I am using different machine learning models to model a noisy dataset for some study. I came across fitrgp model in MATLAB to model the data using gaussian process regression. I am also using dacefit functions to model same data using kriging. I am bit confused about the difference between both, as I understand that models interpolate the data by modeling residuals using kernel.

When I evaluated my training error for both models, I noticed kriging to interpolate (trainset error ~ 1e-14) as expected but GP regression model gives much higher training error (~1e-2). Does this mean that the fitrgp model in MATLAB does regression actually (as it does not capture the noise in data completely). If possible, can someone please explain the difference between both these models in simple terms? Thanks in advance!

  • $\begingroup$ +1 -- but some narrowing of this question would be welcome, because "Kriging" is a broad family of procedures rather than a specific model. What kind of Kriging do you have in mind? You might especially consider the issue of how the variogram (or covariance) is estimated. $\endgroup$ – whuber Mar 13 '20 at 13:11
  • $\begingroup$ fitrgp has many options; I'm less familiar with dacefit. Normally I've thought of fitrgp as both a type of GPR and a type of Kriging (i.e. interchangeable). $\endgroup$ – Sterling Dec 7 '20 at 15:44

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.