Data used in a single GPR fit.
We follow sklearn closely. We use the kernel given in Eq.(S3) of arxiv:1809.09125 :
\[ K(x, x') = \sigma_k^2 exp(-1/2 \sum_i^D (x^{i} - x'^{i})^2/\sigma_i^2) \]
where D is the dimension of the model.
Note that Eq.(S3) also includes the WhiteKernel, which has a noise parameter, but we don't need that here since we only need to evaluate \( K_{x* x} \) when evaluating the fits (See the mean value in Eq.(S2) of same paper). The other term we need is alpha = \( K_{x x}^{-1} {\bf f}\), which involves the WhiteKernel, but is precomputed offline. alpha is a vector of size N, where N is the number of cases in the training data set.
Definition at line 60 of file LALSimNRHybSurUtilities.h.
Data Fields | |
REAL8 | constant_value |
\( \sigma_k^2 \) in kernel. More... | |
REAL8 | y_train_mean |
Mean value before GPR fit, usually zero. More... | |
gsl_vector * | length_scale |
\( \sigma_i \) in kernel. More... | |
gsl_vector * | alpha |
Precomputed \( K_{x x}^{-1} {\bf f}\). More... | |
REAL8 GPRHyperParams::constant_value |
\( \sigma_k^2 \) in kernel.
Definition at line 61 of file LALSimNRHybSurUtilities.h.
REAL8 GPRHyperParams::y_train_mean |
Mean value before GPR fit, usually zero.
Definition at line 62 of file LALSimNRHybSurUtilities.h.
gsl_vector* GPRHyperParams::length_scale |
\( \sigma_i \) in kernel.
Definition at line 63 of file LALSimNRHybSurUtilities.h.
gsl_vector* GPRHyperParams::alpha |
Precomputed \( K_{x x}^{-1} {\bf f}\).
Definition at line 64 of file LALSimNRHybSurUtilities.h.