question about example "Simple_GP_Regression"(question)

[open2333]

Hi there, I am running the example "Simple_GP_Regression", I am using points without noise and I got this result

May I ask how can I get the result that the confidence(variance) is zero at training points(observed data ) like the picture

[david-vicente]

I think that you can try to constrain the noise parameter.

likelihood = gpytorch.likelihoods.GaussianLikelihood(noise_constraint=gpytorch.constraints.Interval(0.0,0.0001))

[david-vicente]

when you force the noise parameter to zero the GP regression will work as an interpolator.

[gpleiss]

In general it's hard to get the noise parameter exactly to zero, as this would cause lots of numerical instabilities. However, if you make it something small (e.g. 1e-6) then you'll get a similar effect.