I'm using GpyTorch to model gaussian processes. Since I'm new to this tool I adapted the doc tutorial in just a few ways to fit my training data.
Unfortunetly my confidence region is for every region of my x-axis the same height. Meaning that even on regions of my x-axis with a lot of uncertainty (for example around 0.25, (https://i.stack.imgur.com/HvBwU.png)), the confidence region is not bigger than in areas with lower uncertainty. The model might has diffrent issues, but I managed to fix most of them.
I tried to get a diffrent confidence regions with multiple combinations from diffrent models and kernels and likelihoods, but none of them worked reliably.
In the screenshot case I used:
gpytorch.models.ExactGP
gpytorch.means.ConstantMean()
gpytorch.kernels.AdditiveKernel(gpytorch.kernels.RBFKernel(eps = 10))
gpytorch.distributions.MultivariateNormal()
gpytorch.likelihoods.GaussianLikelihood()
gpytorch.mlls.ExactMarginalLogLikelihood(likelihood, model)
Any idea how I can bring diffrent confidence regions into the model? Since there are diffrent combinations of model, kernel, likelihood and so on, which parameters have an impact on the confidence interval?
Already a big THANKS!