2018

View page source

Release Notes for 2018

mm-dd

10-08

  1. Remove partial Newton step wish list item (implemented on 2018-06-30 ).

  2. Add multi-threading wish list item.

09-25

Advance to cppad-20180925 . This enabled us to remove the need for two and three levels of AD; i.e. a2_double and a3_double . The arguments and results for the ran_likelihood function are now a1_vector objects instead of a3_vector objects.

08-30

API: Remove the user defined versions of the Jacobian and Hessian of the random likelihood with respect to the random effects; ran_likelihood_jac and ran_likelihood_hes . These functions are no longer part of the API and are not used (even if defined by the user).

08-27

Advance to cppad-20180827

08-22

API: Change the type CppAD::mixed::sparse_rcv to d_sparse_rcv to better be more consistent with the other CppAD mixed type definitions . This will require some minor changes to the user’s code. For example, in the constructor for the derived class, the type for A_rcv has changed from sparse_rcv to d_sparse_rcv .

08-20

Advance to cppad-20180820 (this version has optimization of dynamic parameters).

08-18

  1. Advance to cppad-20810818 .

  2. Use Cppad dynamic parameters (new feature) to speed up the computation of the Hessian of the objective (when quasi_fixed is true). Cppad needs to improve its optimization of dynamic parameter calculations. For example, the optimized tapes that compute the Hessian of H(beta, theta, u) are larger when ( theta , u ) are dynamic parameters than when they are variables. These tapes are about the same size before being optimized.

08-09

Advance to cppad-20810809 . This fixed valgrind warnings and a bug in cppad.

08-08

Advance to cppad-20810730 . This resulted is and speed improvement for the current version of the ar1_xam.sh and capture_xam.sh tests.

07-25

  1. Add discussion of why the default eigen_prefix in bin/run_cmake.sh is different from the other prefixes.

  2. Change the name cppad_cxx_flags to extra_cxx_flags .

  3. Improve discussion of the Paths needed during and after the install process.

  4. Improve detection (during the install ) of when the default settings for extra_cxx_flags do not work.

07-12

  1. The fixed effects optimization will not try to backup and recover when max_iter is minus one for the fixed effects. In addition, when running the debug (release) version, errors during initialize will result in an runtime error (call to fatal_error ). This will gives the user the choice to seem the source code where the error was detected when build_type is debug .

  2. Improve documentation for quasi_fixed .

06-30

  1. Implement the partial Newton step update in definition of the second order accurate approximation for the optimal random effects W(beta, theta, u) . This should lead to a large improvement when the Hessian of the random likelihood with respect to the random effects is sparse.

  2. The checkpoint_newton_step setting in run_cmake.sh has been removed.

06-29

Advance to cppad-20180627 .

06-20

Advance to cppad-20180620 . This should use increase speed and use less memory during the optimization of cppad functions; see optimize_cppad_function .

06-14

  1. Some systems installs of suitesparse install cholmod.h in the systems include path instead of a suitesparse subdirector of the system include path. The install of cppad_mixed now automatically detects both these cases.

  2. Advance to cppad-20180613 . This should use less memory during the optimization of cppad functions; see optimize_cppad_function .

06-09

  1. Change the suitesparse library from a special requirement (using bin/install_suitesparse.sh ) to a system requirement .

  2. The replace option for the existing argument to bin/example_install.sh was not working. This has been fixed.

06-04

Advance to using cppad-20180604 . This will probably have no effect but might reduce the amount of memory cppad_mixed uses.

05-21

Advance to using cppad-20180521 . This should reduce the amount of memory cppad_mixed uses.

05-07

Advance to using cppad-20180507 . This should both increase the speed of cppad_mixed and reduce the amount of memory it uses.

05-03

Advance to using cppad-20180503 . This should reduce the amount of memory used by cppad_mixed , without adversely affecting its speed.

04-06

The limited_memory_max_history setting used to be forbidden as an fixed_ipopt_option because it was chosen by cppad_mixed . This has been changed and now the user can choose this ipopt option.

03-22

In the wish list, remove the Student’s t item and change the sparse_matrix item.

03-10

The install Download procedure was changed to use the following command to determine the version number:

grep '^SET(cppad_mixed_version' CMakeLists.txt

02-20

Some function declarations for friends in cppad_mixed.hpp were missing. This has been fixed.

02-12

Add more discussion of fixed_scale .

02-11

Avoid warning: ‘obj_factor_scaled’ may be used uninitialized in eval_h on some systems.

02-10

  1. The hide_ipopt_scaling option was removed from the cmake command and the scaling is always hidden from ipopt.

  2. Improve discussion of the fixed_scale argument to the optimize_fixed function.

02-08

There was a bug in computing the second derivative of the objective when the fix_likelihood had absolute value terms in its Negative Log-Density Vector . This has been fixed.

02-07

  1. Change the adaptive and trace-adaptive derivative_test options so they check that components of the Jacobian, evaluated by eval_jac_g in ipopt_fixed_eval_jac_g , that are not in the sparsity pattern, are zero.

  2. Change adaptive derivative check to test Hessian evaluated by eval_h in ipopt_fixed_eval_h , when quasi_fixed is false.

  3. Change the ipopt eval_h documentation to specify either upper or lower triangle; e.g., see iRow in ipopt_xam_eval_h .

02-05

  1. The scaling of the fixed effects optimization has been changed to not include components of the fixed effects for which the lower and upper limits are equal; see fixed_scale .

  2. Reduce the changes of a random failure (false negative result) during the testing of lasso.cpp .

01-23

There was a mistake in the removal of use_atomic_cholesky ; see 01-21 below. This has been fixed.

01-22

Improve the instructions for combining both Debug and Release versions of cppad_mixed on the same system.

01-21

Remove the use_atomic_cholesky option from the cmake script.

01-15

The flags -Wpedantic , -Wshadow and -Wconversion were added to the default value of extra_cxx_flags . In addition, fix the corresponding warning messages generated by gcc during the cppad_mixed part of the instal process.

01-14

Change the ran_likelihood_hes.cpp example to use the offset log normal distribution and eliminate the chance of random failure during its test.