Default priors for the kernel hyperparameters

Weakly informative log-normal priors (i.e. Normal priors on the log scale) on length_scale, periodic_scale, long_term_scale and the noise-to-signal ratio nugget_ratio. They act as mild regularisation on an otherwise maximum-likelihood fit, keeping weakly identified parameters (notably long_term_scale) away from the boundary.

Usage

default_kernel_priors()

Value

A named list of list(meanlog, sdlog) priors.