Skip to content

CHANGES.md

Latest commit

 

History

History
724 lines (601 loc) · 28.3 KB

CHANGES.md

File metadata and controls

724 lines (601 loc) · 28.3 KB

Change history for MP-Opt-Model

since version 5.0

4/20/26

  • Move mp.opt_model.is_mixed_integer() logic into new are_all_continuous() method of mp.sm_variable, and make the former a simple wrapper.
  • Fix bug with parameter caching in mp.sm_variable, where calling params() without requesting the variable type would not cache it, so subsequent requests for the variable time incorrectly returned an empty string. Now the variable type is assembled and cached any time there is a non-continuous variable present.

12/11/25

  • Add clear_cached_params() method to mp.set_manager_opt_model and mp.opt_model, to clear aggregated parameters cached by the params() method.

12/10/25

  • Add support for the Gurobi's new PDHG LP solver in qps_gurobi().

11/14/25

  • Add support for the new HiPO interior point LP solver in HiGHS, including have_feature_highs_hipo() to detect availability.

10/28/25

  • Add fix_integer and relax_integer options to miqps_master() and move the implementations from mp.opt_model.solve() to miqps_master(). Note: Since the implementation is in miqps_master() it is not available directly in the individual solver-specific miqps_<solver>() functions.

10/21/25

  • Add fix_integer option for mp.opt_model.solve(). Set to true to fix any integer variables to their initial values, as specified in x0 or opt.x0.
  • Add MILP example from examples/miqp_ex1.m to MILP tests.
  • Fix typo bug in dimension checking in set_params() for quadratic costs.

10/3/25

  • Avoid error displaying mp.opt_model object when variable was added with a sparse dimension argument.

9/25/25

  • Improve reliability of detection of CLP MEX file.

7/18/25

  • Fix failing tests in t_qcqps_master when 'DEFAULT' implies 'FMINCON' (as in MATLAB Online).

Version 5.0 - Jul 12, 2025

7/12/25

  • Release 5.0.
  • Move examples from lib/t to their own examples directory.

7/3/25

  • Add support for Artelys Knitro 15.x which required changes to the prior options handling.
  • INCOMPATIBLE CHANGE: The knitro_opts field of the opt input to nlps_master() and nlps_knitro() and the solve() method of opt_model has been redesigned. It is now a raw Artelys Knitro options struct, so the opts, tol_x and tol_f fields are no longer valid. For tol_x and tol_f, use xtol and ftol, and the contents of opts should be placed directly in the top level of the knitro_opts field.
  • INCOMPATIBLE CHANGE: Remove support for older versions of Knitro, including all references to ktrlink for pre-v9 versions. Currently supports Artelys Knitro version 13.1 and later.

6/21/25

  • Add new mp.struct2object() function to convert a struct back to the object from which it was created. Helps with workarounds to the fact that Octave still (as of 10.x) does not support saving/loading of classdef objects. This function allows objects to implement to_struct() and from_struct() methods to facilitate the process.
  • Add to_struct() and from_struct() methods to mp.opt_model, mp.set_manager, and legacy opt_model classes, to facilitate trivial conversion between objects, which Octave cannot save/load, and structs, which it can.

6/10/25

  • Fix handling of scalar inputs for vector parameters when adding an empty set of variables or linear/quadratic constraints. Now properly "expands" them to an empty vector (issue #16).
  • Update handling by mp.sm_quad_cost of constant term for quadratic costs (issue #15):
    • When H is empty, a scalar k will no longer be expanded (implicitly) to a vector, rather it will result in a scalar cost set.
    • When H is a vector and k is a scalar, k will be expanded explicitly to a vector. The mp.sm_quad_cost_legacy class is unchanged, so this change affects only mp.opt_model, not the legacy opt_model.

5/23/25

  • Add support to qps_master() and miqps_master() for the open-source HiGHS solver for LP, QP and MILP problems, including functions miqps_highs(), qps_highs(), have_feature_highs(), highsver(), and highs_options(). Uses the HiGHSMEX interface for HiGHS.
  • Fix that using miqps_master() with 'DEFAULT' solver to solve an LP/QP problem without a MILP/MIQP solver available incorrectly threw a fatal error stating there was no solver available.

5/7/25

  • Fix crash in mp.sm_quad_cost.display_soln() when linear coefficient is empty.

4/30/25

  • Add support for quadratic constraint and QCQP (quadratically- constrained quadratic programming) problems. Thanks to Wilson Gonzalez Vanegas.
    • Add new top-level wrapper function qcqps_master() to provide a standard unified interface quadratically-constrained quadratic programming (QCQP) problems, with the ability to provide solver- specific input options.
    • Add qcqps_gurobi(), qcqps_knitro(), qcqps_nlps() with interface that matches qcqps_master() to handle implementation for Gurobi, Artelys Knitro, fmincon, IPOPT, and MIPS solvers.
    • Add mpopt2qcqpopt() to set up an options struct for qcqps_master() based on a MATPOWER options struct.
    • Add mp.sm_quad_constraint class to handle quadratic constraints in opt_model.
    • Add qcn property to opt_model for quadratic constraints and automatic detection of a new QCQP problem type that is sent to qcqps_master() to solve.

3/19/25

  • Update miqps_<solver>() functions to avoid changing MIP solution values in price computation stage. It was rounding integer variables, potentionally causing a small discrepancy between the objective value reported by the solver and the value obtained by computing directly from the returned solution x.

3/8/25

  • Update mosek_options() for MOSEK 11.x compatibility.

3/1/25

  • Add support to qps_master() for Artelys Knitro solver for LP and QP problems, including functions qps_knitro(), knitrover(), and artelys_knitro_options(). Thanks to Wilson Gonzalez Vanegas.

2/28/25

  • Throw error if opt_model.parse_soln() is called for an unsolved model.

10/29/24

  • Add relax_integer option for opt_model.solve(). Set to true to easily solve the LP or QP relaxation of a mixed integer LP or QP.

8/17/24

  • Add mp.set_manager_opt_model base class to handle common opt_model functionality, such as handling solutions, for the individual field object subclasses.
  • Refactor opt_model to move lots of functionality into new mp.set_manager_opt_model subclasses:
    • mp.sm_lin_constraint - set manager class for linear constraints
    • mp.sm_nln_constraint - set manager class for nonlinear constraints
    • mp.sm_nln_cost - set manager class for general nonlinear costs
    • mp.sm_quad_cost - set manager class for quadratic costs
    • mp.sm_variable - set manager class for variables
  • Deprecate the following opt_model methods in favor of methods of the individual mp.set_manager objects contained by the opt_model object:
    • add_named_set() --> use mp.set_manager.add()
    • describe_idx() --> use mp.set_manager.describe_idx()
    • getN() --> use mp.set_manager.get_N()
    • init_indexed_name() --> use mp.set_manager.init_indexed_name()
    • set_type_idx_map() --> use mp.set_manager.set_type_idx_map()
    • add_lin_constraint() --> use mp.sm_lin_constraint.add()
    • add_nln_constraint() --> use mp.sm_nln_constraint.add()
    • add_nln_cost() --> use mp.sm_nln_cost.add()
    • add_quad_cost() --> use mp.sm_quad_cost.add()
    • add_var() --> use mp.sm_variable.add()
    • eval_lin_constraint() --> use mp.sm_lin_constraint.eval()
    • eval_nln_constraint() --> use mp.sm_nln_constraint.eval()
    • eval_nln_constraint_hess() --> use mp.sm_nln_constraint.eval_hess()
    • eval_nln_cost() --> use mp.sm_nln_cost.eval()
    • eval_quad_cost() --> use mp.sm_quad_cost.eval()
    • init_indexed_name() --> use mp.set_manager.init_indexed_name()
    • params_lin_constraint() --> use mp.sm_lin_constraint.params()
    • params_nln_constraint() --> use mp.sm_nln_constraint.params()
    • params_nln_cost() --> use mp.sm_nln_cost.params()
    • params_quad_cost() --> use mp.sm_quad_cost.params()
    • params_var() --> use mp.sm_variable.params()
    • set_params() --> use mp.set_manager.set_params()
    • varsets_cell2struct() --> use mp.sm_variable.varsets_cell2struct()
    • varsets_idx() --> use mp.sm_variable.varsets_idx()
    • varsets_len() --> use mp.sm_variable.varsets_len()
    • varsets_x() --> use mp.sm_variable.varsets_x()
  • INCOMPATIBLE CHANGE: Move parsed solution from soln property of opt_model object to soln property of individual child mp.set_manager_opt_model objects. Currently it is still available at the original location, but this is now deprecated.

7/10/24

  • Add option for opt_model.display_soln() to print to a file.

6/24/24

  • Add mp.set_manager class to encapsulate mp_idx_manager functionality into an individual field object representing a specific set type, rather than in the container class.
  • Refactor mp_idx_manager to use new mp.set_manager class.

Version 4.2 - May 10, 2024

5/10/24

  • Release 4.2.

4/23/24

  • Fix bug in test t_opt_model() for MATLAB R2011b and earlier.

4/22/24

  • Fix false positive in have_feature_fsolve() in case where the file is present, but without a valid license.

4/5/24

  • Fix bug in miqps_mosek() where the lower and upper bounds of binary variables got overwritten with 0 and 1, respectively, effectively relaxing any potentially tighter bounds provided as input.
  • Clear cached parameters after updating linear constraints via opt_model.set_params().
  • Add caching of aggregate output parameters in opt_model.params_var().

3/26/24

3/21/24

  • Add to the parse_soln() method of opt_model an optional stash input argument that, if present and true, causes the parsed solution to be stored back in the object, as the solve() method was already doing when opt.parse_soln is true.
  • Add new method has_parsed_soln() to opt_model to check for availability of a parsed solution in the model.

3/1/24

  • Update have_feature_ipopt() to recognize IPOPT MEX installations from Enrico Bertolazzi's mexIPOPT, which include MEX files that have been renamed to architecture-specific names along with an ipopt.m wrapper function to call the appropriate one. Thanks to Carlos Murillo-Sánchez. Note: While MP-Opt-Model no longer requires this, my recommendation is still to simply rename the MEX file to ipopt.<mexext>, with the appropriate architecture-specific extension, and delete the unnecessary ipopt.m entirely.

1/31/24

  • Add convert_lin_constraint() and convert_lin_constraint_multipliers() functions to eliminate code duplication for common task of converting linear constraints and their multipliers between a single set of doubly-bounded inequality constraints and separate sets of equality and upper-bounded inequality constraints.
  • Change solver for CPLEX price computation stage in miqps_cplex() from primal simplex to dual simplex (probably no impact, except it was a simple way to get a newly failing test in another project to pass again).

12/8/23

  • Always skip price computation stage in miqps_<solver>() functions for pure (as opposed to mixed) integer problems.

11/29/23

  • Add support to miqps_master() for calling miqps_<my_solver>() by setting opt.alg to '<MY_SOLVER>' to allow for handling custom MILP/MIQP solvers.

11/8/23

  • Add support to nlps_master() for calling nlps_<my_solver>() by setting opt.alg to '<MY_SOLVER>' to allow for handling custom NLP solvers.

11/6/23

  • Add to opt_model.add_lin_constraint() the option to provide/store the transpose of the A matrix instead of the original. This can potentially save significant memory for sparse matrices with many more columns than rows. E.g. storage constraints in MOST for 8760 hour planning horizon.

10/13/23

  • Update opt_model.display_soln() to avoid displaying an infinite average for quadratic costs when corresponding quantity is zero.

9/13/23

  • Clear cached parameters after updating quadratic costs via opt_model.set_params().

9/12/23

  • Tweak some MI/QPS solver tests to make them more robust for 'DEFAULT' solver under different environments.

3/27/23

  • Update for compatibility with MATLAB R2023a (Optimization Toolbox 9.5) which removed x0 as a valid input to linprog.

3/7/23

  • Add opt_model.is_solved() method to determine if the model has been solved and contains the solution.
  • Add opt_model.display_soln() method to display the results of a solved model, including values, bounds and shadow prices for variables and linear constraints, values and shadow prices for nonlinear constraints, and individual cost components.

Version 4.1 - Dec 13, 2022

12/13/22

  • Release 4.1.

11/14/22

  • Allow for Gurobi's Method option to be set to 5, for deterministic concurrent simplex.

10/27/22

  • Add runtime field to output argument of qps_glpk() and qps_mosek().
  • Add support to qps_master() for calling qps_<my_solver>() by setting opt.alg to '<MY_SOLVER>' to allow for handling custom LP/QP solvers.

7/5/22

  • Update for compatibility with Artelys Knitro 13.1 and later.

4/8/22

  • Relax some test tolerances to prevent failure with Gurobi 9.5.

3/3/22

  • Add elapsed time in seconds to results of the solve() method of opt_model, returned in om.soln.output.et.

2/14/22

  • Skip some fmincon tests using interior point algorithm with finite difference Hessian that began failing in MATLAB R2021b.

Version 4.0 - Oct 18, 2021

10/18/21

  • Release 4.0.

5/8/21

  • Add several enhancements to mp_idx_manager/set_type_idx_map().
    • Speed improvements
    • Add idxs = [] option to specify "all"
    • Add optional group_by_name input argument.

5/3/21

  • Add support for new class of problems - parameterized nonlinear equations (PNE). Either create a model with only equality constraints (no inequalities or costs) and with number of variables equal to 1 more than number of constraints, or call pnes_master() directly. See Section 4.5 of User's Manual for details.

    • Predictor/corrector numerical continuation method for tracing solution curves for PNE problems.
    • Plotting of solution curves.
    • User-defined event functions and callback functions.
    • Warm-start capabilities.

    Thanks to Shrirang Abhyankar and Alexander Flueck for contributions to this feature.

  • Add functions:

    • pnes_master()
    • pne_callback_default()
    • pne_callback_nose()
    • pne_callback_target_lam()
    • pne_detect_events()
    • pne_detected_event()
    • pne_event_nose()
    • pne_event_target_lam()
    • pne_pfcn_arc_length()
    • pne_pfcn_natural()
    • pne_pfcn_pseudo_arc_length()
    • pne_register_callbacks()
    • pne_register_events()
    • mpopt2pneopt()

1/21/21

  • Refactor describe_idx() into a new method, set_type_idx_map(), that returns in information in a programmatically usable form, and an updated describe_idx() that calls the new method, then formats the results in the expected char array(s).

1/5/21

  • Calling the problem_type() or is_mixed_integer() method on an empty model no longer cause a fatal error.

12/16/20

  • Update to use labels from set_types property as headers for opt_model.display() to simplify things and facilitate use by subclasses.

12/1/20

  • Move init_set_types() call out of opt_model constructor to avoid complexity of other workarounds for bug in Octave 5.2 and earlier related to inheritance of methods called during construction. Now init_set_types() is called automatically as needed after object construction and before object use in add_var(), display() and init_indexed_name().

11/24/20

  • Add optional threshold for detecting failure of LEQ solve, by setting the leq_opt.thresh option. If the absolute value of any element of the solution vector exceeds the threshold, exitflag is set to 0, indicating failure.

11/17/20

  • Fix examples of om.set_params() usage in manual and help text.

Version 3.0 - Oct 8, 2020

10/8/20

  • Release 3.0.

9/23/20

  • Move have_feature() to MP-Test and respective feature detection functions to MP-Test, MIPS, and MATPOWER.
    • MP-Test
      • have_feature()
      • have_feature_matlab()
      • have_feature_octave()
    • MIPS
      • have_feature_lu_vec()
      • have_feature_pardiso_legacy()
      • have_feature_pardiso_object()
      • have_feature_pardiso()
    • MATPOWER
      • have_feature_e4st()
      • have_feature_minopf()
      • have_feature_most()
      • have_feature_pdipmopf()
      • have_feature_regexp_split()
      • have_feature_scpdipmopf()
      • have_feature_sdp_pf()
      • have_feature_smartmarket()
      • have_feature_syngrid()
      • have_feature_tralmopf()

9/18/20

  • Add have_feature() as a modular, extensible alternative to have_fcn(), where the detection of a feature named <tag> is implemented by the function have_feature_<tag>().
  • Make have_fcn() a simple wrapper around the new have_feature().

9/16/20

  • Add set_params() method to modify parameters of existing variables, costs and constraints of an MP-Opt-Model object.
  • Calling params_var() method with empty idx no longer results in fatal error.
  • For opt_model, fixed incorrect evaluation of constant term in vector valued quadratic costs with constant term supplied as a vector.
  • Simplified logic to determine whether a quadratic cost for an MP-Opt-Model object is vector vs. scalar valued. If the quadratic coefficient is supplied as a matrix, the cost is scalar varied, otherwise it is vector valued.

9/14/20

  • Allow v0, vl, and vu inputs to opt_model.add_var() method, and l and u inputs to opt_model.add_lin_constraint() to be scalars that get expanded automatically to the appropriate vector dimension.

9/11/20

  • Add get_soln() method to opt_model for extracting solved results for a given named set of variables, constraints or costs.
  • Add parse_soln() method which returns a struct with a complete set of solution vector and shadow price values for a solved model.

9/10/20

  • Add caching of problem_type() return value.

9/1/20

  • Add support for OSQP solver from https://osqp.org for LP and QP problems, including functions qps_osqp(), osqpver(), and osqp_options().

8/31/20

  • Save the results of solve() method to the soln field of the MP-Opt-Model object.

8/28/20

  • Add eval_lin_constraint() method to evaluate the constraint values for the full set or an individual named subset of linear constraints.

8/27/20

  • Starting point supplied to solve() via opt.x0 is no longer ignored for nonlinear equations.

Version 2.1 - Aug 25, 2020

8/25/20

  • Release 2.1.
  • Add core NLEQ solver function nleqs_core() with arbitrary, user-defined update function, used to implement Gauss-Seidel and Newton solvers, nleqs_gauss_seidel() and nleqs_newton().

8/20/20

  • Add linear equation ('LEQ') problem type for models with equal number of variables and linear equality constraints, no costs, and no inequality or nonlinear equality constraints. Solved via mplinsolve().
  • The solve() method of opt_model can now automatically handle mixed systems of equations, with both linear and nonlinear equality constraints.

7/30/20

  • Add fast-decoupled Newton's and Gauss-Seidel solvers for nonlinear equations. Use alg = 'FD' and alg = 'GS' with nleqs_master(). See also nleqs_fd_newton() and nleqs_gauss_seidel().

7/29/20

  • Allow solve() method to pass along number of requested output arguments *_master() solver functions.
  • INCOMPATIBLE CHANGE: In output return value from nleqs_newton(), changed the normF field of output.hist to normf, for consistency in using lowercase f everywhere.

Version 2.0 - Jul 8, 2020

7/8/20

  • Release 2.0.

7/3/20

  • Add to eval_nln_constraint() method the ability to compute constraints for a single named set.

7/2/20

  • Skip evaluation of gradient if eval_nln_constraint() is called with a single output argument.
  • Add params_nln_constraint() method, and add documentation to the manual for it and params_nln_cost().

7/1/20

  • Add mpopt2nleqopt() to create or modify an nleqs_master() options struct from a MATPOWER options struct.
  • Add table of valid have_fcn() input tags to User's Manual.

6/24/20

  • Add support for nonlinear equations (NLEQ) to opt_model. For problems with only nonlinear equality constraints and no costs, the problem_type() method returns 'NLEQ' and the solve() method calls nleqs_master() to solve the problem.
  • Add tests for solving LP/QP, MILP/MIQP, NLP and NLEQ problems via opt_model.solve().

6/17/20

  • Add nleqs_master() function as unified interface for solving nonlinear equations, including implementations for fsolve and Newton's method in functions nleqs_fsolve() and nleqs_newton(), respectively.

6/16/20

  • Add new 'fsolve' tag to have_fcn() to check for availability of fsolve() function.

5/11/20

  • Remove redundant MIPS tests from test_mp_opt_model.

Version 1.0 - May 8, 2020

5/8/20

  • Release 1.0.

5/7/20

  • Add MP-Opt-Model User's Manual in docs, with LaTeX source in docs/src.
  • Add usage examples to README.md.

4/30/20

  • Add README.md, CHANGES.md, AUTHORS, CONTRIBUTING.md, LICENSE.
  • Refactor opt_model class to inherit from new abstract base class mp_idx_manager, which can be used to manage the indexing of other sets of parameters, etc. in other contexts.

Version 0.8 - Apr 29, 2020

4/29/20

  • Version 0.8.
  • Add mpomver() to define MP-Opt-Model version information.
  • INCOMPATIBLE CHANGE: Renamed the following functions and modified the order of their input args so that the MP-Opt-Model object appears first. Ideally, these would be defined as methods of the opt_model class, but Octave 4.2 and earlier is not able to find them via a function handle (as used in the solve() method) if they are inherited by a subclass.
    • opf_consfcn() --> nlp_consfcn()
    • opf_costfcn() --> nlp_costfcn()
    • opf_hessfcn() --> nlp_hessfcn()
  • Add GitHub Actions CI workflow and Travis-CI configuration.
  • Add test_mp_opt_model() to run all tests.
  • Remove MATPOWER dependencies.
  • Move code related to solver interfaces, opt_model and a few other functions like have_fcn() and nested_struct_copy() from main MATPOWER repository to new MP-Opt-Model repository.

⬆ In MP-Opt-Model repository ⬆

⬇ In MATPOWER repository ⬇

4/28/20

  • Move deprecated opt_model methods and code related to legacy user-defined OPF costs from opt_model to opf_model.
  • INCOMPATIBLE CHANGE: Modify order of default output arguments of opt_model.get_idx() (again), removing the one related to legacy costs.

3/18/20

  • Add nlpopf_solver() based on the new solver() method of opt_model. This single function replaces mipsopf_solver(), fmincopf_solver(), ipoptopf_solver(), and ktropf_solver().
  • Convert dcopf_solver() to use the new solver() method of opt_model instead of calling qps_matpower() manually.
  • Add new top-level wrapper function nlps_matpower() to provide a standard interface for MATPOWER's nonlinear program (NLP) solvers (fmincon, IPOPT, Artelys Knitro, and MIPS), with calling syntax similar to mips(). It includes the ability to pass in solver-specific input options.
  • Add nlps_fmincon(), nlps_ipopt() and nlps_knitro(), with interface that matches nlps_matpower() to handle implementation for fmincon, IPOPT, and Artelys Knitro solvers, respectively.
  • Add mpopt2nlpopt() to set up an options struct for nlps_matpower() based on a MATPOWER options struct.
  • Add three new methods to opt_model class:
    • is_mixed_integer() - returns true if the model includes any binary or integer variables
    • problem_type() - returns one of the following strings, based on the characteristics of the variables, costs and constraints in the model:
      • 'NLP' - nonlinear program
      • 'LP' - linear program
      • 'QP' - quadratic program
      • 'MILP' - mixed-integer linear program
      • 'MIQP' - mixed-integer quadratic program
    • solve() - solves the model using qps_matpower(), miqps_matpower(), or nlps_matpower(), depending on the problem type ('MINLP' problems are not yet implemented)

3/12/20

  • Fix bug in ktropf_solver() where Artelys Knitro was still using fmincon options.

3/6/20

  • Fix issue with missing objective function value from miqps_mosek() and qps_mosek() when return status is "Stalled at or near optimal solution."

3/4/20

  • Remove unused input arguments from opf_consfcn() and opf_hessfcn().

2/27/20

  • Add copy() method to opt_model class to get around issues with inheritance in constructors that was preventing copy constructor from working in Octave 5.2 and earlier (see also Octave bug 52614.

2/26/20

  • Significant performance improvement for CPLEX on small problems by eliminating call to cplexoptimset(), which was a huge bottleneck.
  • Fix CPLEX 12.10 compatibility issue #90.

2/18/20

  • Artelys Knitro 12.1 compatibility fix.

8/15/19

  • Improve performance of opt_model.add_named_set(). (See issue #79.) Thanks to Baraa Mohandes.
  • Refactor code in opt_model.params_lin_constraint() and opt_model.params_quad_cost() to speed up sparse matrix construction when there are lots of constraint or cost sets. Results in significant speedups for some problems during problem setup in MOST. (See pull request #70.) Thanks to Daniel Muldrew.

Version 0.7.0 - Jun 20, 2019

6/20/19

  • This change history begins with the code that was part of the MATPOWER 7.0 release, which is tagged as version 0.7.0 in the MP-Opt-Model repository.