chiark / gitweb /
stevenj [Fri, 20 Jul 2012 20:20:35 +0000 (16:20 -0400)]
stevenj [Fri, 20 Jul 2012 20:14:33 +0000 (16:14 -0400)]
stevenj [Fri, 20 Jul 2012 18:19:42 +0000 (14:19 -0400)]
stevenj [Fri, 20 Jul 2012 18:13:45 +0000 (14:13 -0400)]
stevenj [Fri, 20 Jul 2012 18:04:19 +0000 (14:04 -0400)]
just unconditionally disable compilation of matlab & octave plugins unless --enable-shared is turned on, since test was unreliable and the 32-bit systems where static libraries worked are disappearing anyway
Ignore-this:
27823a53f73a6c56ad7c0312c950b518
darcs-hash:
20120720180419-c8de0-
c22087237d77b38bcbfe44bd7780c11fec9fb549.gz
stevenj [Mon, 16 Jan 2012 14:49:19 +0000 (09:49 -0500)]
in Matlab/Octave interface, make returning NaN from the objective equivalent to an nlopt_force_stop; thanks to Norman Violet for the suggestion
Ignore-this:
7aacd84c708a986dce9f793c08c8ec65
darcs-hash:
20120116144919-c8de0-
a067fa3d0dab251b263603c122f1038b636c34ac.gz
stevenj [Thu, 8 Dec 2011 02:00:30 +0000 (21:00 -0500)]
bug fix to convergence test (only relevant to sbplx); thanks to Douglas Bates
Ignore-this:
d163aaae150e0a9c8ae268109abdf48
darcs-hash:
20111208020030-c8de0-
0b3b621b4fbefb980ec7188f0a5be99a5b1d9adc.gz
stevenj [Mon, 28 Nov 2011 22:42:47 +0000 (17:42 -0500)]
CCSA trust-region problem should use dual_opt parameters (= user local_optim params, if any)
Ignore-this:
20371a4fa65358541c8453144d95c013
darcs-hash:
20111128224247-c8de0-
d37ecaf11d95b07924adbc10ba46a32040185947.gz
stevenj [Mon, 28 Nov 2011 19:09:23 +0000 (14:09 -0500)]
fix maximization with preconditioner (passed wrong data)
Ignore-this:
dfb13aacefe82d18fbbfc097d0bd60f6
darcs-hash:
20111128190923-c8de0-
c064a70c22f04b91ac5453482208db240f355d1b.gz
stevenj [Sun, 27 Nov 2011 19:20:58 +0000 (14:20 -0500)]
stevenj [Sat, 26 Nov 2011 19:54:13 +0000 (14:54 -0500)]
lower tolerance for dual optimization in MMA/CCSAQ; thanks to Christophe Leruste for the problem report
Ignore-this:
b8b4e324f9626f48a8e71d15dcb04c0e
darcs-hash:
20111126195413-c8de0-
7d5070a069d030f85c5dc42fe643af015290ae9e.gz
stevenj [Sat, 26 Nov 2011 19:45:06 +0000 (14:45 -0500)]
use reciprocal quadratic formula, which is accurate everywhere, rather than switching between quadratic formula and Taylor expansion
Ignore-this:
61a4e0cf96604cc766bad8540402b95f
darcs-hash:
20111126194506-c8de0-
3099879e86db6a94cba3ebd07aad2ccbfc968ff3.gz
stevenj [Wed, 23 Nov 2011 19:33:10 +0000 (14:33 -0500)]
stevenj [Wed, 23 Nov 2011 19:09:39 +0000 (14:09 -0500)]
stevenj [Wed, 16 Nov 2011 19:20:13 +0000 (14:20 -0500)]
stevenj [Tue, 15 Nov 2011 23:07:34 +0000 (18:07 -0500)]
bug in timer, thanks to William Vaughn for the patch
Ignore-this:
80c91b13714b274d94f7a14bc2792a0
darcs-hash:
20111115230734-c8de0-
b84fa13e892b3f30f272ff2fd552aca2e5207151.gz
stevenj [Tue, 15 Nov 2011 22:49:45 +0000 (17:49 -0500)]
stevenj [Tue, 15 Nov 2011 22:47:59 +0000 (17:47 -0500)]
added prototype matlab precond interface (for objective only)
Ignore-this:
483c95a6036a94ae9f5a61201730de80
darcs-hash:
20111115224759-c8de0-
6d0296c52baf1a8c579303faf3f4e1e55fb832f0.gz
stevenj [Tue, 15 Nov 2011 22:19:29 +0000 (17:19 -0500)]
added (untested) interface for specifying preconditioners
Ignore-this:
246b0790dff127b54fbf72d34d2da402
darcs-hash:
20111115221929-c8de0-
a27654751310b4251bdf6961ca7f1f3fc13164f4.gz
stevenj [Tue, 15 Nov 2011 21:14:09 +0000 (16:14 -0500)]
added CCSAQ algorithm; internal support for preconditioners (untested) not yet exported in NLopt API
Ignore-this:
a4b95966acb185468bd87b0805d156f4
darcs-hash:
20111115211409-c8de0-
a4e57acee53c6bd901a6cb0aa422982cd15c0190.gz
stevenj [Tue, 15 Nov 2011 20:41:17 +0000 (15:41 -0500)]
when adding mconstraints, allow tol==NULL as synonym for zero tolerances
Ignore-this:
db03978604ff683b95100eb50764dab9
darcs-hash:
20111115204117-c8de0-
f16c8f5ab44f1aa3ade48b350675d4a094c25e03.gz
stevenj [Thu, 9 Jun 2011 20:11:13 +0000 (16:11 -0400)]
stevenj [Thu, 9 Jun 2011 17:26:14 +0000 (13:26 -0400)]
--with-cxx should not use -no-undefined, since it deppends on the C++ std libs (which prevents us from building --with-cxx libs for Windows); thanks to Volker Lorrmann for the bug report
darcs-hash:
20110609172614-c8de0-
c4ece0c600b6cbe6155387daa7b92e4466bfbabc.gz
stevenj [Thu, 9 Jun 2011 16:34:27 +0000 (12:34 -0400)]
guile and python libs will NOT have no undefined symbols, since they refer to things in the guile and python shared libraries, respectively
darcs-hash:
20110609163427-c8de0-
2ecf408d789a768334a857b0e9909af0ca038e9f.gz
stevenj [Wed, 8 Jun 2011 17:23:00 +0000 (13:23 -0400)]
whoops, missed one static variable in luksan; thanks to Gert Wollny for the bug report
darcs-hash:
20110608172300-c8de0-
d09a1cd26ebc53d58b795086b21663c580eaf8a7.gz
stevenj [Wed, 8 Jun 2011 17:21:27 +0000 (13:21 -0400)]
stevenj [Thu, 26 May 2011 18:20:41 +0000 (14:20 -0400)]
F77 api for upper-case linkage should use set/get vector_storage, not subspace_dim
darcs-hash:
20110526182041-c8de0-
97bb6e5ae2dcd0d4390c2a6461073e148e654bde.gz
stevenj [Thu, 26 May 2011 17:50:54 +0000 (13:50 -0400)]
stevenj [Thu, 26 May 2011 17:48:15 +0000 (13:48 -0400)]
version bump to 2.2.2, updated NEWS
darcs-hash:
20110526174815-c8de0-
1d6b6104346ee393bf2f96c9215f0408a49ca4f0.gz
stevenj [Thu, 26 May 2011 17:42:13 +0000 (13:42 -0400)]
report correct count in test program
darcs-hash:
20110526174213-c8de0-
47ec8a3c4bdcd70ffbb0c68a58ef32a4ea8153a3.gz
stevenj [Thu, 26 May 2011 17:37:17 +0000 (13:37 -0400)]
stevenj [Wed, 25 May 2011 19:19:57 +0000 (15:19 -0400)]
make luksan routines re-entrant; thanks to Gert Wollny for the bug report (fixes gentoo bug #368685)
darcs-hash:
20110525191957-c8de0-
474429e4fdda69649248a32806f57f027760b645.gz
stevenj [Tue, 29 Mar 2011 22:59:57 +0000 (18:59 -0400)]
bug fix in cdirect -- typo caused slope to be underestimated in some cases, and will miss some potentially optimal rectangles. Thanks to Sinisa Hristov for the bug report
darcs-hash:
20110329225957-c8de0-
96a76c698ce0a934f7f0c2f5334aa259f073b0ef.gz
stevenj [Tue, 22 Mar 2011 20:13:21 +0000 (16:13 -0400)]
stevenj [Tue, 22 Mar 2011 20:13:00 +0000 (16:13 -0400)]
fixes to windows setup.py; thanks to Robert G for the patch
darcs-hash:
20110322201300-c8de0-
7caeb9ab8e893d3c8ec0be13c96959be1234c8d0.gz
stevenj [Mon, 7 Mar 2011 19:37:28 +0000 (14:37 -0500)]
use elimdim for praxis, nelder mead, and sbplx
darcs-hash:
20110307193728-c8de0-
19aaeaee247e0487cd6a0994c53614506fe7e3af.gz
stevenj [Mon, 7 Mar 2011 19:23:50 +0000 (14:23 -0500)]
check for ub - lb small case for potential minimizers
darcs-hash:
20110307192350-c8de0-
62dfd6579e4c6a0a3e0c9e93f7936203d9e173bf.gz
stevenj [Thu, 3 Mar 2011 00:03:54 +0000 (19:03 -0500)]
elimdim wrappers to handle lb==ub in derivative-free routines (for which it is inconvenient to handle this case directly in the algorithm)
darcs-hash:
20110303000354-c8de0-
c25c18691c4d3ae542c8989375d5e6c113269de4.gz
stevenj [Wed, 2 Mar 2011 21:08:57 +0000 (16:08 -0500)]
stevenj [Sun, 13 Feb 2011 02:53:21 +0000 (21:53 -0500)]
stevenj [Sun, 13 Feb 2011 01:58:59 +0000 (20:58 -0500)]
-b option to fix some params by equating the bounds
darcs-hash:
20110213015859-c8de0-
c548cf8cac5f028a05601f2da9c2f34f8c0acc86.gz
stevenj [Thu, 13 Jan 2011 15:22:48 +0000 (10:22 -0500)]
add missing xtol check to SLSQP (which caused an erroneeous roundoff_limited return code); thanks to Alexander Riess for the bug report
darcs-hash:
20110113152248-c8de0-
70b46f86af4489280ee6df2dc0c3a054e2cd7ecf.gz
stevenj [Thu, 13 Jan 2011 15:14:30 +0000 (10:14 -0500)]
in slsqp make sure that we count only function evals and not other callback steps
darcs-hash:
20110113151430-c8de0-
b375418b8ca98ef138ba2e00da22789b62f998f2.gz
stevenj [Tue, 7 Dec 2010 20:18:02 +0000 (15:18 -0500)]
deprecated API should support xtol_abs==NULL (for 0 tol) for backward compatibility
darcs-hash:
20101207201802-c8de0-
d0e23236914bbfae02d15c5f0c3ee61361a9da79.gz
stevenj [Sat, 20 Nov 2010 18:44:03 +0000 (13:44 -0500)]
default to mf=10 in luksan codes, like in Fortran, not mf=4, and remove pointless -5 from maxeval upper bound, and remove FIXME since this is now user-settable
darcs-hash:
20101120184403-c8de0-
4748332bf66ad0aafc31126b97ffc5a3f105543c.gz
stevenj [Sat, 20 Nov 2010 18:37:59 +0000 (13:37 -0500)]
rename subspace_dim to vector_storage, since not necessarily a subspace (and can be > n)
darcs-hash:
20101120183759-c8de0-
48667b8ad265694082acd9cc926b02c5746941e0.gz
stevenj [Thu, 18 Nov 2010 23:42:23 +0000 (18:42 -0500)]
add get/set subspace dimension for low-storage quasi-Newton methods
darcs-hash:
20101118234223-c8de0-
d7c2ef4736197f0a7e56a568a5987b3a7b5cea7e.gz
stevenj [Thu, 28 Oct 2010 23:04:31 +0000 (19:04 -0400)]
stevenj [Thu, 28 Oct 2010 23:00:02 +0000 (19:00 -0400)]
fix support for maxtime in Luksan algorithms and ORIG_DIRECT; thanks to Jurgen Werner for the bug report
darcs-hash:
20101028230002-c8de0-
c51903a73ffc18f3ddd4e5d68edc3580d1676f3d.gz
stevenj [Sat, 16 Oct 2010 21:00:36 +0000 (17:00 -0400)]
include nlopt_optimize_usage.h in dist tarball, for Windows users; thanks to Per-Olaf Sturesson for the suggestion
darcs-hash:
20101016210036-c8de0-
e090de40c0610862c1bd607e499ee3b1f2796a5d.gz
stevenj [Mon, 6 Sep 2010 18:01:28 +0000 (14:01 -0400)]
stevenj [Mon, 6 Sep 2010 17:56:43 +0000 (13:56 -0400)]
add missing Windows exporting junk for nlopt_algorithm_name; thanks to Ofek Shilon for the bug report
darcs-hash:
20100906175643-c8de0-
dc82178a1444e3cf632cb44d405483d720e82163.gz
stevenj [Wed, 1 Sep 2010 01:18:27 +0000 (21:18 -0400)]
fix compiler warnings about prototype mismatches in options.c; thanks to Benoit Scherrer
darcs-hash:
20100901011827-c8de0-
0671d98605879992d794934fc521518a73c7e98b.gz
stevenj [Mon, 30 Aug 2010 23:55:48 +0000 (19:55 -0400)]
add missing "inline" keywords to C++ functions so that nlopt.hpp can be included in multiple compilation units; thanks to Steve Jaffe for the bug report
darcs-hash:
20100830235548-c8de0-
6d4a9c8ca4900c947b2d0f4361ab6da4756188fb.gz
stevenj [Thu, 19 Aug 2010 07:36:43 +0000 (03:36 -0400)]
fixed bug in set/get xtol_abs, thanks to David Rivest-Hénault for the bug report
darcs-hash:
20100819073643-c8de0-
80e9c5691ef76d73f33255b0b615dced6374467a.gz
stevenj [Fri, 13 Aug 2010 17:39:59 +0000 (13:39 -0400)]
stevenj [Fri, 13 Aug 2010 17:35:48 +0000 (13:35 -0400)]
max/min macros are already defined(!) by stdlib.h(!) in visual studio; thanks to Benoit Scherrer for the report about the compiler warning
darcs-hash:
20100813173548-c8de0-
f0815042dc233cd3279c81517ffdaf9ef247d7e5.gz
stevenj [Thu, 5 Aug 2010 19:49:18 +0000 (15:49 -0400)]
stevenj [Fri, 30 Jul 2010 14:35:20 +0000 (10:35 -0400)]
NLOPT_DLL_EXPORT option to compile with MS dllexport flag, thanks to Benoit for the suggestion
darcs-hash:
20100730143520-c8de0-
4b368e6403b28260a9221769b452f898b6945f5c.gz
stevenj [Fri, 30 Jul 2010 14:34:25 +0000 (10:34 -0400)]
update: Nocedal code is available under GPL
darcs-hash:
20100730143425-c8de0-
b09d1a93d78c8c37456c7ce0b11e3599e776c28d.gz
stevenj [Fri, 30 Jul 2010 02:35:46 +0000 (22:35 -0400)]
handle missing C99/IEEE copysign function, e.g. for WWin32; thanks to Benoit Scherrer for the bug report
darcs-hash:
20100730023546-c8de0-
2d82e58615ba8036734a2b104aefb4a53a88f9b5.gz
stevenj [Fri, 30 Jul 2010 00:54:19 +0000 (20:54 -0400)]
stick to C89 and avoid mixed declarations and code; thanks to Benoit Scherrer for the bug report (fails with MSVC)
darcs-hash:
20100730005419-c8de0-
39dc04e5a2be81512bb3ca87c6fa397192e5bfd7.gz
stevenj [Thu, 29 Jul 2010 01:34:11 +0000 (21:34 -0400)]
lcc doesn't support dllimport, thanks to Laurent Vanbeylen for the bug report
darcs-hash:
20100729013411-c8de0-
d6b68ec8cf89f3ce2a42fdb28f2d934e9448d5d3.gz
stevenj [Thu, 15 Jul 2010 21:51:05 +0000 (17:51 -0400)]
stevenj [Thu, 15 Jul 2010 21:32:58 +0000 (17:32 -0400)]
thread-safety in nlopt_seconds(), assuming we have __thread or equivalent
darcs-hash:
20100715213258-c8de0-
d2cdf53cb97a6abc8c587f00d9027851b8ef6d29.gz
stevenj [Thu, 15 Jul 2010 21:31:24 +0000 (17:31 -0400)]
stevenj [Thu, 15 Jul 2010 21:10:33 +0000 (17:10 -0400)]
stevenj [Thu, 15 Jul 2010 20:44:05 +0000 (16:44 -0400)]
added --without-threadlocal to disable thread-local keyword for MinGW compilation
darcs-hash:
20100715204405-c8de0-
f21292dda1bd153f87aa9d772ddca945a2231a0e.gz
stevenj [Thu, 15 Jul 2010 20:23:15 +0000 (16:23 -0400)]
fix SLSQP to work with unbounded dimensions
darcs-hash:
20100715202315-c8de0-
e93f00932885115f2a0f036bc27f5d8e7334bcfc.gz
stevenj [Thu, 15 Jul 2010 19:37:14 +0000 (15:37 -0400)]
stevenj [Thu, 15 Jul 2010 19:35:04 +0000 (15:35 -0400)]
replace 1e20 with HUGE_VAL in ORIG_DIRECT code (silly hard-coded upper bound)
darcs-hash:
20100715193504-c8de0-
bddb4c45a2839ae5fa41c56d5a6754411c646d1f.gz
stevenj [Thu, 15 Jul 2010 19:34:06 +0000 (15:34 -0400)]
support forced stops in ORIG_DIRECT
darcs-hash:
20100715193406-c8de0-
57853932f731ace7580a80e8fdb66cbb6f753cb6.gz
stevenj [Thu, 15 Jul 2010 18:10:44 +0000 (14:10 -0400)]
stevenj [Thu, 15 Jul 2010 16:20:51 +0000 (12:20 -0400)]
allow BOBYQA and COBYLA to use unequal initial step sizes in different dimensions (internally rescaling the coordinates as needed); thanks to Tom Fiddaman for pointing out the problem when different coordinates have very different units
darcs-hash:
20100715162051-c8de0-
5685b685d14f5042f824bbe5bff7f3aeed0fb808.gz
stevenj [Thu, 15 Jul 2010 16:15:40 +0000 (12:15 -0400)]
fix failure in C++ (and some other front-ends) when de-allocating an nlopt object with a local optimizer; thanks to Jurgen Werner for the bug report
darcs-hash:
20100715161540-c8de0-
8aa02bad85ddebaae8e38e05825c23c3d850a702.gz
stevenj [Tue, 13 Jul 2010 17:09:40 +0000 (13:09 -0400)]
stevenj [Mon, 12 Jul 2010 23:33:39 +0000 (19:33 -0400)]
added python docstring for module; thanks to Sebastian Walter for the suggestion
darcs-hash:
20100712233339-c8de0-
b8da98e494e6441fec52320cbc5408ad83d956d2.gz
stevenj [Mon, 12 Jul 2010 23:01:44 +0000 (19:01 -0400)]
stevenj [Mon, 12 Jul 2010 22:31:17 +0000 (18:31 -0400)]
fix memory leak for error condition
darcs-hash:
20100712223117-c8de0-
1d1a5920fdadab5ae0f7f25154d7b3d823bd6715.gz
stevenj [Mon, 12 Jul 2010 22:30:58 +0000 (18:30 -0400)]
print out which bound was violated for bounds violation
darcs-hash:
20100712223058-c8de0-
28424889fa63e686e5377c0625b0891bc2446b84.gz
stevenj [Mon, 12 Jul 2010 22:30:28 +0000 (18:30 -0400)]
bug fixes to slsqp, now seems to work
darcs-hash:
20100712223028-c8de0-
5102cafe2ad38f4e100fb986cb3085cb0e7f2aba.gz
stevenj [Sat, 10 Jul 2010 21:59:47 +0000 (17:59 -0400)]
added slsqp code (compiles, not yet tested)
darcs-hash:
20100710215947-c8de0-
2175b5e6e4b4518814a525d95ce2a4643eac9104.gz
stevenj [Sat, 10 Jul 2010 21:59:37 +0000 (17:59 -0400)]
prepare to handle forced-stop in ORIG_DIRECT (not done yet)
darcs-hash:
20100710215937-c8de0-
b67e75ac8c2281b62aff33d86556069b4dc33c95.gz
stevenj [Fri, 9 Jul 2010 16:56:46 +0000 (12:56 -0400)]
add missing %catches statements for recent C++ API additions, especially the add_*constraint functions; thanks to Dmitrey Kroshko for the bug report
darcs-hash:
20100709165646-c8de0-
f7dab6f3bc3dc7900a26674a3ba6f98bdb78a3ef.gz
stevenj [Fri, 9 Jul 2010 16:34:03 +0000 (12:34 -0400)]
if user specifies OCT_INSTALL_DIR, make M_INSTALL_DIR (if unspecified) default to the same directory
darcs-hash:
20100709163403-c8de0-
5a5f3e0c9f02e3f9995138880a3d74883f96896d.gz
stevenj [Fri, 9 Jul 2010 15:14:55 +0000 (11:14 -0400)]
added GUILE_INSTALL_DIR var to change Guile installation directory
darcs-hash:
20100709151455-c8de0-
7f3529d9594bc00f7e7c7fa08a5eeb6314c39aba.gz
stevenj [Thu, 8 Jul 2010 21:23:53 +0000 (17:23 -0400)]
in functions with constraints, make sure forced_stop stops immediately, before evaluating any more user callbacks
darcs-hash:
20100708212353-c8de0-
5060b668f7e91e93ca182f3c2a8ee7618b4d5032.gz
stevenj [Thu, 8 Jul 2010 21:23:40 +0000 (17:23 -0400)]
stevenj [Thu, 8 Jul 2010 20:26:02 +0000 (16:26 -0400)]
use 2-dimensional array for grad arg of python mconstraint
darcs-hash:
20100708202602-c8de0-
815718d964a5c6e92ca5e254ba519a65c0f921e2.gz
stevenj [Thu, 8 Jul 2010 19:36:54 +0000 (15:36 -0400)]
stevenj [Thu, 8 Jul 2010 19:16:46 +0000 (15:16 -0400)]
elim. memory leaks in C++/Python/Guile/Fortran interfaces if add_constraint gives an error or you set the objective multiple times
darcs-hash:
20100708191646-c8de0-
4f06d4e9c27831699b6bd7add1291be14a348197.gz
stevenj [Thu, 8 Jul 2010 19:04:36 +0000 (15:04 -0400)]
fix memory leak in python wrapper on exception
darcs-hash:
20100708190436-c8de0-
0b50fd5776481816285a42d5b06679666377d0ea.gz
stevenj [Wed, 7 Jul 2010 23:10:56 +0000 (19:10 -0400)]
return error if more than n equality constraints are specified
darcs-hash:
20100707231056-c8de0-
74c179fa38352cee82f7ee5e209602be5aafe27b.gz
stevenj [Wed, 7 Jul 2010 23:10:33 +0000 (19:10 -0400)]
yikes, fix argument check for tol argument in add_constraint functions
darcs-hash:
20100707231033-c8de0-
d1d4a50c4e1ac45220449d5b0057a758b9441dff.gz
stevenj [Wed, 7 Jul 2010 15:21:34 +0000 (11:21 -0400)]
stevenj [Wed, 7 Jul 2010 15:17:05 +0000 (11:17 -0400)]
better checking for Python and Numpy include directories; thanks to Nathaniel Smith for the tip about numpy.get_include()
darcs-hash:
20100707151705-c8de0-
ce47965d05109ef499e8cbe8b395a682818c0804.gz
stevenj [Tue, 6 Jul 2010 23:51:46 +0000 (19:51 -0400)]
missing nlopt- prefix for version_* functions
darcs-hash:
20100706235146-c8de0-
37ee2ead46ceae2a966292480f9c89e34793f252.gz
stevenj [Tue, 6 Jul 2010 21:16:17 +0000 (17:16 -0400)]
stevenj [Tue, 6 Jul 2010 21:14:18 +0000 (17:14 -0400)]
fix python mconstraint, change C++ mconstraint to infer m from tol.size()
darcs-hash:
20100706211418-c8de0-
cd6dc26c140e257683b533f02d497ffa132db1e1.gz
stevenj [Tue, 6 Jul 2010 20:58:19 +0000 (16:58 -0400)]
python interface for mconstraint feature
darcs-hash:
20100706205819-c8de0-
4544309b57ae3e6abec33e0b145ca178ffbeee2c.gz