|
|
|
ego
The Efficient Global Optimization (EGO) algorithm
solves costly box-bounded global optimization problems with additional linear, nonlinear and integer constraints.
The idea of the EGO algorithm
is to first fit a response surface to data collected by evaluating the
objective function at a few points. Then, EGO balances between finding
the minimum of the surface and improving the approximation by sampling
where the prediction error may be high.
Main features
-
EGO implements the algorithm EGO by D. R. Jones,
Matthias Schonlau and William J. Welch:
Efficient Global Optimization of Expensive Black-Box Functions,
Journal of Global Optimization, 13:455-492, 1998.
-
EGO is extended to handle
noncostly linear and nonlinear constraints.
-
Costly constraints could be treated by adding these with
penalties to the objective function.
-
EGO saves all information from the run on the file
cgoSave.mat.
This makes warm start possible.
The solver rbfSolve
is using the same format and file.
This makes it possible to run any number of iterations and combinations
using both the solvers
rbfSolve and ego.
-
EGO could be started with:
-
1. the users own set of starting points.
-
2. all corner points, optionally adding the mid point.
-
3. some corner points (Gutmann), optionally adding the mid point.
-
4. a latin hypercube experimental design procedure (default).
-
5. a new ellipsoid experimental design algorithm (Holmström).
-
6. a warm start with points from file.
-
EGO is using four types of function transformation. Default EGO tries
to find the best possible transformation from the initial set of data.
|
|