Thomas Capricelli
977ed615a6
be sure that there's no name clash between NumericalDiff::df and the
...
original functor df()
2009-09-28 17:45:45 +02:00
Thomas Capricelli
7968737247
fix tests : we perform slightly worse because we do use one more function
...
evaluation in our numericaldiff than what (c)minpack did
2009-09-28 04:13:57 +02:00
Thomas Capricelli
d3850641a1
remove some duplicated code LevenbergMarquardt::minimizeNumericalDiff*() by
...
using the generic Eigen NumericalDiff recently introduced.
LevenbergMarquardt::lmdif1(), which is provided as a convenience method for
people porting code from (c)minpack, is now a static function
2009-09-28 03:26:42 +02:00
Thomas Capricelli
87be19de4a
central sheme for numerical diff
2009-09-28 02:55:30 +02:00
Thomas Capricelli
206b5e3972
starting work on a Numerical differenciation module
2009-09-28 02:43:07 +02:00
Thomas Capricelli
52026eb800
cleaning
2009-09-28 02:35:42 +02:00
Thomas Capricelli
8e8997d403
use dynamic type in functor, as NonLinear only knows about this currently
2009-09-28 01:19:29 +02:00
Thomas Capricelli
bee14ee8e6
use operator() so that to be coherent with eigen AutoDiff functor
2009-09-28 00:32:31 +02:00
Thomas Capricelli
956d65ea63
define a generic functor and makes other ones inherit it
2009-09-28 00:18:14 +02:00
Thomas Capricelli
251c0f45ac
remove references to adolc and split tests functions for clarity
2009-09-27 23:54:06 +02:00
Mark Borgerding
a39de276a9
added the test case for FFTW
2009-09-14 01:52:26 -04:00
Thomas Capricelli
72746838ad
merge with tip
2009-09-10 02:37:08 +02:00
Thomas Capricelli
c990938415
eigenization of fcn_chkder + bugfix
2009-08-29 02:46:19 +02:00
Thomas Capricelli
c1265ebbfe
fix bounds using c standard instead of fortran's
2009-08-29 02:36:13 +02:00
Thomas Capricelli
4f7daf942c
fix indentation for fcn_chkder
2009-08-29 02:30:18 +02:00
Thomas Capricelli
16d08b2b0f
check number of evaluation even in the case of *1(), now we have it..
2009-08-26 14:47:10 +02:00
Thomas Capricelli
458947af5e
move Parameters as a class member, simplify calling sequence. Convenience
...
methods from minpack ( "*1()" ) get their original name back : they are
only useful when porting, anyway. Still, i prefer to keep them.
2009-08-26 14:23:05 +02:00
Thomas Capricelli
eac9293449
split every algorithm in *Init() + while(running) { *OneStep() }
2009-08-25 23:43:33 +02:00
Thomas Capricelli
bbd44ef0ad
move more stuff into Parameters
2009-08-25 23:37:27 +02:00
Thomas Capricelli
470ea55834
put nfev/njev as internal variables as well
2009-08-25 22:13:08 +02:00
Thomas Capricelli
fa0183e7c7
make diag be an internal variable too
2009-08-25 21:59:10 +02:00
Thomas Capricelli
e465ea82e1
define and use struct Parameters
2009-08-25 21:50:01 +02:00
Thomas Capricelli
d59cc0ad82
merge files
2009-08-25 17:25:56 +02:00
Thomas Capricelli
613a464320
cleaning
2009-08-25 16:48:09 +02:00
Thomas Capricelli
6c1a9703b1
move most of results vectors/matrices inside respective classes.
2009-08-25 16:08:09 +02:00
Thomas Capricelli
201f58e528
merge both c methods lmstr/lmstr1 into one class
...
LevenbergMarquardtOptimumStorage with two methods.
2009-08-25 14:18:38 +02:00
Thomas Capricelli
3f1b81e129
merge both c methods lmdif/lmdif1 into one class
...
LevenbergMarquardtNumericalDiff with two methods.
2009-08-25 14:09:19 +02:00
Thomas Capricelli
a736378331
cleaning
2009-08-25 14:03:30 +02:00
Thomas Capricelli
d880e6f774
merge both c methods hybrj1/hybrj into one class HybridNonLinearSolver with
...
two methods. hybrd stuff renamed to HybridNonLinearSolverNumericalDiff.
2009-08-25 13:56:25 +02:00
Thomas Capricelli
a043708e87
merge both c methods hybrd/hybrd1 into one class HybridNonLinearSolver with
...
two methods.
2009-08-25 13:48:25 +02:00
Thomas Capricelli
602b13815f
merge both c methods lmder/lmder1 into one class LevenbergMarquardt with
...
two methods.
2009-08-25 13:40:45 +02:00
Thomas Capricelli
45442b8d41
some more work on Functors
2009-08-24 21:48:22 +02:00
Thomas Capricelli
15d2c3af90
playing with functors
2009-08-24 19:45:35 +02:00
Thomas Capricelli
6f567f10be
cleaning
2009-08-24 19:19:30 +02:00
Thomas Capricelli
4e62e29869
cleaning covar
2009-08-24 17:49:37 +02:00
Thomas Capricelli
17905c7399
eigenization of ei_covar()
2009-08-24 17:47:35 +02:00
Thomas Capricelli
92a5bb4539
clean debug stuff
2009-08-24 16:14:42 +02:00
Thomas Capricelli
e65a7c7c70
misc cleaning
2009-08-24 09:28:29 +02:00
Thomas Capricelli
950eb4a254
various cleaning and homogeneization
2009-08-24 08:28:31 +02:00
Thomas Capricelli
134dea76d3
beautify functors for lmdif, lmstr, hybrj, hybrd
2009-08-23 04:57:48 +02:00
Thomas Capricelli
acd757737a
beautify Functor for lmder : we now have f,df,debug methods
2009-08-23 04:39:22 +02:00
Thomas Capricelli
878f15b8a5
* use eigen object for callbacks for hybrd and lmdif
...
* use Functor instead of argument for ei_fdjac*()
2009-08-23 04:06:16 +02:00
Thomas Capricelli
8a27e774f8
use eigen objects for hybrj and lmstr
2009-08-23 03:14:42 +02:00
Thomas Capricelli
3251e12258
use eigen objects for the lmder callback
2009-08-23 03:02:03 +02:00
Thomas Capricelli
2727099906
remove redundant code, fix bounds in those loops that still come from
...
fortran
2009-08-23 02:32:08 +02:00
Thomas Capricelli
1225704753
we do not need/use the 'void *p' parameter
2009-08-23 01:59:20 +02:00
Thomas Capricelli
93fabbff5e
use blueNorm() instead of norm()
2009-08-22 07:05:10 +02:00
Thomas Capricelli
aa3a7c3303
raw import of covar() : this is the last one, and we now do not depend on
...
the cminpack library anymore.
2009-08-22 06:44:41 +02:00
Thomas Capricelli
16061a46db
Now that the main algorithms are imported into eigen, we import subroutines
...
used by those algorithms (aka "second level").
This is a row import : we copy/paste the files from cminpack and make
very few changes :
* template<Scalar> them (replace double)
* dpmpar() replaced by c++ standard equivalent
* abs/fabs/sqrt/min/max replaced by ei_* or std::*
* use eigen norms instead of enorm()
Important Notes:
* The use of stableNorm() was not enough in some cases, but using
blueNorm() instead fixed the problems (some tests gave bad results,
either in number of iterations or precision of the results)
* As a whole, the only test that changed is testNistMGH17() : it now takes
some few steps less to get the same result. So this is a small improvement.
After this commit, the only remaining dependency from the cminpack
static library is 'covar', only used from the tests.
2009-08-22 06:40:22 +02:00
Hauke Heibel
ab6eb6a1a4
Adaptions from .lazy() towards .noalias().
...
Added missing casts.
2009-08-31 17:29:37 +02:00
Thomas Capricelli
20480a5438
merging ei_lmdif() and lmdif_template()
2009-08-21 04:24:59 +02:00
Thomas Capricelli
2e3d17c3ce
be (hopefully) smarter with indices convention : we keep the c convention
...
(0->n-1) as much as possible, and only convert at borders with
fortran-expecting methods, that will eventually dissapear.
2009-08-21 04:16:37 +02:00
Thomas Capricelli
1715e2cb3b
merging ei_lmder and lmder_template into ei_lmder() which takes eigen
...
argument, but still uses f2c code inside.
2009-08-21 02:34:40 +02:00
Thomas Capricelli
9294d33a11
use references intead of pointers for njev/nfev
2009-08-21 00:23:26 +02:00
Thomas Capricelli
b09ebe01da
* porting lmdif1 to eigen
...
* qtf was missing in lmdif signature (this is an output of the method)
2009-08-20 22:59:09 +02:00
Thomas Capricelli
8d2f6ad7e1
iwa is not really an argument, but just an old fashioned 'work array' :
...
remove it from the eigen API
2009-08-20 22:46:38 +02:00
Thomas Capricelli
9a876806e1
use eigen stableNorm() instead of cminpack 'enorm'. The results are mostly
...
slightly better in tests (one test needs 15 iterations intead of 16, for
the same result). Some numerical results have improved slightly, too.
If one uses blueNorm() instead, an assert for 'overflow' is raised from
blueNorm()
2009-08-20 21:04:38 +02:00
Thomas Capricelli
6953cad81d
remove unneeded "Eigen::", we already 'use' Eigen namespace
2009-08-19 20:06:34 +02:00
Thomas Capricelli
369693aa1c
oops, forgot those ones
2009-08-19 20:02:49 +02:00
Thomas Capricelli
3093e92da5
machine_epsilon is now called epsilon in latest eigen
2009-08-19 19:53:08 +02:00
Thomas Capricelli
47ac354190
merge with the main dev branch
2009-08-19 19:46:37 +02:00
Thomas Capricelli
fb54cfb013
import main files from cminpack as *.h files:
...
* function names are changed by appending _template
* it uses basic templating : template<typename T>
* wrappers now use those versions instead of the ones from cminpack
* lot of external methods from cminpack are still used
* tests pass though they are unchanged (they use wrappers)
2009-08-19 18:38:45 +02:00
Thomas Capricelli
703198a1a6
wrapper for chkder() : this was the last wrapper missing
2009-08-19 18:32:37 +02:00
Gael Guennebaud
ef13c3e754
add normalize and normalized overloads in AlignedVector3
2009-08-16 11:51:46 +02:00
Thomas Capricelli
3f63d6f97f
fix BoxBOD in the first case : now all tests pass
2009-08-14 19:04:09 +02:00
Thomas Capricelli
3856c2d84d
oops, i missed one : real last difficult nist test : Eckerle4
2009-08-14 18:54:28 +02:00
Thomas Capricelli
93627fefcf
last 'hard' test from Nist : ratkowsky3
2009-08-14 18:31:04 +02:00
Thomas Capricelli
91f61f7679
fix bad urls
2009-08-14 17:48:04 +02:00
Thomas Capricelli
d8c671f475
yet another (difficult) Nist test : Thurber
2009-08-14 17:46:28 +02:00
Thomas Capricelli
e057d1ef47
tweak precision for Chwirut2 test
2009-08-14 17:25:39 +02:00
Thomas Capricelli
56127cfb1a
add yet another easy Nist test : Chwirut2
2009-08-14 17:21:31 +02:00
Thomas Capricelli
f7cd4c8923
cleaning : wa1 used in 'covar' needs not be the same as in lmder* and all.
...
it's just an old-fashioned way to re-use memory without allocation...
2009-08-13 16:29:17 +02:00
Jitse Niesen
f71f878bab
Add support for matrix exponential of floats and complex numbers.
2009-08-12 15:44:22 +01:00
Thomas Capricelli
7b922eb634
BoxBOD : oops.. shame on me, i did a mistake in the derivative.... now we need 16
...
iterations instead of 7693 ;-)
the first test still fails though.
2009-08-12 02:34:22 +02:00
Thomas Capricelli
fd307b8f3f
fix a bug in BoxBOD Nist test : we now get the actual value for 'start 2'
...
'start 1' still fails though :/
2009-08-12 02:27:44 +02:00
Thomas Capricelli
3c675609bf
add another Nist test of 'hard' difficutly : Bennett5
2009-08-12 02:13:04 +02:00
Thomas Capricelli
54d09a8122
add another Nist test of 'hard' difficutly : MGH09
2009-08-12 01:50:56 +02:00
Thomas Capricelli
5ac17b4680
add another Nist test of medium difficutly : MGH17
2009-08-11 20:24:02 +02:00
Thomas Capricelli
d1bc9144cb
wrapper for lmstr1 and lmstr + eigenization of calling tests
2009-08-10 17:37:27 +02:00
Thomas Capricelli
bb1204a145
wrapper for lmdif1 + eigenization of calling test
2009-08-10 17:16:43 +02:00
Thomas Capricelli
80372c18ee
wrapper for lmdif (+test call eigenization)
2009-08-10 16:54:53 +02:00
Thomas Capricelli
4a26baa718
wrapper for hybrj1
2009-08-10 16:32:45 +02:00
Thomas Capricelli
1d53ce8d48
wrapper for hybrj
2009-08-10 16:21:22 +02:00
Thomas Capricelli
120235deef
add another (actuallY) difficult NIST test : BoxBOD.
...
The first try fails, the second one passes, but with a very bad accuracy
(~4 digits only).
anyway, my aim is to check we do not change cminpack while portint, so i
keep this test.
2009-08-10 14:11:55 +02:00
Thomas Capricelli
bcfe874968
add another 'difficult'-rated NIST test, which passes
2009-08-10 13:47:18 +02:00
Thomas Capricelli
7d65bd42eb
fix testNistHahn1 : i had swapped x[] and y[].... :/
2009-08-10 13:46:43 +02:00
Thomas Capricelli
b71aa34946
a Nist test rated 'difficult', which passes.
2009-08-10 13:05:30 +02:00
Thomas Capricelli
9b1130b82a
another nist test with difficulty 'leverage', it passes.
2009-08-10 12:49:44 +02:00
Thomas Capricelli
7ecbbc9aa4
another nist test with difficulty 'leverage', this one passes
2009-08-10 12:34:51 +02:00
Thomas Capricelli
1045bc17f5
another nist test ('average' difficulty), which fails. It is disabled until
...
further notice.
2009-08-10 12:08:31 +02:00
Thomas Capricelli
c7a72958ba
add an easy test from the NIST set :
...
http://www.itl.nist.gov/div898/strd/nls/data/misra1a.shtml
2009-08-10 10:47:55 +02:00
Thomas Capricelli
ec2b9f90a3
hybrd : wrapper + eigenize test
2009-08-10 03:39:50 +02:00
Thomas Capricelli
50c192961c
eigenize lmder + some other small fixes
2009-08-09 05:07:59 +02:00
Thomas Capricelli
a6625f22d4
eigenize the test for lmder1, clean functor stuff.
...
(and check the tests still pass, of course, that's the whole point..)
2009-08-09 03:54:36 +02:00
Thomas Capricelli
ceeb023ff2
use template parameter Scalar instead of VectorType, fix a segfault.
2009-08-09 03:33:04 +02:00
Thomas Capricelli
7db4052749
eigenize the test a little more
2009-08-09 03:16:24 +02:00
Thomas Capricelli
f19eda7cf6
first test for a basic wrapper (and only wrapper!) for cminpack functions
2009-08-09 03:07:34 +02:00
Thomas Capricelli
2b9f110639
actually use eigen include file
2009-08-09 01:12:14 +02:00
Thomas Capricelli
b695113a81
Add all other file from cminpack/examples as tests.
...
Important : one test was failing because cminpack-1.0.2 does x[3]=1. on x
which is of size 3. Probably because fortran indices are shifted wrt to C
indices and someone forgot to fix this one.
This is correct in this commit and this is the only change I've done on files
from cminpack examples.
(i've also reported the bug to cminpack author)
2009-08-08 23:41:54 +02:00
Thomas Capricelli
d646d99366
Start of module "NonLinear". We start out of cminpack-1.0.2
...
(http://devernay.free.fr/hacks/cminpack.html )
The first test is adapted from the example/ directory.
Some stuff is hard coded for our initial tests.
2009-08-08 22:18:48 +02:00
Mark Borgerding
4d6b962ba4
added FindFFTW, but I don't think it's right yet
2009-06-10 22:16:32 -04:00
Mark Borgerding
1c54340174
more work on ei_fftw_impl
2009-05-31 15:44:57 -04:00
Mark Borgerding
03ed6f9bfb
refactored ei_kissfft_impl to maintain a cache of cpx fft plans
2009-05-25 23:06:49 -04:00
Mark Borgerding
210092d16c
changed name from simple_fft_traits to ei_kissfft_impl
2009-05-25 20:35:24 -04:00
Mark Borgerding
326ea77390
added FFT inverse complex-to-scalar interface (not yet optimized)
2009-05-23 22:50:07 -04:00
Mark Borgerding
3047988172
scalar forward FFT optimized for even size, converts to cpx for odd
2009-05-23 12:50:07 -04:00
Mark Borgerding
9c0fcd0f62
started real optimization, added benchmark for FFT
2009-05-23 10:09:48 -04:00
Mark Borgerding
8b4afe3deb
added non-optimized real forward fft (no inverse yet)
2009-05-22 22:37:59 -04:00
Benoit Jacob
6347b1db5b
remove sentence "Eigen itself is part of the KDE project."
...
it never made very precise sense. but now does it still make any?
2009-05-22 20:25:33 +02:00
Mark Borgerding
be1b2ad4ec
removed unused code
2009-05-19 00:26:19 -04:00
Mark Borgerding
92ca9fc032
initial pass of FFT module -- includes complex 1-d case only
2009-05-19 00:21:04 -04:00
Gael Guennebaud
159c99a288
fix adolc unit test for dynamic sizes
2009-05-12 07:38:46 +00:00
Benoit Jacob
eac79b6d2e
CREDIT Hauke Heibel, fix MSVC warnings
2009-05-09 03:41:17 +00:00
Benoit Jacob
3b79d99f71
*make coeff() return a const reference (required with the recent change with CoeffReturnType)
...
*fix a unused variable warning
2009-05-09 03:39:31 +00:00
Benoit Jacob
834eb5bfc8
new unsupported module by Jitse Niesen: matrix exponential
2009-05-05 20:46:55 +00:00
Gael Guennebaud
c25fc50d54
add an AlignedVector3 module suitable for vectorization of 3D vectors
2009-05-03 17:19:19 +00:00
Gael Guennebaud
acb32c69d4
* update BVH to explicitely use aligned_allocator
...
* fix warning in StdVector
2009-04-23 11:33:36 +00:00
Gael Guennebaud
ea41a18b80
clean asm comments
2009-04-09 21:29:31 +00:00
Jos van den Oever
5a628567b0
Compile fix.
2009-04-05 14:26:47 +00:00
Gael Guennebaud
0170eb0dbe
add an auto-diff module in unsupported. it is similar to adolc's forward
...
mode but the advantage of using Eigen's expression template to compute
the derivatives (unless you nest an AutoDiffScalar into an Eigen's
matrix).
2009-04-01 14:43:37 +00:00
Gael Guennebaud
4bb5221d22
Add BVH module in unsupported (patch from Ilya Baran)
...
(I thought I committed it a week ago but it seems the command failed)
2009-03-18 20:06:06 +00:00
Gael Guennebaud
40774c625e
add a proof of concept autodiff jacobian helper class based on adolc
...
with unit test and FindAdolc cmake module
2009-02-27 16:19:13 +00:00
Gael Guennebaud
95db32fcdc
setup the unsupported directory structure.
...
The unsupported module documentation is automatically generated in:
build/doc/unsupported/
with bidirectional cross references.
I leave a class Foo in AdolcForward module to illustrate the
cross-reference behavior. I will remove it in the next commit.
2009-02-04 15:37:00 +00:00