PTMLocScaleDataGen

PTMLocScaleDataGen#

class liesel_ptm.PTMLocScaleDataGen(shape, loc_fn=None, scale_fn=None, ncov=0, zmin=-2.5, zmax=2.5, use_norm_at_zero=True)[source]#

Bases: DataGenerator

Draws random samples from a location-scale transformation model.

Parameters:

  • shape (Any) – A vector of shape parameters \(\boldsymbol{\delta}\) used to define the transformation function.

  • loc_fn (Optional[Callable[[Any], Any]]) – A function taking in a 2d array of covariates, returning a 1d array of locations. (default: None)

  • scale_fn (Optional[Callable[[Any], Any]]) – A function taking in a 2d array of covariates, returning a 1d array of scales. (default: None)

  • ncov (int) – An integer, giving the number of covariates to generate. (default: 0)

  • zmin (float) – Lower and upper boundary knot locations, respectively. (default: -2.5)

  • zmax (float) – Lower and upper boundary knot locations, respectively. (default: 2.5)

  • use_norm_at_zero (bool) – If True, the transformation function is shifted such that \(h(0)=0\) before inversion. Should be kept this way, this option exists mainly for testing purposes. (default: True)

Examples

Example without covariates:

>>> import liesel_ptm as ptm
>>> import jax

>>> key1 = jax.random.PRNGKey(21)
>>> key2 = jax.random.PRNGKey(42)
>>> shape = ptm.sample_shape(key1, nshape=10, scale=0.5).sample

>>> dg = ptm.PTMLocScaleDataGen(shape=shape)
>>> sample = dg.sample(key2, nobs=10)
>>> sample_df = dg.to_df(sample)
>>> sample_df[["y", "pdf"]].head()
        y       pdf
0 -2.283257  0.087928
1  0.430342  0.645286
2  0.091028  0.449210
3  0.552843  0.651007
4  0.443428  0.648257

Example with one covariate, using the same shape:

>>> dg = ptm.PTMLocScaleDataGen(shape=shape, loc_fn=lambda x: 1.5*x, ncov=1)
>>> sample = dg.sample(key2, nobs=10)
>>> sample_df = dg.to_df(sample)
>>> sample_df[["y", "pdf"]].head()

Methods

dfgrid

Evaluates a number of quantities given fitting arrays of z and x.

loc

Location function.

normalization

Normalization function \(h(\varepsilon)\).

normalization_inv

Inverse of the normalization function.

sample

Draws random samples from the transformation model implied by shape, loc(), and scale().

scale

Scale function.

to_df

Takes a data dictionary as returned by sample() and turns it into a pandas.DataFrame.

transformation

Full transformation function, including the scale and location parts.

transformation_deriv

Derivative of the full transformation function, including the scale and location parts.

transformation_inverse

Inverse of the full transformation function, including the scale and location parts.

Attributes

knots

Knots of the transformation function's spline segment.

shape

Shape parameters \(\boldsymbol{\delta}\).
Contents