Skip to main content

Table 2 Summary of the 9 algorithms tested: name, space over which it is defined (mixed versus continuous with latent variables), metamodel used, acquisition criterion, optimizer of the acquisition criterion

From: A comparison of mixed-variables Bayesian optimization approaches

Name

Formulation

Metamodel

Acq. crit.

Optimizer of the acq. crit.

LV-EGO

LV

GP

EI

Restarted COBYLA

LV-RFO

LV

randomForest toolbox

EI

Focus-search (from mlrMBO)

ALV-EGO-ge or -gi

LV

GP

EI

DoE (for \(\lambda _t\) and \(\rho _t\)) and restarted COBYLA

ALV-EGO-le or -li

LV

GP

EI

Restarted COBYLA

MS-RFO

MS

randomForest toolbox

EI

Focus-search (from mlrMBO)

MS-ES

MS

none

\(-y(x,u)\)

Evolution strategy (from Li et al. [15] in CEGO implementation Zaefferer [30])

MS-MKES

MS

GP (sym. compound disc. kernel)

EI

Evolution strategy (from Li et al. [15] in CEGO implementation Zaefferer [30])