How to access kernels’ parameter values in multioutput Gaussian processes and multitask Gaussian processes

[GHU2021]

Hi,

If we use ExactGPModel with for example ScaleKernel and RBFKernel, we can use the following commands to access kernels’ parameter values (values of lengthscale, outputscale) and noise variance:
model.covar_module.base_kernel.lengthscale.item()
model.covar_module.raw_outputscale
model.likelihood.noise.item()

Similarly, if we use MultitaskGPModel or BatchIndependentMultitaskGPModel, how can we access to kernels' parameter values (e.g., if we define the addition of RBF kernel and linear kernel) in these models after we optimized these models?

I appreciate you help.

[gpleiss]

Drop the .item(). E.g.

model.covar_module.base_kernel.lengthscale
model.covar_module.raw_outputscale
model.likelihood.noise

[GHU2021]

Dear Dr. Pleiss,
Thank you for your response.
.item() is necessary because I need to have values of kernel parameters and write them in a file.
Assume that I use MultitaskGPModel as follows:

class MultitaskGPModel(gpytorch.models.ExactGP):
def init(self, train_x, train_y, likelihood, L, M):
super(MultitaskGPModel, self).init(train_x, train_y, likelihood)
self.mean_module = gpytorch.means.MultitaskMean(gpytorch.means.ConstantMean(), num_tasks=L )
self.covar_module = gpytorch.kernels.MultitaskKernel(
gpytorch.kernels.RBFKernel(active_dims=list(range(0,M)))+gpytorch.kernels.RBFKernel(active_dims=list(range(M,2*M))), num_tasks=L, rank=1
)
def forward(self, x):
mean_x = self.mean_module(x)
covar_x = self.covar_module(x)
return gpytorch.distributions.MultitaskMultivariateNormal(mean_x, covar_x)

Then, to check lengthscale value of the first RBF kernel, I should use
model.covar_module.data_covar_module.kernels[0].lengthscale.item()
and to access to noise variance of Multitask GP, I should use
model.likelihood.noise.item()