2020

View page source

Release Notes for 2020

mm-dd

12-22

  1. If build_type was debug, ipopt was creating a debug.out file. This file is no longer being created.

  2. Fix new warning generated by git-2.29.2 pull command.

  3. Build external packages in external sub-directory instead of build/external , so one can remove the build directory without possibly having to completely rebuild the externals.

  4. Better error messaging if an error occurs during example_install.sh .

11-30

  1. Add /opt/local to the list of prefixes that are searched for cholmod.h . This was necessary to get the install to work on a Mac system that uses ports.

  2. Improve the error messaging when example_install.sh fails.

11-23

Fix the pkg-config cppad_mixed.pc file by separating the private part of the libs and requires fields.

11-21

  1. Remove use of sudo when running example_install.sh as root.

  2. Use apt-get instead of apt to list installed debian packages (this avoids a warning when running example_install.sh).

  3. Fix searching for suitesparse library on debian systems. (It seems debian recently added the hardware type to the library path.)

  4. Use multi-process make to speed up the example_install.sh , and bin/ipopt_install.sh when more that one processor is available.

11-19

  1. The example_install.sh script was modified so that on Mac OS it uses the system version of lapack and gcc (instead of using the extra package manager to install them).

  2. The run_cmake.sh script, which is used by example_install.sh , was modified to automatically sets PKG_CONFIG_PATH .

11-18

Extend example_install.sh so it also works on mac with port.

11-05

Fix some problems with example_install.sh on mac OS using brew.

11-04

There was a problem with the make check command when cholmod.h was in the include directory (instead of a suitesparse subdirectory). This has been fixed.

11-03

  1. Change cppad_prefix to cmake_install_prefix .

  2. Remove eigen_prefix from run_cmake.sh . The example install script bin/install_eigen.sh uses the prefix cmake_install_prefix/eigen to install ipopt.

  3. The file cppad_mixed.pc was added to the installation.

  4. There were some unused variable warnings in the files cholmod_factor.cpp and ldlt_cholmod.cpp were fix when build_type was release . These have been fixed.

11-02

Remove ipopt_prefix from run_cmake.sh . The example bin/install_ipopt.sh uses the prefix cmake_install_prefix to install ipopt.

10-31

Extend example_install.sh so that it handles mac OS use brew.

10-21

Advance to cppad-20201021 (this fixes a warning on some compilers).

10-10

Fix a problem with bin/install_ipopt.sh . To be specific, it needs to run get.ASL and get.Mumps to get supplements to the ASL and Mumps git repositories.

10-06

  1. Change bin/install_ipopt.sh to install a shared version of the corresponding libraries (even if debug is enabled).

  2. Change the cppad_mixed base class destructor to be virtual so that deleting a derived class object calls it. This avoid compiler warnings under certain circumstances.

  3. Advance to eigen-3.3.7. One notable part of this conversion is that Eigen::Index is no longer int . The examples and tests were changed to remove this assumption.

08-31

Fix a bug in the computation of the first derivative of the fixed effects objective when ldlt_cholmod was yes . To be more specific, the permutation for the LDLT factor was being used where the inverse permutation was required.

08-21

The following improvements we made to the install process:

  1. Fix the check of ipopt_prefix in bin/example_install.sh .

  2. Fix bin/install_eigen.sh because eigen moved from bitbucket to gitlab.

  3. Change bin/install_ipopt.sh to directly run configure instead of using coinbrew .

07-02

Add the error_msg result to the sample_fixed and sample_random routines.

06-30

  1. API Change : The last (6th) argument to the sample_fixed function (the optimal random effects random_opt ) is no longer being used and has been removed.

  2. Improve error message when fixed effects information matrix is not positive definite during a sample_fixed call. In addition, always make this condition a fatal error.

  3. Change sample_fixed.cpp , capture_xam.cpp , and ar1_xam.cpp to use hes_fixed_obj (instead of the deprecated routine information_mat ) to calculation the fixed effects information matrix hes_fixed_obj .

06-07

  1. Advance to ipopt-3.13.2 and cppad-20200606.

  2. Change the CppAD options used for optimizing cppad functions . This is intended to be an improvement for large problems, but the ar1_xam.sh and capture_xam.sh speed tests did not show much change.

  3. The ndebug_defined output was added to the capture_xam.cpp and ar1_xam.cpp examples.

  4. The following speed test parameters were increased: the number_random effects in ar1_xam.sh , the number_random effects in capture_xam.sh , the number_locations effects in capture_xam.sh

05-30

Some more improvements were made to the adaptive and trace-adaptive derivative_test options. To be specific, the scale of the function values (derivative values) were taking into account when checking derivatives (Hessians).

05-29

Add an error message when the Hessian with respect to the random effects is not positive definite during the initialization of the second order approximation for the objective, when quasi_fixed is false.

05-27

The adaptive and trace-adaptive derivative_test options were improved. To be specific, the larger step size was used unless significant improvement is achieved by a smaller step size. This reduces the amount of numerical round off to consider in the comparison.

03-28

The release version of the hes_random_obj routine would crash when there were no random effects. This has been fixed.

03-25

The debug version of the hes_random_obj routine would fail with an assert where there were no random effects. This has been changed so it returns an empty matrix (the same as the release version).

03-23

Change the notation ‘total objective’ to fixed effects objective and define the Random Effects Objective .

03-22

  1. The specifications for the Covariance of the random effect computed by sample_random was edited.

  2. information_mat was deprecated and replaced by hes_fixed_obj .

  3. The hes_random_obj routine was added.

03-18

It is no longer necessary to have the problem Lagrange multipliers in the solution argument to sample_fixed .

03-15

Change the sample_fixed routine to ignore all the constraints except where the lower and upper limits for a fixed effect are equal, Constant Fixed Effects . This is an over estimate of the variance but is faster to calculate.

API Change

The new sample_fixed routine may result in samples that are not within the lower and upper limits for the fixed effects; fixed_lower and fixed_upper . You may want to modify these samples to be within these limits before you use them.