/cutest-mirror

Permalink
Switch branches/tags
Nothing to show
Find file Copy path
Fetching contributors…
Cannot retrieve contributors at this time
Raw Blame History
191 lines (134 sloc) 7.99 KB
A grand unified Matlab gateway for the CUTEst tools.
This interface brings together the unconstrained, constrained,
dense and sparse versions of the CUTEst tools.
In order to unify the tools and be able to use the same Matlab commands on
both constrained and unconstrained problems, the tool names in this
interface differ from the those in the old Fortran gateway routine.
Getting started
---------------
Make sure you have correctly set the following environment variables:
MYMATLAB directory containing Matlab. Matlab's mex executable
should be found under $MYMATLAB/bin.
MYMATLABARCH CUTEst architecture built with Matlab support.
Let's try and decode a problem and load it up in Matlab:
$ cutest2matlab LUBRIFC # This is the shell prompt
$ matlab &
>> prob = cutest_setup() % This is the Matlab prompt
prob =
n: 3751 % number of variables
m: 2500 % number of constraints
nnzh: 8749 % number of nonzeros in Hessian of Lagrangian
nnzj: 1572494 % number of nonzeros in constraint Jacobian
x: [3751x1 double] % initial guess
bl: [3751x1 double] % lower bounds on variables
bu: [3751x1 double] % upper bounds on variables
v: [2500x1 double] % initial Lagrange multipliers
cl: [2500x1 double] % left-hand side of constraints
cu: [2500x1 double] % right-hand side of constraints
equatn: [2500x1 logical] % array indicating equality constraints
linear: [2500x1 logical] % array indicating linear constraints
name: 'LUBRIFC ' % problem name
>> [f,g] = cutest_obj(prob.x);
If the above does not work as expected for you, please e-mail
dominique.orban@gerad.ca with a description of the commands you run, the
results you get and error messages, if any.
Decoding a problem from inside Matlab
-------------------------------------
At the Matlab prompt, type
>> setenv('DYLD_LIBRARY_PATH', '/usr/local/bin:/opt/local/lib:')
>> !sifdecoder LUBRIFC
The problem should be decoded in the current directory. More programmatically,
you may use constructions such as
>> probname='LUBRIFC';
>> cmd = ['sifdecoder ', probname]; % Note the trailing blank.
>> unix(cmd);
In order to create a MEX file, you can proceed similarly
>> cmd=['runcutest -p matlab -D ', probname];
>> unix(cmd);
Troubleshooting
---------------
If Matlab complains that
Symbol not found: __gfortran_transfer_array_write
Referenced from: /private/tmp/cutest-matlab/mcutest.mexmaci64
Expected in: /Applications/Matlab/MATLAB_R2015b.app/sys/os/maci64/libgfortran.3.dylib
(the paths may look different on your system) the solution consists in editing `.matlab7rc.sh`. On OSX, this file is found in /Applications/Matlab/MATLAB_R2015b.app/bin. Open the file and search for your MEX architecture (e.g., `maci64`) and look for the line
LDPATH_PREFIX=''
Insert the path where your `libgfortran` is found, e.g.,
LDPATH_PREFIX='/usr/local/lib/gcc/5'
and restart Matlab.
Reference
---------
The table below may be obtained with the command 'help cutest' at the
Matlab prompt.
--------------------------------------------------------------------------
Matlab Tool CUTEst library function(s) Purpose
--------------------------------------------------------------------------
dims cdimen Obtain problem dimensions
setup usetup / csetup Setup problem data structure
obj uofg / cofg Evaluate objective function value
and its gradient if requested
grad ugr / cgr Evaluate the objective gradient
objcons cfn Evaluate objective and constraints
cons ccfg / ccifg Evaluate constraint bodies
and their gradients if requested.
Evaluate a single constraint value
and its gradient if requested
scons ccfsg / ccifsg Evaluate constraint bodies and
Jacobian in sparse format.
Evaluate a single constraint value
and its gradient as a sparse vector
lag clfg Evaluate Lagrangian function value
and its gradient if requested
lagjac cgr Evaluate Jacobian and gradient of
either objective or Lagrangian
slagjac csgr Evaluate Jacobian in sparse format
and gradient of either objective or
Lagrangian as a sparse vector
Jprod cjprod Evaluate the matrix-vector product
between the Jacobian and a vector
Jtprod cjprod Evaluate the matrix-vector product
between the transpose Jacobian and
a vector
hess udh / cdh Evaluate the Hessian matrix of the
Lagrangian, or of the objective if
the problem is unconstrained
ihess udh / cidh Evaluate the Hessian matrix of the
i-th problem function (i=0 is the
objective function), or of the
objective if problem is unconstrained
hprod uprod / cprod Evaluate the matrix-vector product
between the Hessian of the
Lagrangian
(or the objective if unconstrained)
and a vector
gradhess ugrdh / cgrdh Evaluate the gradient of either the
objective or the Lagrangian, the
Jacobian (or its transpose) and the
Hessian of the Lagrangian in dense
format
sphess ush / csh Evaluate the Hessian matrix of the
Lagrangian, or of the objective if
the problem is unconstrained, in
sparse format
isphess ush / cish Evaluate the Hessian matrix of the
i-th problem function (i=0 is the
objective function), or of the
objective if problem is
unconstrained, in sparse format
varnames varnames Obtain variable names as strings
connames cnames Obtain constraint names as strings
terminate uterminate / cterminate Remove existing internal workspace
--------------------------------------------------------------------------
* To create the Matlab/CUTEst interface, use the cutest2matlab command or,
for more control, the runcutest command with the -p matlab option.
This will create a binary file mcutest.mexglx (32bit Linux),
mcutest.mexa64 (64bit Linux) mcutest.mexmaci (32bit OSX) or
mcutest.mexmaci64 (64bit OSX).
* Both this binary and the Matlab .m files in $CUTEST/src/matlab must be
placed on the Matlab search path.
* See the man page for runcutest for more details of other options, and the
Matlab help files for the tools mentioned above.
CUTEr version:
D. Orban, Montreal, January 2007
CUTEst version additions:
Nick Gould, January 2013