10.2.2.1. TEGO

The TMaster subclass for EGO is TEGO. Its constructor has two standard arguments, a TDataServer and a TRun pointer. Three kinds of TDataServer can be passed to the class :

• An empty tds with the input TAttribute declared: initial points are random;

• A tds filled with a sampler where input TAttribute are declared: initial points are defined by user but they need to be evaluated.

• A tds filled with a launcher or relauncher where both input and output TAttribute are declared: initial filling phase can be skipped;

The TEGO objects have a method named setSize with two integer arguments: the first one, used in the case of an empty tds, gives the number of random points needed for the construction of the first surrogate model; the second one gives the maximum evaluation number of the expensive code.

Two different solvers can be defined, one for the surrogate model construction (using setModeler method) and one for the next point search (using setSolver method). If they are not defined a default solver is used.

Optimization loop ends when either:

• max number evaluation is reached;

• expected improvement objective is lower than a threshold;

• a Cholesky decomposition problem occurs in the surrogate model construction.

In these version, results are not filtered: all evaluated points are saved in the TDataServer