fitrgp hyperparameter optimization for noisy functions
2 views (last 30 days)
Show older comments
hyperparameter optimization for fitrgp hyperparameters in matlab there is {'BasisFunction','KernelFunction','KernelScale','Sigma','Standardize'} that can be optimized. However, in noisy cases we have a sigma parameter (different than kernel sigma) added to the diagonal of the kernel based on Rasmussen, 2006. How can I specify or optimize that sigma besides sigmaf and sigmal which are the kernel hyperparameters?
0 Comments
Answers (2)
Aditya Patil
on 21 Sep 2020
From my understanding, you want to set the noise variance() parameter mentioned in the book. You can do so using 'Sigma' name value pair. Note that 'Sigma' sets the standard deviation. You can find more about it in the documentation here
0 Comments
Hyunjung Lee
on 4 Dec 2020
The nugget or jitter can be "addressed" by adding a small number to the main diagonal of the correlation matrix.
I agree with Aditya Patil. In fitrgp, that can be done via 'Sigma'. For example,
tau=1e-3; %some small value
gpmodel=fitrgp(X,y,'FitMethod','none','Sigma',tau);
By setting 'FitMethod' to 'none', the nugget term remains as the value you assign.
0 Comments
See Also
Products
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!