Gaussian process regression

SMT implements several surrogate models related to Gaussian process regression:

• KRG implements classic gaussian process regression.

• KPLS and KPLSK are variants using PLS dimension reduction to address high-dimensional training data.

• GPX is a re-implementation of KRG and KPLS using Rust for faster training/prediction operations.

• GEKPLS leverages on derivatives training data to improve the surrogate model quality.

• MGP takes into account the uncertainty of the hyperparameters defined as a density function.

• SGP implements sparse methods allowing to deal with large training dataset as others implementations have a time complexity of \(O(n^3)\) as well as a \(O(n^2)\) memory cost in the number \(n\) of training points.

• CCKRG implements cooperative components Kriging, a way of fitting a high-dimensional ordinary Kriging model by sequential lower-dimensional component model fits. For each component, only the associated hyperparameters are optimized. All other hyperparameters are set to a so-called cooperative context vector, which contains the current best hyperparameter values.

Here below, the links to the dedicated pages:

• Kriging (KRG)
• KPLS
• KPLSK
• GPX
• GEKPLS
• Marginal Gaussian Process (MGP)
• Sparse Gaussian Process (SGP)
• Cooperative Components Kriging (CoopCompKRG)