chiark / gitweb /
stevenj [Thu, 17 Jun 2010 15:50:09 +0000 (11:50 -0400)]
STDCALL needs to go after return type for MSVC++; thanks to Dave Katz for the bug report
darcs-hash:
20100617155009-c8de0-
a0de840fc8e724787c233ceeb367925776201937.gz
stevenj [Thu, 17 Jun 2010 04:26:35 +0000 (00:26 -0400)]
stevenj [Wed, 16 Jun 2010 23:24:29 +0000 (19:24 -0400)]
stevenj [Wed, 16 Jun 2010 23:10:06 +0000 (19:10 -0400)]
added --with-mthreads option to use -mthreads flag when building MinGW, since that is reportedly needed for __thread to give threadlocal storage
darcs-hash:
20100616231006-c8de0-
4b0e74542f7fc4a99d65d452be1bb9393d9514ae.gz
stevenj [Wed, 16 Jun 2010 23:04:11 +0000 (19:04 -0400)]
stevenj [Wed, 16 Jun 2010 22:17:07 +0000 (18:17 -0400)]
added --without-python and --without-guile options
darcs-hash:
20100616221707-c8de0-
d313274f5c6f2663debde1db97ca180f794e672f.gz
stevenj [Wed, 16 Jun 2010 21:47:52 +0000 (17:47 -0400)]
fixed get_array functions in Fortran
darcs-hash:
20100616214752-c8de0-
6806563b789b0b6e62652ee39c6c68755567d080.gz
stevenj [Wed, 16 Jun 2010 21:35:17 +0000 (17:35 -0400)]
whoops, fixed to configure with --disable-shared
darcs-hash:
20100616213517-c8de0-
d0dacca994721dfbd3e1b65e221d20fa206a1607.gz
stevenj [Wed, 16 Jun 2010 18:57:31 +0000 (14:57 -0400)]
doc update -- main MLSL no longer has default sub-algorithm
darcs-hash:
20100616185731-c8de0-
442e58dd00cc159038bbbe7deca4defb0171fd7c.gz
stevenj [Wed, 16 Jun 2010 00:45:10 +0000 (20:45 -0400)]
change lb/ub field in Matlab interface to lower_bounds/upper_bounds, for consistency
darcs-hash:
20100616004510-c8de0-
9ef715297bbfeaf62e5ee94485294500f41dd1cf.gz
stevenj [Tue, 15 Jun 2010 23:06:44 +0000 (19:06 -0400)]
stevenj [Tue, 15 Jun 2010 22:29:11 +0000 (18:29 -0400)]
stevenj [Tue, 15 Jun 2010 22:25:04 +0000 (18:25 -0400)]
add NLOPT_G_MLSL variants that require local_opt to be specified
darcs-hash:
20100615222504-c8de0-
5f416d50c5d8431fe641785cc1458a78178b13db.gz
stevenj [Mon, 14 Jun 2010 22:53:07 +0000 (18:53 -0400)]
move new AUGLAG constants to end of list in order to preserve backwards compatibility
darcs-hash:
20100614225307-c8de0-
92350fe85ba959cce1d7d25007eb6ce768a20603.gz
stevenj [Mon, 14 Jun 2010 22:51:38 +0000 (18:51 -0400)]
add AUGLAG constants to other langs
darcs-hash:
20100614225138-c8de0-
6c10fbcfd0f55f1639c42c3c1bcbe4be66d0c8af.gz
stevenj [Mon, 14 Jun 2010 22:38:23 +0000 (18:38 -0400)]
add numpy.i to dist, since it isn't shipped in a standard location I can rely on
darcs-hash:
20100614223823-c8de0-
d739ffa9cb40287ddd2dd9439c52f318bb66fb87.gz
stevenj [Mon, 14 Jun 2010 22:29:55 +0000 (18:29 -0400)]
stevenj [Mon, 14 Jun 2010 22:27:41 +0000 (18:27 -0400)]
scheme callbacks now take grad as vector argument, and must use side-effects to modify it
darcs-hash:
20100614222741-c8de0-
d8e26b0d530e3644688b4170c5760abea7985c4f.gz
stevenj [Mon, 14 Jun 2010 22:04:16 +0000 (18:04 -0400)]
detect a roundoff error condition in COBYLA
darcs-hash:
20100614220416-c8de0-
ac9ecd7d4180261175fac3db4db73bc491bc37ad.gz
stevenj [Mon, 14 Jun 2010 20:55:27 +0000 (16:55 -0400)]
copy initial_step (if any) from opt to local_opt
darcs-hash:
20100614205527-c8de0-
045ada4f1a8285d48c3b50339928fcab59e3c2d2.gz
stevenj [Mon, 14 Jun 2010 20:42:37 +0000 (16:42 -0400)]
set_initial_step(opt, NULL) resets to default
darcs-hash:
20100614204237-c8de0-
cd7a3a4bdbbd4d67c592fd103779f3a97aa51fc8.gz
stevenj [Mon, 14 Jun 2010 20:38:57 +0000 (16:38 -0400)]
stevenj [Mon, 14 Jun 2010 20:29:46 +0000 (16:29 -0400)]
turn off auglag_verbose by default
darcs-hash:
20100614202946-c8de0-
3af042b584be591df33cb5bd8eb6145c502d5f7c.gz
stevenj [Mon, 14 Jun 2010 20:26:11 +0000 (16:26 -0400)]
NLOPT_AUGLAG, no LN/LD, which requires local-opt algorithm to be specified; remove "convergence" tests for no progress towards feasibility in auglag, which seem to cause incorrect termination ....
darcs-hash:
20100614202611-c8de0-
2ee921f6892ca3d1824974367eba310ec6d21612.gz
stevenj [Mon, 14 Jun 2010 18:40:16 +0000 (14:40 -0400)]
updated NEWS with new 2.0 features
darcs-hash:
20100614184016-c8de0-
d4aa6956ca06f17ef1aca932d97bcbc968cc16dc.gz
stevenj [Mon, 14 Jun 2010 18:37:16 +0000 (14:37 -0400)]
disable Python/Guile wrappers unless --enable-shared
darcs-hash:
20100614183716-c8de0-
9cba4159abbf8bc1d3e73c76341331116b6bfbb3.gz
stevenj [Mon, 14 Jun 2010 18:21:29 +0000 (14:21 -0400)]
stevenj [Mon, 14 Jun 2010 16:32:08 +0000 (12:32 -0400)]
stevenj [Sun, 13 Jun 2010 02:07:45 +0000 (22:07 -0400)]
map NLopt's C++ exceptions to more Scheme-like equivalents
darcs-hash:
20100613020745-c8de0-
0eefd9e80ec3fab27aed5991ee9f1b364b2bae32.gz
stevenj [Sun, 13 Jun 2010 02:07:26 +0000 (22:07 -0400)]
bug fix in guile wrapper, to match latest C++ api
darcs-hash:
20100613020726-c8de0-
c70e2d5c022756ef40e0b493e46dbfb9dcd37849.gz
stevenj [Sun, 13 Jun 2010 01:39:28 +0000 (21:39 -0400)]
stevenj [Sat, 12 Jun 2010 14:15:30 +0000 (10:15 -0400)]
catch and rethrow nlopt exceptions from objective
darcs-hash:
20100612141530-c8de0-
c91903c5377395424f255f5983b49d5095bdd3e3.gz
stevenj [Sat, 12 Jun 2010 04:34:14 +0000 (00:34 -0400)]
bug fix - python strides are in bytes, not in doubles (thanks to Zhichao for the bug report)
darcs-hash:
20100612043414-c8de0-
be0b85376f235b5e48551c6c917ad687ec41eaeb.gz
stevenj [Fri, 11 Jun 2010 04:44:42 +0000 (00:44 -0400)]
make prepending nlopt_ the exception, not the rule
darcs-hash:
20100611044442-c8de0-
ab8bc3e581c5809c5d321b4db90a741997b3de31.gz
stevenj [Fri, 11 Jun 2010 04:44:33 +0000 (00:44 -0400)]
stevenj [Fri, 11 Jun 2010 04:31:35 +0000 (00:31 -0400)]
stevenj [Fri, 11 Jun 2010 04:15:52 +0000 (00:15 -0400)]
stevenj [Fri, 11 Jun 2010 04:12:12 +0000 (00:12 -0400)]
stevenj [Mon, 7 Jun 2010 03:30:53 +0000 (23:30 -0400)]
python plugin compiles (untested)
darcs-hash:
20100607033053-c8de0-
9dd5b50f08cc6ae4708bf76a1dd75dd42b84aa2b.gz
stevenj [Sat, 5 Jun 2010 00:19:16 +0000 (20:19 -0400)]
replace free_f_data flag with more general munging feature, use for proper refcounting of function pointers in SWIG guile/python wrappers
darcs-hash:
20100605001916-c8de0-
37678738283a4bc9aa454687ec1bf49ecc7adca7.gz
stevenj [Fri, 4 Jun 2010 22:51:47 +0000 (18:51 -0400)]
beginnings of Python wrapper (not tested, not compiled, memory leak still in callback reference)
darcs-hash:
20100604225147-c8de0-
110484f6dd1de418e3e3db7cfb7c16bc2988e29b.gz
stevenj [Fri, 4 Jun 2010 22:45:23 +0000 (18:45 -0400)]
stevenj [Fri, 4 Jun 2010 20:44:31 +0000 (16:44 -0400)]
stevenj [Fri, 4 Jun 2010 02:03:46 +0000 (22:03 -0400)]
support exceptions in SWIG wrappers
darcs-hash:
20100604020346-c8de0-
b94b0853daeba61580499c4385089737b6d9d221.gz
stevenj [Thu, 3 Jun 2010 23:14:01 +0000 (19:14 -0400)]
provide alternate optimize() method that does not areturn results via reference arguments, to make it easier for SWIG to translate
darcs-hash:
20100603231401-c8de0-
f74e227905ebfc20c7bb432282a5f52bdc9349ae.gz
stevenj [Thu, 3 Jun 2010 22:55:50 +0000 (18:55 -0400)]
use thread_id, if available, in nlopt_srand_time to ensure that different threads get different seed even if they are called simultaneously
darcs-hash:
20100603225550-c8de0-
963cf98deef902085d264eb09e4067f0e8e32742.gz
stevenj [Thu, 3 Jun 2010 13:26:30 +0000 (09:26 -0400)]
libnlopt_guile has no inter-library dependencies
darcs-hash:
20100603132630-c8de0-
b7f58c6b0bb073d28cbeae2f157c32c8ae4d6ee8.gz
stevenj [Thu, 3 Jun 2010 02:58:24 +0000 (22:58 -0400)]
Guile wrapper seems mostly functional, sans exceptions (needed wrapper around opt::optimize to return triplet)
darcs-hash:
20100603025824-c8de0-
bfd2efa65d9993029b48081e35b7fb28fb331a22.gz
stevenj [Thu, 3 Jun 2010 02:06:55 +0000 (22:06 -0400)]
initial stab at guile-wrapper/SWIG Makefile
darcs-hash:
20100603020655-c8de0-
45c3100538d44d29af80f43ac4c3d88cb06f148e.gz
stevenj [Wed, 2 Jun 2010 17:39:26 +0000 (13:39 -0400)]
no way to rethrow last-caught exception outside of "catch" block
darcs-hash:
20100602173926-c8de0-
c802cf401c6123da9638d637599d1e2d293938ed.gz
stevenj [Wed, 2 Jun 2010 17:23:33 +0000 (13:23 -0400)]
use memcpy, return inf on exception from objective
darcs-hash:
20100602172333-c8de0-
6b496612c4caef9edcf942e4a6397431223173af.gz
stevenj [Wed, 2 Jun 2010 17:13:17 +0000 (13:13 -0400)]
slight optimization in std::vector<double> wrappers (alllocate temporaries only once)
darcs-hash:
20100602171317-c8de0-
7e2e715b15d5222bc52bb2b35a56691008dcc254.gz
stevenj [Wed, 2 Jun 2010 06:22:44 +0000 (02:22 -0400)]
first stab at SWIG wrappers (unfinished Makefile), starting with Guile
darcs-hash:
20100602062244-c8de0-
d776ad5f27c68cc6b6d5799d341d4bb0b0264c58.gz
stevenj [Wed, 2 Jun 2010 05:25:37 +0000 (01:25 -0400)]
no need for opt::reinit since we have = operator
darcs-hash:
20100602052537-c8de0-
526082e36d2563e1506b393e5564b7b0129a89d4.gz
stevenj [Wed, 2 Jun 2010 03:50:21 +0000 (23:50 -0400)]
added fortran wrapper for nlopt_copy
darcs-hash:
20100602035021-c8de0-
d1fabb203b45275b539e487a4868933cf9551be9.gz
stevenj [Wed, 2 Jun 2010 02:44:02 +0000 (22:44 -0400)]
take f_data in std::vector-style objective/constraints
darcs-hash:
20100602024402-c8de0-
d24dad28db4a9059944be55dd5870d6ecc7aceaf.gz
stevenj [Wed, 2 Jun 2010 01:45:55 +0000 (21:45 -0400)]
streamlined C++ header by using only std::vector and little or no double*; also, try to be somewhat exception-safe
darcs-hash:
20100602014555-c8de0-
8b4cefb5e66fddd028f137a6d6b7b19ebca5445b.gz
stevenj [Wed, 2 Jun 2010 00:36:32 +0000 (20:36 -0400)]
stevenj [Tue, 1 Jun 2010 22:47:59 +0000 (18:47 -0400)]
stevenj [Tue, 1 Jun 2010 15:17:31 +0000 (11:17 -0400)]
added new Fortran API, removed unnecessary --with-windows-f77-mangling
darcs-hash:
20100601151731-c8de0-
094eca40a2ca0b04dbb029cb111b2284ede822da.gz
stevenj [Mon, 31 May 2010 21:49:14 +0000 (17:49 -0400)]
fixed generation of nlopt.f (again)
darcs-hash:
20100531214914-c8de0-
0bd388d69350800bad22d18211e5def40a0121e1.gz
stevenj [Mon, 31 May 2010 16:59:10 +0000 (12:59 -0400)]
improve C++ header exceptions, rename NLOPT_FORCE_STOP to NLOPT_FORCED_STOP
darcs-hash:
20100531165910-c8de0-
347b8ba2ff62d825467d4baf6b21508f5364d1e0.gz
stevenj [Fri, 28 May 2010 21:41:12 +0000 (17:41 -0400)]
use thread-local storage for Mersenne twister to make thread-safe (for compilers supporting __thread or __declspec(thread))
darcs-hash:
20100528214112-c8de0-
4ca1313eec0a24eb1bc4e0b4aea0bc0c93c75788.gz
stevenj [Fri, 28 May 2010 16:44:40 +0000 (12:44 -0400)]
stevenj [Fri, 28 May 2010 16:37:37 +0000 (12:37 -0400)]
stevenj [Fri, 28 May 2010 16:36:00 +0000 (12:36 -0400)]
added nlopt_force_stop termination
darcs-hash:
20100528163600-c8de0-
aa1d026191abea2ffd7870265933673fe18ab890.gz
stevenj [Wed, 12 May 2010 15:18:12 +0000 (11:18 -0400)]
stevenj [Wed, 12 May 2010 15:01:53 +0000 (11:01 -0400)]
stevenj [Wed, 12 May 2010 04:00:09 +0000 (00:00 -0400)]
stab at Matlab nlopt_optimize interface
darcs-hash:
20100512040009-c8de0-
edd99ecc6904eca7c0fbb4d42a263b0ddb8cd00e.gz
stevenj [Wed, 12 May 2010 03:48:48 +0000 (23:48 -0400)]
note that n argument is now unsigned
darcs-hash:
20100512034848-c8de0-
7a7f45a53656b351d4372ed22104bab823569cf7.gz
stevenj [Wed, 12 May 2010 02:14:54 +0000 (22:14 -0400)]
replace Octave nlopt_minimize_constrained with wrapper around new nlopt_optimize, and document the latter
darcs-hash:
20100512021454-c8de0-
d8c1513f66476aa316e32561e5dc648aca70e06f.gz
stevenj [Wed, 12 May 2010 02:14:30 +0000 (22:14 -0400)]
remove obsolete NLOPT_MINF_MAX_REACHED
darcs-hash:
20100512021430-c8de0-
95df1ca4c461239c6ddbd8cc0e87104d0101039f.gz
stevenj [Wed, 12 May 2010 02:13:51 +0000 (22:13 -0400)]
stevenj [Sat, 8 May 2010 00:12:55 +0000 (20:12 -0400)]
stevenj [Fri, 7 May 2010 23:48:05 +0000 (19:48 -0400)]
support inequality constraints directly in ORIG_DIRECT
darcs-hash:
20100507234805-c8de0-
7419b76ac7a7dde36b4c7cc380ea3be74ba583ef.gz
stevenj [Tue, 6 Apr 2010 02:50:26 +0000 (22:50 -0400)]
initial stab at new-style Octave api
darcs-hash:
20100406025026-c8de0-
653797e7b13d5e004082e273187d950c8db7e436.gz
stevenj [Tue, 6 Apr 2010 02:49:33 +0000 (22:49 -0400)]
stevenj [Mon, 5 Apr 2010 19:17:49 +0000 (15:17 -0400)]
put a few more functions into the C++ API
darcs-hash:
20100405191749-c8de0-
2bb5cb5b80befa74a8a666be82866d91817a9153.gz
stevenj [Mon, 5 Apr 2010 07:28:11 +0000 (03:28 -0400)]
catch a vector-size mismatch in C++
darcs-hash:
20100405072811-c8de0-
307c0f8c22e707816cbe2da178b2ff92fc6a596f.gz
stevenj [Mon, 5 Apr 2010 07:19:53 +0000 (03:19 -0400)]
stevenj [Mon, 5 Apr 2010 07:16:50 +0000 (03:16 -0400)]
add set_max_objective to automate the sign flip for maximization
darcs-hash:
20100405071650-c8de0-
c7d3b05da048406f5004a4c9c46dccd6d2a41100.gz
stevenj [Mon, 5 Apr 2010 06:30:16 +0000 (02:30 -0400)]
stevenj [Mon, 5 Apr 2010 05:58:31 +0000 (01:58 -0400)]
stevenj [Mon, 5 Apr 2010 05:41:36 +0000 (01:41 -0400)]
change new API to use unsigned where it makes sense
darcs-hash:
20100405054136-c8de0-
a106e736e68aa365986a1cc91b9faf8492dd6ea1.gz
stevenj [Mon, 5 Apr 2010 04:49:54 +0000 (00:49 -0400)]
stevenj [Mon, 5 Apr 2010 04:49:17 +0000 (00:49 -0400)]
in MMA, only stop on minf_max if feasible!
darcs-hash:
20100405044917-c8de0-
be7df864ba060befce69c335e9116b01a2f82f10.gz
stevenj [Mon, 5 Apr 2010 01:34:43 +0000 (21:34 -0400)]
stevenj [Mon, 5 Apr 2010 00:56:32 +0000 (20:56 -0400)]
stevenj [Mon, 5 Apr 2010 00:54:22 +0000 (20:54 -0400)]
use constraint tolerance for feasibility check of minf convergence check in COBYLA
darcs-hash:
20100405005422-c8de0-
9fee12e2bca611ba6c4662af0cf2ab49b14a2749.gz
stevenj [Mon, 5 Apr 2010 00:34:52 +0000 (20:34 -0400)]
pseudo-randomize simplex steps in COBLYA, to avoid accidentally taking steps that don't improve conditioning (which seems to happen with bound constraints where the optimum is against the bounds, sometimes); note that the algorithm remains deterministic
darcs-hash:
20100405003452-c8de0-
af9df1f48a6f2adb1d3990aa66320235ee556cab.gz
stevenj [Mon, 5 Apr 2010 00:33:44 +0000 (20:33 -0400)]
allow COBYLA to increase trust-region radius (rho) if predicted improvement was approximately right and simplex is ok, following suggestion in SAS manual for PROC NLP
darcs-hash:
20100405003344-c8de0-
d288279ade2dfafe7b9bcf9facf7807288f28010.gz
stevenj [Mon, 5 Apr 2010 00:33:28 +0000 (20:33 -0400)]
stevenj [Sun, 4 Apr 2010 22:24:32 +0000 (18:24 -0400)]
don't call MLSL recursively (e.g. in auglag)
darcs-hash:
20100404222432-c8de0-
d8e9e542f9b3e8add61ec28c5346858e37bd3577.gz
stevenj [Sun, 4 Apr 2010 22:18:35 +0000 (18:18 -0400)]
return error in global search if domain is not finite
darcs-hash:
20100404221835-c8de0-
748279ced777a37c6a5ada00892bba75da7cdd54.gz
stevenj [Sun, 4 Apr 2010 21:51:25 +0000 (17:51 -0400)]
improved (I think) stopping criteria for auglag
darcs-hash:
20100404215125-c8de0-
c139af2024d66a76717073417d5c15c7db314647.gz
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