sample_gp_function

safeopt.sample_gp_function(kernel, bounds, noise_var, num_samples, interpolation='kernel', mean_function=None)

Sample a function from a gp with corresponding kernel within its bounds.

Parameters:

kernel: instance of GPy.kern.*

bounds: list of tuples

[(x1_min, x1_max), (x2_min, x2_max), …]

noise_var: float

Variance of the observation noise of the GP function

num_samples: int or list

If integer draws the corresponding number of samples in all dimensions and test all possible input combinations. If a list then the list entries correspond to the number of linearly spaced samples of the corresponding input

interpolation: string

If ‘linear’ interpolate linearly between samples, if ‘kernel’ use the corresponding mean RKHS-function of the GP.

mean_function: callable

Mean of the sample function

Returns:

function: object

function(x, noise=True) A function that takes as inputs new locations x to be evaluated and returns the corresponding noisy function values. If noise=False is set the true function values are returned (useful for plotting).