bask.utils
.guess_priors¶
- bask.utils.guess_priors(kernel)[source]¶
Guess suitable priors for the hyperparameters of a given kernel.
This function recursively explores the given (composite) kernel and adds suitable priors each encountered hyperparameter.
Here we use a half-Normal(0, 2.0) prior for all ConstantKernels and WhiteKernels, and an round-flat(0.1, 0.6) prior for all lengthscales. Change of variables is applied, since inference is done in log-space.
- Parameters:
- kernelKernel object.
Can be a single kernel (e.g. Matern), a Product or Sum kernel, or a CompoundKernel.
- Returns:
- priorslist of functions.
The function returns the list of priors in the same order as the vector theta provided by the kernel. Each prior evaluates the logpdf of its argument.