Program: MKLSZOPTIMDRV
mklszoptimdrv......driver for MKLSGRIDNEWVERTICAL
call: [reliability,efficiency,evalcnt]=mklszoptimdrv(matchthresh,zets,stepx,stepy,drmin,plotmode);
matchthresh: (number) [percent]
lsopt.iter_matchthresh value used by MKLSGRIDNEWVERTICAL
zets: (numeric array) [km]
z values for the initial sampling
The smallest z value in here must be zero.
stepx: (numeric array) [km]
list of depth values, the step function has its jumps at
these depths
stepy: (numeric array) [km]
function values of the step function.
At depth stepx(i) the function jumps to value stepy(i)
drmin: (number) [km]
The minimum allowed distance between function samples: the
search is stopped when the shortest sample distance is shorter
than this.
plotmode: (string) [flag]
string flag to swtich on and off the production of
control plots during tests.
possible values: 'on' and 'off'
result: reliability: (number) [percent]
number of samples with maximum function value found
by MKLSGRIDNEWVERTICAL algorithm in relation to the
number found by unform sampling
A reliability of e.g. 40% means that
MKLSGRIDNEWVERTICAL returned only 40% as many
maximum points as the uniform grid search did. This
means that it missed some solutions.
The reliability therefore has to be maximized.
efficiency: (number) [percent]
number of function evaluations used by MKLSGRIDNEWVERTICAL
divided by the number of function evaluations in a uniform
sampling with the same maximal grid density.
An efficiency of e.g. 40% means that
MKLSGRIDNEWVERTICAL used only 100%-40%=60% of the function
evaluations that were used in the uniform grid
search. This means it saved 40% of the computational
cost.
The efficiency therefore has to be maximized.
evalcnt: (number) [counter]
number of function evaluations.
Obviously, there is a trade off between raliability and efficieny:
you can be very efficient when you miss a lot of the solutions...
This function drives the optimization of a step function by
MKLSGRIDNEWVERTICAL and evaluates the reliability and efficieny of the sampling
algorithm.
For the evaluation, a uniform grid search is conducted. The sampling
density of this search is equal to the highest density achieved in the
non-uniform search. The number of samples with maximal function value and
the number of function evaluations are compared between the two methods.
Martin Knapmeyer, 30.09.2005, 28.04.2006
Read M-File Source Code
eof.