Table 5. Recommended hyperparameter... | Emerald Publishing

Table 5.

Recommended hyperparameter settings

Hyperparameter	Recommended setting	Notes
Utility function	Upper confidence bound (UCB)	Expected improvement (EI) also performs well, but use $ξ ≲ 10^{- 3}$
κ	2.0	Minor efficiency gains can be made by lowering κ, but at the risk of becoming stuck
ν	$\frac{5}{2}$	$ν = \frac{3}{2}$ also performs well
l	0.1	Since this is just the initial guess for the Gaussian process optimizer, the choice for l is often not significant
n_s	10	For very simple optimization problems (e.g. single variable, under 10 iterations total), n_s can be lowered, at the risk of becoming stuck

Hyperparameter	Recommended setting	Notes
Utility function	Upper confidence bound(UCB)	Expected improvement (EI) also performs well, but use $ξ ≲ 10^{- 3}$
κ	2.0	Minor efficiency gains can be made by lowering κ, but at the risk of becoming stuck
ν	$\frac{5}{2}$	$ν = \frac{3}{2}$ also performs well
l	0.1	Since this is just the initial guess for the Gaussian process optimizer, the choice for l is often not significant
n_s	10	For very simple optimization problems (e.g. single variable, under 10 iterations total), n_s can be lowered, at the risk of becoming stuck

Source:

Author’s own work