Hi everyone! I want to use cmipack to fit my data to different models (least squares nonlinear problem) but actually I see that routine returns 1 and initial guess vector is not changed, but the fitting problem is quite simple. I think that probably I'm doing something wrong.

My fitting problem is the following. Find parameters of gaussian function (amplitude, mu, and sigma) to fit the data:

  // abscissa
  const std::vector<double> x1{
    -3.01969624096814,
    -2.20911406128415,
    -1.39853188160016,
    -0.587949701916172,
    0.222632477767818,
    1.03321465745181,
    1.8437968371358,
    2.65437901681979,
    3.46496119650378
  };
  
  // data
  const std::vector<double> y1{
    22, 183, 917, 2446, 3239, 2263, 778, 140, 12
  };

It's obvious gaussian function as we can see. In cmipack we can use several functions to fit the data. I've tried lmdif1, when the derivatives are calculated automatically using forward finite difference approximation. In this case I need just define the function to minimize:

struct FData
{
  std::vector<double> x;
  std::vector<double> dataToModel;
};
int gaussian_cminpack(void* p, int m, int n, const double* x,
                       double* fvec, int flag)
{
  auto myData = reinterpret_cast<FData*>(p);

  for (int i = 0; i < m; ++i)
  {
    double y = x[2]  * exp(-0.5 * sqr((myData->x[i] - x[0]) / x[1])) / (sqrt(2 * PI) * x[1]);
    fvec[i] = abs(y - sqr(myData->dataToModel[i]));
  }
  return 0;
}

Then I just following the documentation and

  double iGuess[]{ 0.01, 0.6, 3239 };

  FData data{ x1, y1 };
  double fvec[9];
  int iwa[3];
  double wa[51];
  int lwa = 51;

  auto r = lmdif1(gaussian_cminpack, &data, 9, 3, iGuess, fvec, 0.01, iwa, wa, lwa);

I get r = 1 and iGuess is not changed. Why?

Also, I've tried to use lmder1, when I should give the jacobian 9x3 matrix and also I don't get some adequate result. Can you help me please?

The error informations are:
error LNK2001: unresolved external symbol __imp_hybrd1
error LNK2001: unresolved external symbol __imp_dpmpar

The library path and dependencies are both set in the project, how to fix it?

Hi @devernay , I found that in the version 1.3.8, cmake fails to find BLAS, and says cannot find MKL tool.

while in the master branch, find_package(CBLAS) has been replaced by find_package (BLAS). I tried the master branch version and it succeeded to find BLAS (both MKL and OpenBLAS). Since the previous 1.3.8 version was released in 2021, it might be suitable to release a new version containing some modern "features", at least for the BLAS issue. Anyway, thank you for the great library!

https://homebrew.bintray.com/bottles/cminpack-1.3.4.yosemite.bottle.tar.gz

http://devernay.free.fr/hacks/cminpack/cminpack-1.3.4.tar.gz

The above links are broken, so it can't use brew to install in MacOs
Could you fix it ?

Had no success with the instructed definition

-DSHARED_LIBS=ON

but

-DBUILD_SHARED_LIBS=ON

cd /home
wget https://github.com/devernay/cminpack/archive/v1.3.9.tar.gz
tar -xf v1.3.9.tar.gz
cd v1.3.9.tar.gz
make check

make[2]: *** [Makefile:177: testhydrv] Error 1
*** Running CHK tests (calling MINPACK from C using ../libcminpacks.a)
for x in schkdrv_; do echo $x; ./$x < testdata/chkder.data > $x.out; diff -u ref/echo $x |sed -e s/_/c/.ref $x.out ; done
schkdrv_
--- ref/schkdrvc.ref 2024-05-29 05:09:16.000000000 +0800
+++ schkdrv_.out 2024-06-03 21:38:27.276142410 +0800

make[2]: *** [Makefile:181: testchkdrv] Error 1
make[2]: Target 'testdrv' not remade because of errors.
make[2]: Leaving directory '/root/cminpack-1.3.9/examples'
make[1]: [Makefile:269: checkfloat] Error 2 (ignored)
make clean LIBSUFFIX=s 2>&1 > /dev/null
make[1]: Leaving directory '/root/cminpack-1.3.9/examples'

[root@localhost cminpack-1.3.9]#cat /etc/os-release
NAME="CentOS Linux"
VERSION="8 (Core)"
ID="centos"
ID_LIKE="rhel fedora"
VERSION_ID="8"
PLATFORM_ID="platform:el8"
PRETTY_NAME="CentOS Linux 8 (Core)"
ANSI_COLOR="0;31"
CPE_NAME="cpe:/o:centos:centos:8"
HOME_URL="https://www.centos.org/"
BUG_REPORT_URL="https://bugs.centos.org/"

Currently, double-precision and single-precision builds default to sharing the same library name. As a consequence, downstream packagers (linux, conda) typically just package the double-precision version, and the single precision version is unavailable. In fact, it is not that difficult to provide both single and double precisions in e.g. a linux package; the single-precision library can simply use a separate name, and this is readily achieved with e.g. cmake ... -DBUILD_SHARED_LIBS=ON -DCMAKE_C_FLAGS=-D__cminpack_float__ -DCMAKE_RELEASE_POSTFIX=s which builds the single precision shared lib as libcminpacks.so (the "s" suffix being the one suggested in https://www.math.utah.edu/software/minpack.html#linking-local too).
Would you be willing to declare "officially" that it is preferred that downstream packagers provide both single and double precision versions of the library, with the single-precision using the "s" suffix (for example)?

Hello,
I think the cmake is not linking to cblas as it should:

CMakeFiles/cminpack.dir/objects.a(enorm.c.obj): In function `enorm':
/home/xantares/projects/aur-scripts/mingw-w64-cminpack/src/cminpack-1.3.6/enorm.c:53: undefined reference to `cblas_dnrm2'

Adding target_link_libraries(cminpack PUBLIC cblas) manually solves the problem, but this needs cleaner detection, for example if needing openblas. Maybe be linking to blas/cblas is private.

Especially when the number of problems m is greater than that of variables n. I've tried some simple nonlinear equations and it runs very well, but once the problem becomes complex, results are not correct always. I adjusted the initial value of variables x and some parameters like factor, but it turned out useless.
So I wonder if there is any improvement of the function 'lmdif' or 'lmder' to deal with more complex problems? Or what I can modify to make it work better?

I've simply cloned the repo (master & v1.3.6) and used CMake to build and run the tests. But one test fails on consistently on all platforms (Linux, Mac, Windows). I could not find any open issues about a failing test therefore I thought I'd ask whether this is known and/or to be expected.

Steps to reproduce:

Clone repository and change to project's root folder. In this folder run:

cmake -H. -Bbuild
cmake --build build -- -j
cmake -E env CTEST_OUTPUT_ON_FAILURE=1 cmake --build build --target test

Result:

96% tests passed, 1 tests failed out of 24

Total Test time (real) =   0.54 sec

The following tests FAILED:
	 16 - tlmdifc (Failed)
Errors while running CTest
Makefile:108: recipe for target 'test' failed
make: *** [test] Error 8

The test output written to build/examples/tlmdifc.out is:

 final l2 norm of the residuals     0.09063596

 number of function evaluations        21

 exit parameter                         1

 final approximate solution

     0.08241058       1.133037       2.343695
 covariance

   0.0001531202    0.002869942   -0.002656662
    0.002869942     0.09480937    -0.09098997
   -0.002656662    -0.09098997     0.08778729

These results are from a run on Ubuntu Bionic amd64 but the very same test also failed for me on Mac and Windows.

Kudos for any/all for making this package available.

I have recently modified lmdif/lmder to use the Adept automatic differentiation library, please see sherpa/sherpa#577 for details.

The CMINPACK_VERSION variable contains a semicolon separated list of strings:

i.e. 1.3.9;STRING;CMinpack\ version. See https://cmake.org/cmake/help/latest/command/set.html#set :

Multiple arguments will be joined as a semicolon-separated list to form the actual variable value to be set.

After

the pkgconfig file contains the following Version field:

Version: 1.3.8;STRING;CMinpack version

Hi Frédéric

Issue 1: Would you mind adding a debug postfix to the debug builds?

if (BUILD_SHARED_LIBS)
  message (STATUS "Building shared libraries.")
  set(CMAKE_DEBUG_POSTFIX -gd)
else ()
  message (STATUS "Building static libraries.")
  set(CMAKE_RELEASE_POSTFIX _s)
  set(CMAKE_RELWITHDEBINFO_POSTFIX _s)
  set(CMAKE_DEBUG_POSTFIX _s-gd)
  set(CMAKE_MINSIZEREL_POSTFIX _s)
  if(WIN32)
	add_definitions(-DCMINPACK_NO_DLL)
  endif(WIN32)
endif ()

Issue 2: There is a warning in CMake because of this condition.

if (OS_LINUX OR ${CMAKE_SYSTEM_NAME} STREQUAL "FreeBSD")
	option (USE_FPIC "Use the -fPIC compiler flag." ON)
else (OS_LINUX)
	option (USE_FPIC "Use the -fPIC compiler flag." OFF)
endif (OS_LINUX)

Warning:

CMake Warning (dev) in CMakeLists.txt:
  A logical block opening on the line

    C:/src/cminpack-1.3.6/CMakeLists.txt:23 (if)

  closes on the line

    C:/src/cminpack-1.3.6/CMakeLists.txt:27 (endif)

  with mis-matching arguments.
This warning is for project developers.  Use -Wno-dev to suppress it.

I would change it to:

if (OS_LINUX OR ${CMAKE_SYSTEM_NAME} STREQUAL "FreeBSD")
	option (USE_FPIC "Use the -fPIC compiler flag." ON)
else ()
	option (USE_FPIC "Use the -fPIC compiler flag." OFF)
endif ()

Cheers
Andreas

Hi @devernay! First, thanks for maintaining this project, it's awesome to have an actually maintained "alternative" to the very old Fortran MINPACK!
I'd like to know if there are plans for a new 1.3.7 release (or if it could be released soon), because I'd like to create packages for cminpack on conda-forge. Thanks!

In cminpack, two types of function pointer are used. For example, as for hybrd, we could pass a function pointer to it, which is like
void(*fun)(const int *n, const real *x, real *fvec, int *iflag) or int (*fun)(void *p, int n, const real *x, real *fvec, int iflag). The void *p in the latter type is used to pass additional parameters without using global variables.
But if we want to use lambda, functor, or something else, we cannot make it. So, why not use some templates, so that we are not only able to pass a static type function pointer to hybrd, but also lambda, functor or other stuff.

I'm compiling a fortran package in linux and windows.
It perfectly link with minpack in linux, but the next functions are missing in windows:
qrfac_' qform_' dogleg_' r1updt_' r1mpyq_' r1mpyq_'
I'm using the package stored in conda-forge

The parameter "nfev" is the number of fcn calls. But how can I know the iteration numbers using hybrd function?

Currently the use of pkg-config in CMinpackConfig.cmake in WIN32 is making it append ;cminpack;m to CMINPACK_LIBRARIES, and this causes linkage problems, since these libraries cannot be found.

Output of the relevant part from CMake:

-- Found PkgConfig: C:/ProgramData/chocolatey/bin/pkg-config.exe (found version "0.28")
-- Checking for module 'cminpack'
--   Found cminpack, version 1.3.6
-- Found CMinpack: C:/prefix/lib/cminpack.lib
-- CMinPack found (include: C:/prefix/include/cminpack-1, libs: optimized;C:/prefix/lib/cminpack.lib;debug;C:/prefix/lib/cminpack.lib;cminpack;m)

I'm trying to port some python code that uses hybrd to C++, and finding it very hard to follow what's going on in this package:

There seems to be two versions of every file - hybrd_.c and hybrd.c, for example. I think that hybrd_.c is the one that's been significantly updated?

http://devernay.free.fr/hacks/cminpack/ says this is a significant rework, but the github readme says:

This is a C version of the minpack minimization package. It has been derived from the fortran code using f2c and some limited manual editing.

As far as I can tell, http://devernay.free.fr/hacks/cminpack/ links to this github project, so is the github readme out of date?

Additionally, it seems like the comments in the various files have never been updated. I'm trying to figure out how to use the void* facility:

The function to be minimized and all the Minpack functions now take an extra "void*" argument, which can be used to pass any pointer-to-struct or pointer-to-class, and you can put all you extra parameters and data in that struct. Just cast this pointer to the appropriate pointer type in your function, and there they are! There is no need for global variables anymore. Be careful if you access the same object from different threads, though (a solution is to protect this extra data with a mutex).

But I'm getting lost somewhere in the #define macro maze, and the comments in https://github.com/devernay/cminpack/blob/master/hybrd1.c seem to be the ones from the original fortran. I also can't seem to find any documentation of the changes, or the new function signatures after all the precompiler munging.

It seems like it'd be optimal to take the original versions, which are kept for historical reasons (?), and at least put them in a subdirectory so they don't clutter up everything anymore.

There's a lot of precompiler magic. For example, the first parameter in hybrd is __cminpack_decl_fcn_nn__, but I have no idea where it's coming from. cminpack.h then #defines it to just cminpack_func_nn fcn_nn, neither component of which is defined anywhere, and I have no idea how this file ever compiles. I can only assume something else string-munges to cminpack_func_nn or fcn_nn (holy crap that's confusing).

For that matter, just naively reading the headers is confusing:

__cminpack_attr__
int CMINPACK_EXPORT __cminpack_func__(hybrd1)( __cminpack_decl_fcn_nn__
        void *p, int n, __cminpack_real__ *x, __cminpack_real__ *fvec, __cminpack_real__ tol,
        __cminpack_real__ *wa, int lwa );

Without preprocessing that, does it take the first parameter of __cminpack_decl_fcn_nn__ void *p, where __cminpack_decl_fcn_nn__ is some sort of accessibility specifier, or (as I infer from the documentation) is __cminpack_decl_fcn_nn__ somehow a separate argument despite the lack of a comma? How does that work?

I don't know if I have a coherent question here aside from ".....wat", but in general, I think the documentation needs a little cleanup. I've used doxygen for generating "what it looks like in use" docs before, with great success.

I'm gonna try to convince the precompiler to tell me what the functions I'm calling actually look like.

Hi,

I'm considering using your package for a problem I have.

1. Can your tool be used in visual studio 2015, windows 10?
2. I can see some CUDA source files, are the minimization algorithms implemented in CUDA as well? or they're just generic examples?
3. Can your tool fit a functions f : R^m -> R^n? or is it restricted only to scalar functions?
4. If 3 is yes, is there a bound on the number of parameters to be fit? I have to fit functions that have 192^2 parameters.

Thank you

Is this intended?

36-202% port -v installed cminpack
The following ports are currently installed:
  cminpack @1.3.8_0 requested_variants='' platform='darwin 10' archs='ppc' date='2023-03-08T16:08:00+0800'
  cminpack @1.3.9_0 requested_variants='' platform='darwin 10' archs='ppc' date='2024-06-01T18:47:24+0800'
  cminpack @1.3.10_0 (active) requested_variants='' platform='darwin 10' archs='ppc' date='2024-09-13T09:51:02+0800'

36-202% port contents cminpack
Port cminpack @1.3.10_0 contains:
  /opt/local/include/cminpack-1/cminpack.h
  /opt/local/include/cminpack-1/minpack.h
  /opt/local/lib/libcminpack.1.3.9.dylib
  /opt/local/lib/libcminpack.1.dylib
  /opt/local/lib/libcminpack.dylib
  /opt/local/lib/libcminpackld.1.3.9.dylib
  /opt/local/lib/libcminpackld.1.dylib
  /opt/local/lib/libcminpackld.dylib
  /opt/local/lib/libcminpacks.1.3.9.dylib
  /opt/local/lib/libcminpacks.1.dylib
  /opt/local/lib/libcminpacks.dylib
  /opt/local/lib/pkgconfig/cminpack.pc
  /opt/local/lib/pkgconfig/cminpackld.pc
  /opt/local/lib/pkgconfig/cminpacks.pc
  /opt/local/share/cminpack/CMinpackConfig-macports.cmake
  /opt/local/share/cminpack/CMinpackConfig.cmake
  /opt/local/share/cminpack/CopyrightMINPACK.txt

There is nothing in MacPorts to set the version manually here.