contrib.alternative_solutions TODO

[emma58]

Summary

I'm using this issue to compile outstanding improvements for contrib.alternative_solutions not addressed in #3270:

• Transition from APPSI to contrib.solver
• Once solvers are standardized, switch to having users pass in configured solvers rather than separately passing solver_options arguments
• Break balas.enumerate_binary_solutions function into smaller functions that can be tested
• Implement random and norm objectives in lp_enum.
• Add handling for mixed-integer case in lp_enum_solnpool
• Support variables argument being something other than None in lp_enum.enumerate_linear_solutions
• Refactor shifted_lp.get_shifted_linear_model to use the compile_standard_form writer from repn.plugins.standard_form
• Refactor balas and obbt tests to inherit from unittest.TestCase and reduce bare assertions.

[jlgearh]

Some additional issues include:

• Add handling for mixed-binary case in lp_enum
• Do more testing for shifted_lp, specficially check the constraints in the reformulated problem
• Parallelize the main for loop in obbt
• Add more tests to test_lp_enum_solnpool (at least those from test_lp_enum). As part of this, handle the case where duplicate solutions can get returned. The pentagonal pyramid test returns 6 solutions. The issue is that we actually get back 7. The same solution get’s returned twice, once in presolve and once in B&B.
• See if the heuristics option can be (re)enabled for lp_enum. This was disabled since Gurobi can produce solutions that are not extreme points during heuristic searches.
• See if the algorithms can be updated to allow for subsets of variables to explored for lp_enum and lp_enum_solnpool.

[whart222]

Some additional issues:

• Refactor the library to use a custom SolutionPool class. Return this solution pool class, rather than a list of Solution objects.
• Develop summary techniques that analyze and visualize a SolutionPool (e.g. consensus patterns)