1Blurb:: 2Select the built in Gaussian Process surrogate 3Description:: 4A second version of GP surrogates was available in prior versions of 5Dakota. 6<b> For now, both versions are supported but the \c dakota version is 7deprecated and intended to be removed in a future release. </b> 8 9Historically these models were drastically different, 10but in %Dakota 5.1, they became quite similar. They now differ in that 11the Surfpack GP has a richer set of features/options and tends to be 12more accurate than the %Dakota version. Due to how the Surfpack GP 13handles ill-conditioned correlation matrices (which significantly 14contributes to its greater accuracy), the \c Surfpack GP can be a 15factor of two or three slower than %Dakota's. As of %Dakota 5.2, the 16Surfpack implementation is the default in all contexts except Bayesian 17calibration. 18 19More details on the \c gaussian_process dakota 20model can be found in \cite McF08. 21 22%Dakota's GP 23deals with ill-conditioning in two ways. First, when it encounters a 24non-invertible correlation matrix it iteratively increases the size of 25a "nugget," but in such cases the resulting approximation smooths 26rather than interpolates the data. Second, it has a \c point_selection 27option (default off) that uses a greedy algorithm to select a 28well-spaced subset of points prior to the construction of the GP. In 29this case, the GP will only interpolate the selected 30subset. Typically, one should not need point selection in trust-region 31methods because a small number of points are used to develop a 32surrogate within each trust region. Point selection is most 33beneficial when constructing with a large number of points, typically 34more than order one hundred, though this depends on the number of 35variables and spacing of the sample points. 36 37This 38differs from the \c point_selection option of the %Dakota GP which 39initially chooses a well-spaced subset of points and finds the 40correlation parameters that are most likely for that one subset. 41 42 43Topics:: 44Examples:: 45Theory:: 46Faq:: 47See_Also:: 48