Figure 2.

Hyperparameter calibration procedure

Steps involved in the hyperparameter calibration procedure are (A) the acquisition of a valid dataset (previously collected data, semi-synthetic or synthetic).

(B) The dataset is used to simulate the GP-BO algorithm search procedure using different values of the hyperparameter of interest, each for a given number of repetitions (10 or 50 is typically used). Other hyperparameters are fixed to default values determined as best in previous calibration procedures.

(C) Once the simulations completed, the average performance at the final query is computed across repetitions for models associated with each hyperparameter value.

(D) The performances associated with each hyperparameter values are compared.

(E) The hyperparameter value associated with the highest performance is selected to be used in the online optimization procedure.