OptimizationProblems.jl problem classification

It is possible to access information on the problems implemented in OptimizationProblems.jl without loading the problems using the package's own classification. ​

using OptimizationProblems

Each problem has its own metadata structure, and there is a global metadata structure regrouping all the information.

Problem's metadata

Each problem's metadata is accessible with OptimizationProblems.nameoftheproblem_meta and regroups in a Dict most of the essential information regarding each problem.

OptimizationProblems.AMPGO02_meta

Dict{Symbol, Any} with 17 entries:
  :has_equalities_only    => false
  :origin                 => :unknown
  :has_inequalities_only  => false
  :defined_everywhere     => missing
  :has_fixed_variables    => false
  :variable_ncon          => false
  :nvar                   => 1
  :is_feasible            => true
  :minimize               => true
  :ncon                   => 0
  :name                   => "AMPGO02"
  :best_known_lower_bound => -Inf
  :objtype                => :other
  :best_known_upper_bound => 0.839498
  :has_bounds             => false
  :variable_nvar          => false
  :contype                => :unconstrained

See ? OptimizationProblems.meta for more documentation on the various entries and their default values.

This structre is completed by getters to access the number of variables, get_nameoftheproblem_nvar,the number of constraints, get_nameoftheproblem_ncon, the number of linear constraints, get_nameoftheproblem_nlin, the number of nonlinear constraints, get_nameoftheproblem_nnln, the number of equality constraints, get_nameoftheproblem_nequ, and the number of inequality constraints, get_nameoftheproblem_nineq.

OptimizationProblems.get_AMPGO02_nvar()

1

For scalable problems the entry :variable_nvar (and/or :variable_ncon) is set as true and one can access the number of variables by passing the parameters to the getter functions. By default, the number of variables set in the meta is obtained using OptimizationProblems.default_nvar as a parameter to define the problem.

OptimizationProblems.arglina_meta

Dict{Symbol, Any} with 17 entries:
  :has_equalities_only    => false
  :origin                 => :unknown
  :has_inequalities_only  => false
  :defined_everywhere     => missing
  :has_fixed_variables    => false
  :variable_ncon          => false
  :nvar                   => 100
  :is_feasible            => true
  :minimize               => true
  :ncon                   => 0
  :name                   => "arglina"
  :best_known_lower_bound => -Inf
  :objtype                => :least_squares
  :best_known_upper_bound => 500.0
  :has_bounds             => false
  :variable_nvar          => true
  :contype                => :unconstrained

OptimizationProblems.get_arglina_nvar(n = 10)

10

OptimizationProblems.PureJuMP.arglina(n = 10)

A JuMP Model
Minimization problem with:
Variables: 10
Objective function type: Nonlinear
Model mode: AUTOMATIC
CachingOptimizer state: NO_OPTIMIZER
Solver name: No optimizer attached.
Names registered in the model: x

Global meta

This package collects all the metadata in a single DataFrame.

OptimizationProblems.meta

Then, it is very simple to filter problems using queries on DataFrame. We refer to the documentation of DataFrames.jl for tutorials. For instance, if one wants to select unconstrained scalable problems and use (:nvar, :name).

meta = OptimizationProblems.meta
names_pb_vars = meta[(meta.variable_nvar .== true) .& (meta.ncon .== 0), [:nvar, :name]]

Then, one can prepare a list of problems using the selected ones.

using ADNLPModels
adproblems = (
  eval(Meta.parse("ADNLPProblems.$(pb[:name])()")) for pb in eachrow(names_pb_vars)
)

Base.Generator{DataFrames.DataFrameRows{DataFrames.DataFrame}, Main.var"#1#2"}(Main.var"#1#2"(), 81×2 DataFrameRows
 Row │ nvar   name
     │ Int64  String
─────┼──────────────────
   1 │    91  NZF1
   2 │   100  arglina
   3 │   100  arglinb
   4 │   100  arglinc
   5 │   100  argtrig
   6 │   100  arwhead
   7 │     3  bard
   8 │   100  bdqrtic
  ⋮  │   ⋮        ⋮
  75 │   100  srosenbr
  76 │   100  tointgss
  77 │   100  tquartic
  78 │   100  tridia
  79 │   100  vardim
  80 │    31  watson
  81 │   100  woods
         66 rows omitted)