The following comma-separated files contain detailed results from the application
of DFO solvers to the problems solved in the working paper:
``Derivative-free optimization: A review of algorithms and comparison of software
implementations'' by Luis Miguel Rios and Nikolaos V. Sahinidis.
Results from two types of runs are provided:
1) Problems solved over the entire feasible space.
Ten instances were solved from randomly generated starting points.
(A few solvers discard the starting point and hence provide identical results
in all 10 runs.)
2) Problems solved over a small box in the neighborhood of the global solution.
These instances were solved from a single starting point (near-global solution).
The results are in file names ending in '_local'.
For both types of runs, the following format is used for the results files:
Column 1 'problem' contains the name of the test problem solved
Column 2 'solver' contains the name of the solver used
Column 3 'time' contains the cpu time taken by the solver
Column 4 'solution' contains the best solution found by the solver
Column 5 'calls' contains the number of function calls used by the solver
Column 6 'best call' contains the iteration counter when the best solution was found by the solver
Columns 7-56 contain the best objective function value found by the solver every 50 function calls.
Random starting points:
File 'starting_points.xls' contains the information to generate the starting points.
These starting points were generated using a matrix of [0,1] randomly generated elements.
The matrix can be found in the sheet 'random.numbers'. Each column of the matrix is used
to generate a starting point. That is, the 10 instances used the first 10 columns of the matrix.
Elements in rows are used to generate each of the starting variable's values.
The starting point for variable x_i is obtained using the formula:
x_i = lb_i + y_i (ub_i - lb_i)
where
lb_i, ub_i are the lower and upper bounds for the variable x_i
y_i is the [0,1] randomly generated number at the i^th row
Lower and upper bounds for the variables can be found in the respective problemdata files.