block band structure

[apozharski]

Lifting now maintains structure if a vdx mpcc is passed.

[nosnoc-ci-bot]

Summary - mpecopt coverage

Summary

Generated on:	03/17/2025 - 20:23:33
Coverage date:	03/17/2025 - 20:23:15
Parser:	Cobertura
Assemblies:	2
Classes:	47
Files:	47
Line coverage:	31.3% (1175 of 3746)
Covered lines:	1175
Uncovered lines:	2571
Coverable lines:	3746
Total lines:	5926
Covered branches:	0
Total branches:	0
Method coverage:	Feature is only available for sponsors
Tag:	21_13909076109

Coverage

- 15.5%

Name	Line	Branch
****	15.5%	****
addaptive_active_set_identification	0%
BNLPProjectionStrategy
check_feasiblity_of_lpec	0%
check_if_B_stationary	0%
create_comparison_table	0%
create_lpec_functions	100%
create_lpec_subproblem	48.9%
create_mpec_functions	68.7%
create_phase_i_nlp_solver	0%
create_phase_i_nlp_solver_dev	0%
determine_multipliers_based_stationary_point	14.6%
fast_B_stationarity_check	0%
find_active_sets	100%
find_active_sets_piece_nlp	68%
find_nonscalar	88.2%
HomotopySolverOptions	100%
HomotopySteering
InitializationStrategy
lpec_fallback_strategy	0%
lpec_fallback_strategy_phase_ii	0%
lpec_solver	20%
LpecSolver
LPECSolverOptions	100%
mpec_active_set_generate_problem_functions	0%
mpec_active_set_nosnoc	0%
mpec_active_set_preprocess	0%
mpec_homotopy_solver	0%
mpec_optimizer	0%
mpecopt_phase_ii	0%
nice_plot_colors	0%
PieceNLPStartegy
plot_lpec	0%
print_iter_details	0%
print_iter_header	100%
print_iter_header_objective	0%
print_iter_line	100%
print_iter_line_objective	0%
print_iter_stats	91.6%
print_iter_stats_objective	0%
print_iter_summary	100%
print_iter_summary_scholtes	0%
print_phase_i	100%
print_phase_ii	100%
project_to_bounds	100%
unfold_struct	0%

mpecopt - 62.4%

Name	Line	Branch
mpecopt	62.4%	****
mpecopt.Options	100%
mpecopt.Solver	61.7%

[nurkanovic]

unrelated to you changes the multipliers should should be in results to be consistent with reg_homotopy?

[apozharski]

Yea I think so.

[nurkanovic]

This change is very welcome. :D

[nurkanovic]

running examples/cardinality_optimization.m shows that

• the new interface is broken - diverging iterates (maybe wrong lower bounds for comps?)
• also the legacy interface mpec_optimizer used in all benchmarks is also broken, changes in mpceopt_phase_ii broke it.
  (as the phase ii is anyway a method in the new interface, keeping the old working as it was would be ok)

[nurkanovic]

also benchmarks are broken due to breaking the mpecopt_phase_ii

Error in mpecopt_phase_ii (line 395)
solution.x = x_k(dims.map_w(1:dims.n_primal_non_lifted));

Error in mpec_optimizer (line 588)
[solution,stats] = mpecopt_phase_ii(mpec_casadi,lpec_casadi,dims,settings,solver_initalization,stats,phase_ii);

Error in mpec_benchmark_dtable_loop (line 84)
        [result,stats] = solver_functions{ii}(mpec_struct,solver_initalization,options);

[nurkanovic]

would make sense to have also phase_I as a method to be more consistent.
can be a seperate PR/issue for requesting refactoring.

[nurkanovic]

please add some comments what are these specalized functons doing and what are they needed for?

[apozharski]

The benchmarks and other bugs should now have been fixed.

[nurkanovic]

LGTM 🚀