ard_num_dims argument doesn't create multi-dimensional GP

[AndrewFalkowski]

Hello,

I am trying to fit a GP with different lengthscales for each feature using the ard_num_dims argument. However, when inspecting the model, I only see a single lengthscale dimension. Is there an additional argument that I am missing here?

class GPModel(gpytorch.models.ExactGP):
    def __init__(self, x_train, y_train, likelihood):
        super(GPModel, self).__init__(x_train, y_train, likelihood)
        self.mean_module = gpytorch.means.ConstantMean()
        self.covar_module = gpytorch.kernels.ScaleKernel(gpytorch.kernels.MaternKernel(nu=2.5, ard_num_dims=4))

    def forward(self, x):
        mean_x = self.mean_module(x)
        covar_x = self.covar_module(x)
        return gpytorch.distributions.MultivariateNormal(mean_x, covar_x)

Calling gp.covar_module.base_kernel.raw_lengthscale shows a single dimension tensor:

Parameter containing:
tensor([[0.]], requires_grad=True)

[AndrewFalkowski]

Fixed the issue!