The fvGP Package

The gpCAM API is upstream of fvGP which is the package providing everything around Gaussian Processes. The package has its own API which we present and explain here (IN PROGRESS). Note, this is not really a full tutorial since fvGP is hardly ever being used directly, all information provided here is just for the curious and advanced users. Below there is an example use case presented for a single-task GP.


Most recent version: 3.1.3

class GP(input_space_dim, points, values,

init_hyperparameters, variances = None, compute_device = "cpu", gp_kernel_function = None, gp_mean_function = None,

sparse = False, normalize_y = False)

The GP class is the core of almost all the functionality in fvGP and therefore gpCAM. It is the base class that is used for multi-task GPs and also for ensemble GPs.

class fvGP(input_space_dim, output_space_dim, output_number, points, values,

init_hyperparameters, value_positions = None, variances = None, compute_device = "cpu",

gp_kernel_function = None, gp_mean_function = None,

sparse = False, normalize_y = False)

A Multi-Output GP Framework which inherits most of its functionality from the GP base class.

class EnsembleGP(input_space_dim, points, values, number_of_GPs, hps_obj, variances = None, compute_device = "cpu",

gp_kernel_functions = None, gp_mean_functions = None, sparse = False,

normalize_y = False)

A class for ensemble GPs. Note hyperparameters are now an instance of the hyperparameters class. See the the examples in ./tests/ for more info.

class EnsembleFvGP(coming soon)

A class for ensemble multi-task GPs

class ManifoldGP(coming soon)


class ManifoldFvGP(coming soon)