PTMCoef.new_rw1_sumzero_wb()

PTMCoef.new_rw1_sumzero_wb()#

classmethod PTMCoef.new_rw1_sumzero_wb(knots, wb_scale, inference=None, scale_inference=None, name='', scale_penalty=True, diagonalize_penalty=True, role='transformation_coef', scale_bijector=<tfp.bijectors.Exp 'exp' batch_shape=[] forward_min_event_ndims=0 inverse_min_event_ndims=0 dtype_x=? dtype_y=?>)[source]#

Create RW1 sum-to-zero coefficients with a Weibull prior on the random walk variance.

Parameters:

  • knots (Any) – Knot vector for the spline basis.

  • wb_scale (float) – Scale parameter for the Weibull prior.

  • inference (Any) – Optional inference specification for the latent parameter. (default: None)

  • scale_inference (Any) – Optional inference for the scale variable. Will be passed to the inference argument of ScaleWeibull, thus acting on the level of the variance paramter (not the scale parameter). (default: None)

  • name (str) – Optional base name for created variables. (default: '')

  • scale_penalty (bool) – Whether to scale the penalty matrix to unit infinity norm. (default: True)

  • diagonalize_penalty (bool) – Whether to diagonalize the penalty via a eigenvalue decomposition. (default: True)

  • role (str) – Role assigned to the latent coefficient variable. (default: 'transformation_coef')

  • scale_bijector (Bijector | None) – Optional bijector applied to the scale variable. Will be passed to the bijector argument of ScaleWeibull, thus acting on the level of the variance paramter (not the scale parameter). (default: <tfp.bijectors.Exp 'exp' batch_shape=[] forward_min_event_ndims=0 inverse_min_event_ndims=0 dtype_x=? dtype_y=?>)

Return type:

Self

Returns:

PTMCoef – Configured coefficient with a Weibull variance parameter.

Contents