chiark / gitweb /
stevenj [Sun, 4 Apr 2010 21:48:51 +0000 (17:48 -0400)]
use opt->dx in local sub-algorithm
darcs-hash:
20100404214851-c8de0-
5e39ad9719b7c5068a704a83e4907e98950c4065.gz
stevenj [Sun, 4 Apr 2010 20:31:47 +0000 (16:31 -0400)]
stevenj [Sat, 3 Apr 2010 16:00:38 +0000 (12:00 -0400)]
new, extensible "object-oriented" API, first draft
darcs-hash:
20100403160038-c8de0-
232c66be6bea9251b847be53380e00a5952f695d.gz
stevenj [Fri, 2 Apr 2010 22:50:04 +0000 (18:50 -0400)]
use stdcall interface by default on Windows, since this seems to be more standard there; thanks to Alan Young for the suggestion
darcs-hash:
20100402225004-c8de0-
40e1743d311901ea1ccf0dea3588d654474b050d.gz
stevenj [Mon, 15 Mar 2010 22:31:59 +0000 (18:31 -0400)]
nlopt-util.h not needed for mex file
darcs-hash:
20100315223159-c8de0-
aba750ebeed8a30dbfc442fa993fc414eb321082.gz
stevenj [Tue, 9 Feb 2010 22:47:41 +0000 (17:47 -0500)]
use NLopt stop criteria in Luksan code: in particular, absolute tolerances were missing (thanks to Greg Nicholas for the bug report)
darcs-hash:
20100209224741-c8de0-
3df6ca15549dfe15027bc582948587adbcaa3b29.gz
stevenj [Fri, 29 Jan 2010 06:00:46 +0000 (01:00 -0500)]
fixed to compile under C++, use C++ compiler in --with-cxx mode (thanks to Greg Nicholas for the suggestion)
darcs-hash:
20100129060046-c8de0-
eaf1103383deae6747c025b25cfd02bedd2cdb1a.gz
stevenj [Fri, 29 Jan 2010 06:02:56 +0000 (01:02 -0500)]
stevenj [Fri, 29 Jan 2010 05:43:34 +0000 (00:43 -0500)]
stevenj [Wed, 27 Jan 2010 22:49:36 +0000 (17:49 -0500)]
stevenj [Wed, 27 Jan 2010 22:43:35 +0000 (17:43 -0500)]
stevenj [Tue, 1 Dec 2009 16:48:34 +0000 (11:48 -0500)]
stevenj [Tue, 1 Dec 2009 16:42:33 +0000 (11:42 -0500)]
stevenj [Thu, 19 Nov 2009 01:52:41 +0000 (20:52 -0500)]
stevenj [Thu, 19 Nov 2009 01:51:24 +0000 (20:51 -0500)]
stevenj [Thu, 19 Nov 2009 00:47:24 +0000 (19:47 -0500)]
added get/set stochastic_population functions for greater control over stochastic algorithms
darcs-hash:
20091119004724-c8de0-
3ae4d1219ac30c2add587bcfb1bb7d029f0bc4a2.gz
stevenj [Thu, 19 Nov 2009 00:23:12 +0000 (19:23 -0500)]
isres should ignore change in f and x on first evaluation
darcs-hash:
20091119002312-c8de0-
40fd94fd0d8d0dd456c2a51e5a20608b70e9ca07.gz
stevenj [Thu, 19 Nov 2009 00:10:16 +0000 (19:10 -0500)]
slightly better way of deciding on "best" solution so far in ISRES, still not entirely satisfactory
darcs-hash:
20091119001016-c8de0-
df385d85010eeb3a3f9220bfbfaa5dc44b35bc8e.gz
stevenj [Wed, 18 Nov 2009 16:41:55 +0000 (11:41 -0500)]
stevenj [Wed, 18 Nov 2009 06:18:57 +0000 (01:18 -0500)]
bug fix in ISRES; it seems to work, albeit worse than CRS2
darcs-hash:
20091118061857-c8de0-
60c4151d03f119b56a96db1f830c23c06602cee3.gz
stevenj [Wed, 18 Nov 2009 05:57:31 +0000 (00:57 -0500)]
make sure minf is initialized to HUGE_VAL before calling any optimization routine
darcs-hash:
20091118055731-c8de0-
e3ad0a0f08aa9c4eadca0572f9b43dae7131ce85.gz
stevenj [Wed, 18 Nov 2009 05:52:59 +0000 (00:52 -0500)]
added ISRES (compiles, but untested)
darcs-hash:
20091118055259-c8de0-
1b704e09a6a759e72345cdd1833c9756e3880663.gz
stevenj [Tue, 17 Nov 2009 21:20:16 +0000 (16:20 -0500)]
stevenj [Tue, 17 Nov 2009 21:17:11 +0000 (16:17 -0500)]
fixes to nlopt stop cases in BOBYQA
darcs-hash:
20091117211711-c8de0-
f08774fcd3cfefe2d33029d6561b60d51b5cae6a.gz
stevenj [Tue, 17 Nov 2009 20:49:52 +0000 (15:49 -0500)]
stevenj [Tue, 17 Nov 2009 20:34:03 +0000 (15:34 -0500)]
add preliminary BOBYQA (not quite working yet)
darcs-hash:
20091117203403-c8de0-
a3f3bb4d9d9a2b3dbe031f9f16221e1b8306575f.gz
stevenj [Sat, 14 Nov 2009 03:05:21 +0000 (22:05 -0500)]
fixed extraneous text accidentally included in nlopt.f FOrtran include file
darcs-hash:
20091114030521-c8de0-
ee88f0b9f3dec44f56369dae6c99604c3c754bde.gz
stevenj [Fri, 13 Nov 2009 00:54:24 +0000 (19:54 -0500)]
stevenj [Fri, 13 Nov 2009 00:47:51 +0000 (19:47 -0500)]
stevenj [Fri, 13 Nov 2009 00:43:18 +0000 (19:43 -0500)]
add (disabled) code to test program to signal on NaN creation, for debugging
darcs-hash:
20091113004318-c8de0-
9d518702b3a469dfd45197b08f06f255c5175f06.gz
stevenj [Fri, 13 Nov 2009 00:43:08 +0000 (19:43 -0500)]
added new NLOPT_ROUNDOFF_LIMITED failure code to indicate when optimizer breaks down due to roundoff errors (which may still indicate a useful result); several checks for breakdown in NEWUOA; bug fix in NEWUOA_BOUND that allowed roundoff errors to push x slightly outside bound constraints
darcs-hash:
20091113004308-c8de0-
8d292e1d67c23c851d98974239cfbfddda16a35e.gz
stevenj [Thu, 12 Nov 2009 22:12:19 +0000 (17:12 -0500)]
if reltol > 0, catch convergence case where new == old == 0
darcs-hash:
20091112221219-c8de0-
04e133b083c9f326593ad295b7309bf931c16bc4.gz
stevenj [Thu, 12 Nov 2009 17:46:47 +0000 (12:46 -0500)]
added double Lorentzian test function (not distributed)
darcs-hash:
20091112174647-c8de0-
33b9fc6d95a397a6ee3239ca8cf5a19b1ac46486.gz
stevenj [Thu, 12 Nov 2009 17:46:19 +0000 (12:46 -0500)]
added Box test function (not distributed)
darcs-hash:
20091112174619-c8de0-
ac18ec39b3066b38a10863df20813946365aa97f.gz
stevenj [Thu, 29 Oct 2009 02:17:37 +0000 (22:17 -0400)]
configure script detects whether Matlab/Octave plugins require nlopt to be compiled as a shared library
darcs-hash:
20091029021737-c8de0-
911c87b2d91fe774b55bceaabc76572d54f9dae4.gz
stevenj [Thu, 29 Oct 2009 01:18:05 +0000 (21:18 -0400)]
stevenj [Thu, 29 Oct 2009 01:16:41 +0000 (21:16 -0400)]
updated Octave install-directory detection for recent Octave versions
darcs-hash:
20091029011641-c8de0-
0c02209397ba9a1e9f38b883a60a0e49ad6af6ad.gz
stevenj [Thu, 29 Oct 2009 01:01:01 +0000 (21:01 -0400)]
stevenj [Thu, 29 Oct 2009 00:53:22 +0000 (20:53 -0400)]
work around stupid qsort_r BSD/GNU incompatibility
darcs-hash:
20091029005322-c8de0-
bc99d7ce1efdc63f5b29848d2e49c46ace31a1f4.gz
stevenj [Thu, 27 Nov 2008 20:37:09 +0000 (15:37 -0500)]
added test function with minimum at side (i.e. one active bound constraint), which is more challenging than the corner case
darcs-hash:
20081127203709-c8de0-
248be9ce8ebbedc63edea548940c330579979276.gz
stevenj [Thu, 27 Nov 2008 19:34:51 +0000 (14:34 -0500)]
modify cobyla to explicitly honor bound constraints
darcs-hash:
20081127193451-c8de0-
1f3c16ad9d01ecd63751bc86004ef59229a8c855.gz
stevenj [Wed, 26 Nov 2008 23:12:16 +0000 (18:12 -0500)]
stevenj [Wed, 26 Nov 2008 22:41:10 +0000 (17:41 -0500)]
stevenj [Wed, 26 Nov 2008 22:40:15 +0000 (17:40 -0500)]
--without-octave and --without-matlab options
darcs-hash:
20081126224015-c8de0-
d8d9bd5711d65aac4dd5fbc6a5740a87a3a3d24a.gz
stevenj [Wed, 26 Nov 2008 04:23:30 +0000 (23:23 -0500)]
initial (undocumented) support for equality constraints via augmented Lagrangian method
darcs-hash:
20081126042330-c8de0-
c6074b36c1bf1a41005c1b1526ce801659db703a.gz
stevenj [Tue, 25 Nov 2008 19:37:07 +0000 (14:37 -0500)]
skeleton of future equality-constraint support
darcs-hash:
20081125193707-c8de0-
5a57e367a994ab4bd65f667197312bfd4e8c0a02.gz
stevenj [Sat, 22 Nov 2008 22:29:58 +0000 (17:29 -0500)]
stevenj [Sat, 22 Nov 2008 21:29:30 +0000 (16:29 -0500)]
better checking for Matlab installation path
darcs-hash:
20081122212930-c8de0-
ed002085375fe8493a94b3a880fd0f99da55c7a2.gz
stevenj [Fri, 14 Nov 2008 01:04:49 +0000 (20:04 -0500)]
disable Octave code if mkoctfile not found, even if OCT_INSTALL_DIR specified
darcs-hash:
20081114010449-c8de0-
49d3266da4b839ebe49046f614614e5680a2dca7.gz
stevenj [Thu, 13 Nov 2008 21:11:51 +0000 (16:11 -0500)]
stevenj [Thu, 13 Nov 2008 20:52:25 +0000 (15:52 -0500)]
MLSL needs a nonzero default ftol_rel and/or xtol_rel
darcs-hash:
20081113205225-c8de0-
d38cc7ac93ec94d3d479f2776154b4d59b51f876.gz
stevenj [Thu, 13 Nov 2008 19:56:30 +0000 (14:56 -0500)]
fix overriding of MEX_INSTALL_DIR
darcs-hash:
20081113195630-c8de0-
b2f6dee660bd644a90671e1b84a44c296669c392.gz
stevenj [Thu, 13 Nov 2008 17:08:06 +0000 (12:08 -0500)]
yikes, bug fix in DIRECT that prevented convergence for Shekel-10 (DIRECT-L is not affected)
darcs-hash:
20081113170806-c8de0-
77c0694921154f382921dca06507d02b39d2fe71.gz
stevenj [Wed, 12 Nov 2008 03:28:33 +0000 (22:28 -0500)]
whoops, removed obsolete Makefile
darcs-hash:
20081112032833-c8de0-
48c4230bceed758dcc0fdde3851b7d52fb7db1b7.gz
stevenj [Wed, 12 Nov 2008 03:24:29 +0000 (22:24 -0500)]
stevenj [Wed, 12 Nov 2008 03:23:22 +0000 (22:23 -0500)]
remove BUGS section of man pages
darcs-hash:
20081112032322-c8de0-
c682036baf55cdc4d16a10f78518013e7a8f57c1.gz
stevenj [Tue, 11 Nov 2008 22:52:06 +0000 (17:52 -0500)]
more conservative ftol test in subplex, based on max delta f in simplices, to avoid false positives in early iterations where the stepssize is too big to make progress
darcs-hash:
20081111225206-c8de0-
28b9f96be4888599e1201b849664e6dc402d640d.gz
stevenj [Tue, 11 Nov 2008 22:35:24 +0000 (17:35 -0500)]
call stop_ftol instead of stop_f if we check minf_max elsewhere
darcs-hash:
20081111223524-c8de0-
9267e6cc8db7693ebc0060f11c831b66e6c9903c.gz
stevenj [Tue, 11 Nov 2008 22:27:35 +0000 (17:27 -0500)]
fix dependency (rebuild Octave & Matlab plugins if libnlopt changes)
darcs-hash:
20081111222735-c8de0-
f28c31d2fcbd2909e43b7254532f9838733413d7.gz
stevenj [Tue, 11 Nov 2008 22:26:52 +0000 (17:26 -0500)]
improve MMA for infeasible starting point -- new points should be preferred over old infeasible ones as long as the infeasibilit decreases
darcs-hash:
20081111222652-c8de0-
edd440d99bc11a1724b26307f03d47eb6cca4544.gz
stevenj [Tue, 11 Nov 2008 22:10:32 +0000 (17:10 -0500)]
bug fix in MMA for infeasible starting point -- feasible points always take precedence over infeasible ones
darcs-hash:
20081111221032-c8de0-
c23046ac2c38fd7781ba3d7b1e6334126ca9e344.gz
stevenj [Tue, 11 Nov 2008 06:34:00 +0000 (01:34 -0500)]
stevenj [Tue, 11 Nov 2008 06:08:07 +0000 (01:08 -0500)]
stevenj [Tue, 11 Nov 2008 05:57:23 +0000 (00:57 -0500)]
stevenj [Tue, 11 Nov 2008 05:20:39 +0000 (00:20 -0500)]
work around NaN in cobyla that occurs if we run it for too many iterations past the point where it has converged to machine precision
darcs-hash:
20081111052039-c8de0-
6c6f2e7b8ec8e368aa1d96eed86b33b11bf257f7.gz
stevenj [Tue, 11 Nov 2008 04:50:39 +0000 (23:50 -0500)]
stevenj [Tue, 11 Nov 2008 02:41:21 +0000 (21:41 -0500)]
first stab at nlopt_minimize_constrained Matlab/Octave wrappers
darcs-hash:
20081111024121-c8de0-
db5892685407b546fc41139d3e2034cd39b898c9.gz
stevenj [Tue, 11 Nov 2008 01:37:10 +0000 (20:37 -0500)]
stevenj [Tue, 11 Nov 2008 00:57:39 +0000 (19:57 -0500)]
fix description of constrained nelder-mead
darcs-hash:
20081111005739-c8de0-
8919dcdbd8209909021bd4148471600a6430ad2b.gz
stevenj [Mon, 10 Nov 2008 22:58:41 +0000 (17:58 -0500)]
stevenj [Mon, 10 Nov 2008 22:40:44 +0000 (17:40 -0500)]
remove Rowan's original subplex (non-free) from source, document sbplx and neldermead
darcs-hash:
20081110224044-c8de0-
9cb8a9dbebf2e7d1da9e5979c72018d9afd932e2.gz
stevenj [Mon, 10 Nov 2008 22:03:44 +0000 (17:03 -0500)]
stevenj [Mon, 10 Nov 2008 21:58:47 +0000 (16:58 -0500)]
don't use fp equality in coincident-point check, fix bug that allowed points to go outside bounds
darcs-hash:
20081110215847-c8de0-
eec3ad3d30f6ab534a236d1af75326a6a9a69e3e.gz
stevenj [Mon, 10 Nov 2008 20:51:43 +0000 (15:51 -0500)]
check for coincident points in simplex
darcs-hash:
20081110205143-c8de0-
500fc85b33d2d417b1001f53ae87a4e109384de4.gz
stevenj [Mon, 10 Nov 2008 02:10:39 +0000 (21:10 -0500)]
keep track of max/min evaluation counts
darcs-hash:
20081110021039-c8de0-
3815b01480dc472b75da7fbeb8846e2a2189fefc.gz
stevenj [Sun, 9 Nov 2008 18:47:20 +0000 (13:47 -0500)]
stevenj [Sun, 9 Nov 2008 18:43:17 +0000 (13:43 -0500)]
added internal nldrmd_minimize_ for use in subplex
darcs-hash:
20081109184317-c8de0-
d5f6b588c3e9ff822b50b221d3b9f432f5a748c8.gz
stevenj [Sun, 9 Nov 2008 18:19:16 +0000 (13:19 -0500)]
fixed nelder-mead xtol/ftol stopping criteria, added diameter reduction test for subplex
darcs-hash:
20081109181916-c8de0-
05b8623dcf3365df05997d136b849d8c02b6adc4.gz
stevenj [Sun, 9 Nov 2008 06:15:14 +0000 (01:15 -0500)]
stevenj [Sat, 8 Nov 2008 21:56:21 +0000 (16:56 -0500)]
added nelder-mead simplex algorithm (in preparation for re-implementing subplex, and possibly recent provably convergent variants of nelder-meade)
darcs-hash:
20081108215621-c8de0-
91e9b421d37162aa2b8f4013f1c88a4003d05215.gz
stevenj [Sat, 8 Nov 2008 06:20:03 +0000 (01:20 -0500)]
fixes for latest autoconf version
darcs-hash:
20081108062003-c8de0-
5793892e3a4f77cbd51397325a5ddb747ab579ea.gz
stevenj [Tue, 21 Oct 2008 20:00:14 +0000 (16:00 -0400)]
handle infeasible initial point in MMA with finite upper bound on dual variables
darcs-hash:
20081021200014-c8de0-
5b4873094814440d9a34eb1a59693b2525c4158a.gz
stevenj [Thu, 2 Oct 2008 22:26:52 +0000 (18:26 -0400)]
support for disabling constraints in MMA by returning NaN
darcs-hash:
20081002222652-c8de0-
17d74a76bfe5a0654cf7f42e79a76a33eeb58ddc.gz
stevenj [Mon, 15 Sep 2008 21:14:31 +0000 (17:14 -0400)]
unset mma_verbose for recursive MMA of dual
darcs-hash:
20080915211431-c8de0-
0832386c70f3139b990f11683baaf19cd90079be.gz
stevenj [Mon, 1 Sep 2008 20:04:58 +0000 (16:04 -0400)]
only print averages when multiple iterations
darcs-hash:
20080901200458-c8de0-
6caef5668ed83a0897ef492107ee36322d18d1ca.gz
stevenj [Mon, 1 Sep 2008 19:57:56 +0000 (15:57 -0400)]
added some more decimal places to known minima
darcs-hash:
20080901195756-c8de0-
c398c7c620fd72e7e2691cda70f9f899d7edd1a4.gz
stevenj [Mon, 1 Sep 2008 06:47:50 +0000 (02:47 -0400)]
stevenj [Mon, 1 Sep 2008 06:40:50 +0000 (02:40 -0400)]
default to COBYLA rather than subplex for local searches, since the latter doesn't handle bound constraints well
darcs-hash:
20080901064050-c8de0-
29a9cca7bc4ac2beec2976d64c1ed7fea96c97f4.gz
stevenj [Mon, 1 Sep 2008 06:36:05 +0000 (02:36 -0400)]
check bounds in testopt, fix bug in NEWUOA_BOUNDS that allowed it to go outside of the bounds in one case
darcs-hash:
20080901063605-c8de0-
466debf023ac975267a65a9f125ada08c771fb49.gz
stevenj [Mon, 1 Sep 2008 06:10:03 +0000 (02:10 -0400)]
stevenj [Mon, 1 Sep 2008 03:27:52 +0000 (23:27 -0400)]
first start at replacing original newuoa routines with nlopt versions that will allow bound constraints
darcs-hash:
20080901032752-c8de0-
2f993fc0527b8c9f04c2e7dccc00ccd281ef0bc5.gz
stevenj [Mon, 1 Sep 2008 03:26:45 +0000 (23:26 -0400)]
whoops, forgot to add NEWUOA code
darcs-hash:
20080901032645-c8de0-
1465f42dcbcffcba847197f52233c4657fdeae4e.gz
stevenj [Sun, 31 Aug 2008 18:11:21 +0000 (14:11 -0400)]
added corner optimization test case, average #evaluations
darcs-hash:
20080831181121-c8de0-
84e1ad830abe38d212aca8b8791312d30b105489.gz
stevenj [Sat, 30 Aug 2008 10:27:37 +0000 (06:27 -0400)]
added NEWUOA, unified starting step-size for derivative-free algorithms
darcs-hash:
20080830102737-c8de0-
2f48550a7c042db585110d7a05b566c02fc7b892.gz
stevenj [Fri, 29 Aug 2008 20:28:43 +0000 (16:28 -0400)]
stevenj [Fri, 29 Aug 2008 03:27:25 +0000 (23:27 -0400)]
stevenj [Fri, 29 Aug 2008 03:12:33 +0000 (23:12 -0400)]
stevenj [Thu, 28 Aug 2008 01:55:58 +0000 (21:55 -0400)]
first stab at experimental cquad algorithm; doesn't work well yet, so not in make dist
darcs-hash:
20080828015558-c8de0-
1f110c33c323e6ecb62c43799d05f8504d33c924.gz
stevenj [Wed, 27 Aug 2008 18:11:48 +0000 (14:11 -0400)]
stevenj [Wed, 27 Aug 2008 16:54:39 +0000 (12:54 -0400)]