-
Notifications
You must be signed in to change notification settings - Fork 1
A mirror of the SIFDecode SVN repository
License
optimizers/sifdecode-mirror
Folders and files
Name | Name | Last commit message | Last commit date | |
---|---|---|---|---|
Repository files navigation
CUTEst is the latest evolution of CUTE, the constrained and unconstrained testing environment for numerical optimization. To get CUTEst working, here is what you need to do: 0. If are reading this, you have probably got this far, but in any case ... To use CUTEst, you will need to install three packages, ARCHDefs, SIFDecode and CUTEst, as well as yor desired subset of the CUTEst test problems in SIF format. To download ARCHDefs, issue the command svn checkout --username anonymous \ http://ccpforge.cse.rl.ac.uk/svn/cutest/archdefs/trunk ./archdefs in a directory you can write to; use a blank password if prompted. To download SIFDecode, issue the command svn checkout --username anonymous \ http://ccpforge.cse.rl.ac.uk/svn/cutest/sifdecode/trunk ./sifdecode in the same directory. To download CUTEst, svn checkout --username anonymous \ http://ccpforge.cse.rl.ac.uk/svn/cutest/cutest/trunk ./cutest in the same directory. This will produce three sub-directories ./archdefs ./sifdecode ./cutest To obtain problems from the CUTEr/st and related test sets, see http://tracsvn.mathappl.polymtl.ca/trac/cuter/wiki/ProblemCollections or http://www.cuter.rl.ac.uk/Problems/mastsif.shtml 1. Now set the three environment variables $ARCHDEFS, $SIFDECODE and $CUTEST to point at the directories ./archdefs, ./sifdecode and ./cutest respectively. For example, if archdefs lies in the directory /home/rsimpson and you use the C (csh) or TC (tcsh) shell, issue the command setenv ARCHDEFS /home/rsimpson/archdefs (etc) while if you have the Bourne (sh) or Bash (bash) shell, use export ARCHDEFS="/home/rsimpson/archdefs" (etc). To make these permanent, place them in your ~/.cshrc or ~/.bashrc shell startup file. 2. Now, move to the ./cutest directory, enter the command $ARCHDEFS/install_optsuite and follow the instructions. You may install as many machine/os/compiler combinations (architectures) as you wish, and you will have access to each independently of the others. After each installation, read the instructions provided to set further envioronment variables as required. ARCHDEFS <- full path to./archdefs SIFDECODE <- full path to ./sifdecode CUTEST <- full path to ./cutest MYARCH <- the "architecture" you wish to use by default MASTSIF <- the directory containing your SIF test examples and optionally (for Matlab use) MATLAB <- your system Matlab installation directory MYMATLABARCH <- the "architecture" you wish to use with Matlab (if any) as indicated by the installation scripts, to ensure that $SIFDECODE/bin and $CUTEST/bin are on your search path ($PATH$) and that $SIFDECODE/man and $CUTEST/man are on your man path ($MANPATH$). See README.csh (for C-shell users) and README.sh (for Bourne, bash and korn shell users). 3. To test that the installation works, issue the commands cd $SIFDECODE/src ; make -f $SIFDECODE/makefiles/$MYARCH test cd $CUTEST/src ; make -f $CUTEST/makefiles/$MYARCH test 4. Both SIFDecode and CUTEst will be updated whenever we fix a bug or add a new interface. From time to time you may wish to check this by issuing the commands cd $SIFDECODE ; svn update ; cd src ; make -f $SIFDECODE/makefiles/$MYARCH cd $CUTEST ; svn update ; cd src ; make -f $CUTEST/makefiles/$MYARCH This will copy all the updates and recompile both packages. To update on a non-default installed version, replace $MYARCH in the above two lines by the string of the form machine.os.compiler that corresponds to your desired architecture - a list of all installed architectures can be found by issuing the command ls -1 $SIFDECODE/versions for SIFdecode and ls -1 $CUTEST/versions for CUTEst. 5. You may change compiler flags and the like by editing the files $CUTEST/makefiles/$MYARCH $CUTEST/bin/sys/$MYARCH (and the same for $SIFDECODE), and then issuing the commands cd $SIFDECODE/src ; make -f $SIFDECODE/makefiles/$MYARCH cleanall make -f $SIFDECODE/makefiles/$MYARCH cd $CUTEST/ src ; make -f $CUTEST/makefiles/$MYARCH cleanall make -f $CUTEST/makefiles/$MYARCH For non-defaults installations, substiture "machine.os.compiler" for $MYARCH. 6. CUTEst contains interfaces to, ** BUT NOT THE SOURCE CODE FOR ** , a number of popular linear and nonlinear optimization packages. To see which, enter the command ls -1 $CUTEST/package The package "pack" has relevant information recorded in two places 1. the directory $CUTEST/src/"pack" 2. the file $CUTEST/package/"pack" The first contains * the fortran or C interface between CUTEst and the package ("pack"_main.f/f90/c), * a dummy call to the package that serves as a test example ("pack"_test.f/f90/c), * a list of options in a file marked "PACK".SPC, where PACK is the capitalized package name, * a README."pack" that gives details of how to obtain the external package, how and where to install it, and what the options in "PACK".SPC actually mean, and * a makefile component (makemaster). The second lists five variables that must be set to use the external package. The user should check that these make sense, and if necessary edit them to be consistent with the installed version of the package as described in the aforementioned README."pack". 7. Once the package has been installed, it should simply be a matter of issuing the command runcutest -p "pack" -D problem where problem.SIF is any example from the current directory (by default) or $MASTSIF (if the problem is not in the current directory). The runcutest command has a number of options, see man runcutest 8. If you wish to use CUTEst with Matlab, you will need to use a mex interface between CUTEst, Matlab and the SIF problem file you wish to evaluate. To create this link, issue the command cutest2matlab problem for your desired problem.SIF file. This will create a a binary file mcutest.mexglx (32bit Linux), mcutest.mexa64 (64bit Linux) mcutest.mex- maci (32bit OSX) or mcutest.mexmaci64 (64bit OSX). Once you have started Matlab, you must add the mcutest.mex* binary and the contents of the directory $CUTEST/src/matlab to your Matlab search path. Then consult help cutest to find the range of CUTEst Matlab tools available. 9. If you wish to add an interace to your own package, "mypack", you will need to create a directory $CUTEST/src/"mypack" along with files $CUTEST/package/"mypack" $CUTEST/src/"mypack"/"mypack"_main.f/f90/c $CUTEST/src/"mypack"/"mypack"_test.f/f90/c $CUTEST/src/"mypack"/"MYPACK".SPC $CUTEST/src/"mypack"/makemaster and optionally $CUTEST/src/"mypack"/README."pack" The simplest way to do this is to pick an existing package that has rouighly the same functionality, to copy its data to the above destinations, and to modify it for your needs. To help you, have a look at man cutest and the man pages that are referred to to see which evaluation tools are available. You can conult the examples in the directory $CUTEST/src/test to see how each CUTEST tool is called, and those in $CUTEST/package/gen as examples of fortran 77, 90 and C calls. If you have further questions, please contact us. And if you would like us to include your interface in the distribution for others to use, we would be delighted to help you. 10. Aside from the man pages, PDF documentation for each tool and the runcutest, cutest2matlab and sifdecoder scripts is avalable in the directories $SIFDECODE/doc/pdf/ and $CUTEST/doc/pdf. 11. Refer to the CUTEst Wiki http://ccpforge.cse.rl.ac.uk/gf/project/cutest/wiki/ for updates/bug fixes/news. Best of luck Nick Gould Dominique Orban Philippe Toint For GALAHAD productions 30 January 2013 This version: 21 March 2013
About
A mirror of the SIFDecode SVN repository
Resources
License
Stars
Watchers
Forks
Packages 0
No packages published