Bounds and starting values for hyperparameters ML estimation; $yN$ is a vector of the $f$-values from the GP training set and $Δy=maxyN-minyN$.
$mμ$ median $f$-value $medyN$ $minyN-2Δy$ $maxyN+2Δy$
$σf2$ 0.5 $exp(-2)$ $exp(25)$
$ℓ$ $exp(-2)$ $exp(25)$
$σn2$ $10-2$ $10-6$ 10
$mμ$ median $f$-value $medyN$ $minyN-2Δy$ $maxyN+2Δy$
$σf2$ 0.5 $exp(-2)$ $exp(25)$
$ℓ$ $exp(-2)$ $exp(25)$
$σn2$ $10-2$ $10-6$ 10