ex4x.gms : External Equation - Example 4x

Description

Example contributed by Meta Voelker and Michael Ferris, UW-Madison

Model for finding the smoothing parameter (window width) used to
estimate the density function for a set of DEA efficiency scores.
The density estimate uses a reflected kernel function. The smoothing
parameter is found by maximizing the log likelihood cross-validation
function on the efficiency scores.

Reference: B.W. Sliverman, Density Estimation for Statistics and
Data Analysis, Chapman and Hall, 1986.

The example is similar to ex4 but there are two extra DLLs that
show the use of constant derivatives in Fortran (ex4xf_cb.for) and C
(ex4xc_cb.c). The execution time should be reduced if CONOPT3 is used
as the NLP solver.


Small Model of Type : GAMS


Category : GAMS Test library


Main file : ex4x.gms

$Title  External Equation - Example 4x (EX4X,SEQ=568)

$ontext

  Example contributed by Meta Voelker and Michael Ferris, UW-Madison

  Model for finding the smoothing parameter (window width) used to
  estimate the density function for a set of DEA efficiency scores.
  The density estimate uses a reflected kernel function. The smoothing
  parameter is found by maximizing the log likelihood cross-validation
  function on the efficiency scores.

  Reference: B.W. Sliverman, Density Estimation for Statistics and
  Data Analysis, Chapman and Hall, 1986.

  The example is similar to ex4 but there are two extra DLLs that
  show the use of constant derivatives in Fortran (ex4xf_cb.for) and C
  (ex4xc_cb.c). The execution time should be reduced if CONOPT3 is used
  as the NLP solver.

$offtext


option limrow = 0, limcol = 0;

* original DMU set names
set ALLI /1*950/;
set I(ALLI);
alias (I,II);

* include objective values from optimizing on initial sample
parameter objval(alli);
$gdxin deasolu
$load objval

* use the command-line option "--nDMU=N" if set, else take them all
* Note that the standard GAMS model, ex4, cannot handle all data.
* You can skip this model by defining --noGAMS=1.
* The GAMS solution is also read off the GDX file, in case the GAMS
* run is skipped

$if not set nDMU $set nDMU 999999
scalar n / %nDMU% /;
n = min (n, card(ALLI));
I(ALLI) = (ord(ALLI) le n);

positive variable h       'smoothing parameter';
variable          cv      'log likelihood function value';
h.lo = 0.00001;
scalar hinit / 10.01 /;
scalar fudge /1e-8/;

equation eqn;
equation eqnX;

eqn.. cv =e= sum {i,
                  log (fudge + sum {ii$[not sameas(ii,i)],
                                    exp(-sqr[(objval(i)-objval(ii))/h]/2) +
                                    exp(-sqr[(objval(i)-2+objval(ii))/h]/2)
                                   }
                      )
                 }
             - n*log(h);

eqnX..       1*h + 2*cv =X= 1;

$                             set pre
$ifi %system.filesys%==unix  $set pre 'lib'
$                             set suf '64'

$set N     ex4x
$set c_cbN %pre%%N%c_cb%suf%
$set f_cbN %pre%%N%f_cb%suf%

model %N%      'GAMS implementation'                         / eqn  /;
model %c_cbN%  'External equations in C, with callbacks'     / eqnX /;
model %f_cbN%  'External equations in F77, with callbacks'   / eqnX /;

parameter report(*,*) 'Solution Summary';
$load report
option report:5;
scalar totdist /0/;

file out / 'ex4x.put' /;
put out n:0:0 /;
put fudge:24:15 /;
loop {I,
     put objval(i):23:13 /;
};
putclose;

$onechoV > runme.gms
h.l = hinit;
h.m = 0;
cv.l = 0;
cv.m = 0;
%1.workspace=1000;
solve %1 maximizing cv using nlp;
abort$(%1.solvestat <> 1) 'problems running model %1';

execerror = 0;
report('h', '%1') = h.l;
report('cv','%1') = cv.l;
execerror = 0;
totdist = totdist + abs(h.l  - report('h', 'ex4'));
totdist = totdist + abs(cv.l - report('cv','ex4'));
$offecho

$                             set ext '.dll'
$ifi %system.filesys%==unix  $set ext '.so'
$ifi %system.platform%==dex  $set ext '.dylib'
$ifi %system.platform%==dax  $set ext '.dylib'

$                             set eq
$ifi %system.filesys%==unix  $set eq "'"

$if set runall  $set runC_cb '1' set runF_cb '1'

$ifthen not set nocomp
$  ifi set runC_cb $call gams complink lo=%gams.lo% --lang=c         --files=ex4xc_cb.c                                     --libname=%c_cbN%%ext%
$  if errorlevel 1 $abort Error compiling C Library
$  ifi set runF_cb $call gams complink lo=%gams.lo% --lang=fortran90 --files=%eq%"gehelper.f90 msg2_f.f90 ex4xf_cb.f90"%eq% --libname=%f_cbN%%ext%
$  if errorlevel 1 $abort Error compiling Fortran90 Library
$endif

$                   batinclude runme %N%
$if set runC_cb    $batinclude runme %c_cbN%
$if set runF_cb    $batinclude runme %f_cbN%

if ((totdist < 1.0E-6),
  display "@@@@ #Test passed.",report;
else
  abort totdist, "@@@@ #Test not passed. Inspect ex4.lst for details.",report;
);