[Docs] GP Regression with Uncertain Inputs

[vijay092]

📚 Documentation/Examples

Hey guys,

Could you please clarify how we could extend GP regression with uncertain inputs to a dataset with multiple features? i.e., when train_x_stdv is matrix at every data point.

I am referring to this example.

train_x_stdv = torch.linspace(0.03, 0.01, 20) will now be a matrix at every datapoint.

How is train_x_distributional = torch.stack((train_x_mean, (train_x_stdv**2).log()), dim=1) modified?

[island-polyu]

Hey, I wonder whether you have solved this problem, because I have the same problem.Thank you.

[gpleiss]

train_x_stdv = torch.linspace(0.03, 0.01, 20) will now be a matrix at every datapoint.

I don't know if this kernel support matrix uncertainties, but it should probably work with a vector of uncertainties. Try setting train_x_stdv to a NxD matrix.