def __init__(self, num_dim, grid_bounds=(-10., 10.), grid_size=64):
variational_distribution = variational.CholeskyVariationalDistribution(
num_inducing_points=grid_size, batch_shape=torch.Size([num_dim])
)
# Our base variational strategy is a GridInterpolationVariationalStrategy,
# which places variational inducing points on a Grid
# We wrap it with a MultitaskVariationalStrategy so that our output is a vector-valued GP
variational_strategy = variational.MultitaskVariationalStrategy(
variational.GridInterpolationVariationalStrategy(
self, grid_size=grid_size, grid_bounds=[grid_bounds],
variational_distribution=variational_distribution,
), num_tasks=num_dim,
)
super().__init__(variational_strategy)
self.covar_module = kernels.ScaleKernel(
kernels.RBFKernel(
lengthscale_prior=priors.SmoothedBoxPrior(
math.exp(-1), math.exp(1), sigma=0.1, transform=torch.exp
)
)
)
self.mean_module = means.ConstantMean()
self.grid_bounds = grid_bounds
def __init__(self, num_dim, grid_bounds = (-100., 100.), grid_size = 100):
variational_distribution = variational.CholeskyVariationalDistribution(num_inducing_points = grid_size, batch_shape = torch.Size([num_dim]))
variational_strategy = variational.MultitaskVariationalStrategy(variational.GridInterpolationVariationalStrategy(self, grid_size = grid_size, grid_bounds = [grid_bounds], variational_distribution = variational_distribution, ), num_tasks = num_dim,)
super().__init__(variational_strategy)
self.covar_module = kernels.ScaleKernel(kernels.RBFKernel(lengthscale_priors = priors.SmoothedBoxPrior(math.exp(-1), math.exp(1), sigma = 0.1, transform = torch.exp)))
self.mean_module = means.ConstantMean()
self.grid_bounds = grid_bounds