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.