots (Any) – The knots used to the define the transformation function.

• y (Any) – The observed response variable.

• normalization_tau2 (Var) – The variance parameter \(\tau^2_\delta\) for the prior of the transformation function’s shape parameters.

• transition_width (float) – The width of the transition segment of the transformation function. A number larger than 0, indicating the length in terms of a multiple of the range of interior knots. Called \(\lambda\). (default: 0.3)

• scale_after_transformation (bool) – Whether to scale the response variable after the transformation. It is not avisable to switch this to True. (default: False)

• scaling_factor (Optional[Var]) – The scaling factor \(\omega\) for the transformation function. If None, the scaling factor will be fixed to one. (default: None)

• shape_param_prior (ShapePrior) – The prior distribution for the shape parameter. (default: <ShapePrior.RANDOM_WALK: 1>)

Methods

all_parameter_names
all_sampled_parameter_names
build_graph	Builds the model graph.
from_nparam	Initializes a penalized transformation model automatically from the desired number of shape parameters.
new_from_graph	Given a Liesel model graph, this creates a new PTMLocScale with updated knots.
optimize_knots	Finds good knots for the transformation function.
setup_engine_builder	Prepares a liesel.goose.EngineBuilder for MCMC sampling.
waic	Assembles a pd.DataFrame that gives information on the widely applicable information criterion for this model.

Attributes

knots	The knots of the transformation function's spline segment.
transition_width	The transition width.
loc_model	Predictor for the location model part.
log_scale_model	Predictor for the log scale model part.
scale_model	The exponential of log_scale_model.
normalization_tau2	The hyperparameter \(\tau^2_\delta\).
scale_after_transformation	Whether to scale the response variable after the transformation.
scaling_factor	The scaling factor \(\omega\) for the transformation function.
shape_param_prior	The prior distribution for the shape parameter.
loc_intercept	Intercept \(\beta_0\) of the location model part.
scale_intercept_exp	Exponential of the intercept \(\exp(\gamma_0)\) of the scale model part.
loc	The response's location, including an intercept.
scale	The response's scale, including an intercept.
residuals	The residuals.
normalization	Organizes the transformation model part \(h(\varepsilon\)).
z_deriv	Derivative of \(h(\varepsilon)\) with respect to the response.
z	Variable for the transformation model part \(h(\varepsilon\)).
refdist	The reference distribution, currently fixed to the standard normal distribution.
response	Response variable.
intercept_names	Names of the intercept nodes.