Parameter Types for Optimization Problems

Introduction

When working with optimization problems, LearningToOptimize.jl supports multiple parameterization strategies. These strategies influence the behavior and performance of the solver and allow flexibility in retrieving information such as dual variables.


Comparison

Parameter TypeSupported ProblemsDual SupportPerformanceNotes
JuMPParameterTypeAllYesModerateGeneral-purpose; slower.
JuMPNLPParameterTypeNLP OnlyNoFastOptimized for NLP problems.
POIParameterType (Default)Linear, ConicYesFastRequires POI-wrapped solvers.

Supported Parameter Types

  1. POIParameterType (Default):

    • Description:
      • Extends MOI.Parameters for linear and conic problems.
      • Compatible with solvers wrapped using ParametricOptInterface.
      • Supports fetching duals w.r.t. parameters.
    • Limitations:
      • Not compatible with nonlinear solvers.
    • Usage Example: Default behavior when using ProblemIterator without specifying param_type.
  2. JuMPParameterType:

    • Description:

      • Adds a variable as a parameter with an additional constraint during solve_batch.
      • Slower compared to other types but supports fetching duals w.r.t. the parameter.
      • Compatible with all problem types.
    • Usage Example:

      using JuMP, LearningToOptimize
      
      model = JuMP.Model(HiGHS.Optimizer)
      @variable(model, x)
      p = @variable(model, _p)
      @constraint(model, cons, x + _p >= 3)
      @objective(model, Min, 2x)
      
      num_p = 10
      problem_iterator = ProblemIterator(
          Dict(p => collect(1.0:num_p));
          param_type = LearningToOptimize.JuMPParameterType
      )
      
      recorder = Recorder{ArrowFile}("output.arrow"; primal_variables = [x], dual_variables = [cons])
      solve_batch(problem_iterator, recorder)
    • Advantages:

      • Works with all solvers and problem types.
      • Duals w.r.t. parameters are available.
  3. JuMPNLPParameterType:

    • Description:

      • Utilizes MOI’s internal parameter structure.
      • Optimized for speed but limited to nonlinear programming (NLP) problems.
      • Does not support fetching duals w.r.t. parameters.
    • Usage Example:

      using JuMP, LearningToOptimize
      
      model = JuMP.Model(Ipopt.Optimizer)
      @variable(model, x)
      p = @variable(model, _p in MOI.Parameter(1.0))
      @constraint(model, cons, x + _p >= 3)
      @objective(model, Min, 2x)
      
      num_p = 10
      problem_iterator = ProblemIterator(
          Dict(p => collect(1.0:num_p));
          param_type = LearningToOptimize.JuMPNLPParameterType
      )
      
      recorder = Recorder{ArrowFile}("output.arrow"; primal_variables = [x], dual_variables = [cons])
      solve_batch(problem_iterator, recorder)
    • Advantages:

      • Fast and efficient for NLP problems.
      • No external wrappers required.