Kernels

Vanguard includes gpytorch.kernels.Kernel subclasses which are recommended for use in controllers.

class vanguard.kernels.ScaledRBFKernel(batch_shape=(), ard_num_dims=None)

The recommended starting place for a kernel.

Parameters:

• batch_shape (Union[tuple[int, ...], Size])

• ard_num_dims (Optional[int])

__init__(batch_shape=(), ard_num_dims=None)

Initialise self.

Parameters:

• batch_shape (Union[tuple[int, ...], Size]) – The batch shape. Defaults to no batching.

• ard_num_dims (Optional[int]) – Set this if you want a separate lengthscale for each input dimension. Defaults to none.

class vanguard.kernels.PeriodicRBFKernel

An RBF kernel with a periodic element.

__init__()

Initialise self.

class vanguard.kernels.TimeSeriesKernel(time_dimension=0)

A kernel suited to time series.

Parameters:

time_dimension (int)

__init__(time_dimension=0)

Initialise self.

Parameters:

time_dimension (int) – The dimension in the data that corresponds to time.