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.