Add options variable_bounds_analysis and constraint_bounds_analysis in NLPModelMeta (#543)

Author: amontoison

README.md

@@ -118,6 +118,8 @@ Attribute          | Type          | Notes
 `minimize`         | `Bool`        | true if `optimize == minimize`
 `islp`             | `Bool`        | true if the problem is a linear program
 `name`             | `String`      | problem name
+`variable_bounds_analysis` | `Bool` | true if the partition of variables into fixed, lower-bounded, upper-bounded, range-bounded, free, and trivially infeasible sets is computed
+`constraint_bounds_analysis`| `Bool` | true if the partition of constraints into equality, lower-bounded, upper-bounded, range-bounded, free, and trivially infeasible sets is computed
 `sparse_jacobian`  | `Bool`        | true if the Jacobian of the constraints is sparse
 `sparse_hessian`   | `Bool`        | true if the Hessian of the Lagrangian is sparse
 `grad_available`   | `Bool`        | true if the gradient of the objective is available

docs/src/index.md

@@ -113,6 +113,8 @@ Attribute          | Type          | Notes
 `minimize`         | `Bool`        | true if `optimize == minimize`
 `islp`             | `Bool`        | true if the problem is a linear program
 `name`             | `String`      | problem name
+`variable_bounds_analysis` | `Bool` | true if the partition of variables into fixed, lower-bounded, upper-bounded, range-bounded, free, and trivially infeasible sets is computed
+`constraint_bounds_analysis`| `Bool` | true if the partition of constraints into equality, lower-bounded, upper-bounded, range-bounded, free, and trivially infeasible sets is computed
 `sparse_jacobian`  | `Bool`        | true if the Jacobian of the constraints is sparse
 `sparse_hessian`   | `Bool`        | true if the Hessian of the Lagrangian is sparse
 `grad_available`   | `Bool`        | true if the gradient of the objective is available

src/nlp/meta.jl

@@ -51,6 +51,8 @@ The following keyword arguments are accepted:
 - `minimize`: true if optimize == minimize
 - `islp`: true if the problem is a linear program
 - `name`: problem name
+- `variable_bounds_analysis`: whether to compute the partition of variables into fixed, lower-bounded, upper-bounded, range-bounded, free, and trivially infeasible sets
+- `constraint_bounds_analysis`: whether to compute the partition of constraints into equality, lower-bounded, upper-bounded, range-bounded, free, and trivially infeasible sets
 - `sparse_jacobian`: indicates whether the Jacobian of the constraints is sparse
 - `sparse_hessian`: indicates whether the Hessian of the Lagrangian is sparse
 - `grad_available`: indicates whether the gradient of the objective is available
@@ -77,6 +79,9 @@ The following keyword arguments are accepted:
 - `nlin`: number of linear constraints
 - `nnln`: number of nonlinear general constraints
 - `nln`: indices of nonlinear constraints
+
+The attributes `ifix`, `ilow`, `iupp`, `irng`, `ifree`, and `iinf` are not computed if `variable_bounds_analysis` is set to `false`.
+The attributes `jfix`, `jlow`, `jupp`, `jrng`, `jfree`, and `jinf` are not computed if `constraint_bounds_analysis` is set to `false`.
 """
 struct NLPModelMeta{T, S} <: AbstractNLPModelMeta{T, S}
   nvar::Int
@@ -123,6 +128,9 @@ struct NLPModelMeta{T, S} <: AbstractNLPModelMeta{T, S}
   islp::Bool
   name::String
 
+  variable_bounds_analysis::Bool
+  constraint_bounds_analysis::Bool
+
   sparse_jacobian::Bool
   sparse_hessian::Bool
 
@@ -155,6 +163,8 @@ function NLPModelMeta{T, S}(
   minimize::Bool = true,
   islp::Bool = false,
   name = "Generic",
+  variable_bounds_analysis::Bool = true,
+  constraint_bounds_analysis::Bool = true,
   sparse_jacobian::Bool = true,
   sparse_hessian::Bool = true,
   grad_available::Bool = true,
@@ -173,14 +183,23 @@ function NLPModelMeta{T, S}(
   @rangecheck 1 ncon lin
   @assert nnzj == lin_nnzj + nln_nnzj
 
-  ifix = findall(lvar .== uvar)
-  ilow = findall((lvar .> T(-Inf)) .& (uvar .== T(Inf)))
-  iupp = findall((lvar .== T(-Inf)) .& (uvar .< T(Inf)))
-  irng = findall((lvar .> T(-Inf)) .& (uvar .< T(Inf)) .& (lvar .< uvar))
-  ifree = findall((lvar .== T(-Inf)) .& (uvar .== T(Inf)))
-  iinf = findall(lvar .> uvar)
+  if variable_bounds_analysis
+    ifix = findall(lvar .== uvar)
+    ilow = findall((lvar .> T(-Inf)) .& (uvar .== T(Inf)))
+    iupp = findall((lvar .== T(-Inf)) .& (uvar .< T(Inf)))
+    irng = findall((lvar .> T(-Inf)) .& (uvar .< T(Inf)) .& (lvar .< uvar))
+    ifree = findall((lvar .== T(-Inf)) .& (uvar .== T(Inf)))
+    iinf = findall(lvar .> uvar)
+  else
+    ifix = Int[]
+    ilow = Int[]
+    iupp = Int[]
+    irng = Int[]
+    ifree = Int[]
+    iinf = Int[]
+  end
 
-  if ncon > 0
+  if (ncon > 0) && constraint_bounds_analysis
     jfix = findall(lcon .== ucon)
     jlow = findall((lcon .> T(-Inf)) .& (ucon .== T(Inf)))
     jupp = findall((lcon .== T(-Inf)) .& (ucon .< T(Inf)))
@@ -236,6 +255,8 @@ function NLPModelMeta{T, S}(
     minimize,
     islp,
     name,
+    variable_bounds_analysis,
+    constraint_bounds_analysis,
     sparse_jacobian,
     sparse_hessian,
     grad_available,
@@ -272,8 +293,10 @@ function NLPModelMeta(
   minimize::Bool = meta.minimize,
   islp::Bool = meta.islp,
   name = meta.name,
-  sparse_jacobian::Bool = true,
-  sparse_hessian::Bool = true,
+  variable_bounds_analysis::Bool = meta.variable_bounds_analysis,
+  constraint_bounds_analysis::Bool = meta.constraint_bounds_analysis,
+  sparse_jacobian::Bool = meta.sparse_jacobian,
+  sparse_hessian::Bool = meta.sparse_hessian,
   grad_available::Bool = meta.grad_available,
   jac_available::Bool = meta.jac_available,
   hess_available::Bool = meta.hess_available,
@@ -302,6 +325,8 @@ function NLPModelMeta(
     minimize = minimize,
     islp = islp,
     name = name,
+    variable_bounds_analysis = variable_bounds_analysis,
+    constraint_bounds_analysis = constraint_bounds_analysis,
     sparse_jacobian = sparse_jacobian,
     sparse_hessian = sparse_hessian,
     grad_available = grad_available,

src/nlp/tools.jl

@@ -22,31 +22,37 @@ end
 
 Returns whether the problem has bounds on the variables.
 """
-has_bounds(meta::AbstractNLPModelMeta) = length(meta.ifree) < meta.nvar
+function has_bounds(meta::AbstractNLPModelMeta)
+    if meta.variable_bounds_analysis
+        return length(meta.ifree) < meta.nvar
+    else
+        return !all(lv -> isinf(lv), meta.lvar) || !all(uv -> isinf(uv), meta.uvar)
+    end
+end
 
 """
     bound_constrained(nlp)
     bound_constrained(meta)
 
 Returns whether the problem has bounds on the variables and no other constraints.
 """
-bound_constrained(meta::AbstractNLPModelMeta) = meta.ncon == 0 && has_bounds(meta)
+bound_constrained(meta::AbstractNLPModelMeta) = (meta.ncon == 0) && has_bounds(meta)
 
 """
     unconstrained(nlp)
     unconstrained(meta)
 
 Returns whether the problem in unconstrained.
 """
-unconstrained(meta::AbstractNLPModelMeta) = meta.ncon == 0 && !has_bounds(meta)
+unconstrained(meta::AbstractNLPModelMeta) = (meta.ncon == 0) && !has_bounds(meta)
 
 """
     linearly_constrained(nlp)
     linearly_constrained(meta)
 
 Returns whether the problem's constraints are known to be all linear.
 """
-linearly_constrained(meta::AbstractNLPModelMeta) = meta.nlin == meta.ncon > 0
+linearly_constrained(meta::AbstractNLPModelMeta) = (meta.ncon > 0) && (meta.nlin == meta.ncon)
 
 """
     equality_constrained(nlp)
@@ -55,7 +61,13 @@ linearly_constrained(meta::AbstractNLPModelMeta) = meta.nlin == meta.ncon > 0
 Returns whether the problem's constraints are all equalities.
 Unconstrained problems return false.
 """
-equality_constrained(meta::AbstractNLPModelMeta) = length(meta.jfix) == meta.ncon > 0
+function equality_constrained(meta::AbstractNLPModelMeta)
+    if meta.constraint_bounds_analysis
+        return (meta.ncon > 0) && (length(meta.jfix) == meta.ncon)
+    else
+        return (meta.ncon > 0) && all(x -> x[1] == x[2], zip(meta.lcon, meta.ucon))
+    end
+end
 
 """
     inequality_constrained(nlp)
@@ -64,23 +76,41 @@ equality_constrained(meta::AbstractNLPModelMeta) = length(meta.jfix) == meta.nco
 Returns whether the problem's constraints are all inequalities.
 Unconstrained problems return true.
 """
-inequality_constrained(meta::AbstractNLPModelMeta) = meta.ncon > 0 && length(meta.jfix) == 0
+function inequality_constrained(meta::AbstractNLPModelMeta)
+    if meta.constraint_bounds_analysis
+        return (meta.ncon > 0) && (length(meta.jfix) == 0)
+    else
+        return (meta.ncon > 0) && all(x -> x[1] != x[2], zip(meta.lcon, meta.ucon))
+    end
+end
 
 """
     has_equalities(nlp)
 
 Returns whether the problem has constraints and at least one of them is an equality.
 Unconstrained problems return false.
 """
-has_equalities(meta::AbstractNLPModelMeta) = meta.ncon ≥ length(meta.jfix) > 0
+function has_equalities(meta::AbstractNLPModelMeta)
+    if meta.constraint_bounds_analysis
+        return (meta.ncon > 0) && (length(meta.jfix) > 0)
+    else
+        return (meta.ncon > 0) && !all(x -> x[1] != x[2], zip(meta.lcon, meta.ucon))
+    end
+end
 
 """
     has_inequalities(nlp)
 
 Returns whether the problem has constraints and at least one of them is an inequality.
 Unconstrained problems return false.
 """
-has_inequalities(meta::AbstractNLPModelMeta) = meta.ncon > 0 && meta.ncon > length(meta.jfix)
+function has_inequalities(meta::AbstractNLPModelMeta)
+    if meta.constraint_bounds_analysis
+        return (meta.ncon > 0) && (meta.ncon > length(meta.jfix))
+    else
+        return (meta.ncon > 0) && !all(x -> x[1] == x[2], zip(meta.lcon, meta.ucon))
+    end
+end
 
 for meth in [
   :has_bounds,

test/nlp/tools.jl

@@ -1,72 +1,80 @@
 @testset "Problem type functions" begin
-  foo_list = [
-    has_bounds,
-    bound_constrained,
-    unconstrained,
-    linearly_constrained,
-    equality_constrained,
-    inequality_constrained,
-    has_equalities,
-    has_inequalities,
-  ]
-  meta_list = [
-    NLPModelMeta(2),
-    NLPModelMeta(2, lvar = zeros(2), uvar = ones(2)),
-    NLPModelMeta(2, ncon = 1, lcon = [0.0], ucon = [0.0]),
-    NLPModelMeta(2, ncon = 1, lcon = [0.0], ucon = [1.0]),
-    NLPModelMeta(2, ncon = 1, lcon = [0.0], ucon = [Inf]),
-    NLPModelMeta(2, ncon = 1, lcon = [-Inf], ucon = [0.0]),
-    NLPModelMeta(2, ncon = 1, lcon = [0.0], ucon = [1.0], lin = [1]),
-    NLPModelMeta(2, ncon = 2, lcon = [0.0, 0.0], ucon = [1.0, 1.0], lin = [1]),
-    NLPModelMeta(2, ncon = 2, lcon = [0.0, 0.0], ucon = [1.0, 0.0], lin = [1]),
-    NLPModelMeta(2, lvar = zeros(2), uvar = ones(2), ncon = 1, lcon = [0.0], ucon = [0.0]),
-    NLPModelMeta(2, lvar = zeros(2), uvar = ones(2), ncon = 1, lcon = [0.0], ucon = [1.0]),
-    NLPModelMeta(2, lvar = zeros(2), uvar = ones(2), ncon = 1, lcon = [0.0], ucon = [Inf]),
-    NLPModelMeta(2, lvar = zeros(2), uvar = ones(2), ncon = 1, lcon = [-Inf], ucon = [0.0]),
-    NLPModelMeta(
-      2,
-      lvar = zeros(2),
-      uvar = ones(2),
-      ncon = 1,
-      lcon = [0.0],
-      ucon = [1.0],
-      lin = [1],
-    ),
-    NLPModelMeta(
-      2,
-      lvar = zeros(2),
-      uvar = ones(2),
-      ncon = 2,
-      lcon = [0.0, 0.0],
-      ucon = [1.0, 1.0],
-      lin = [1],
-    ),
-    NLPModelMeta(
-      2,
-      lvar = zeros(2),
-      uvar = ones(2),
-      ncon = 2,
-      lcon = [0.0, 0.0],
-      ucon = [1.0, 0.0],
-      lin = [1],
-    ),
-  ]
-  results = Bool[
-    0 1 0 0 0 0 0 0 0 1 1 1 1 1 1 1
-    0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
-    1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
-    0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0
-    0 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0
-    0 0 0 1 1 1 1 1 0 0 1 1 1 1 1 0
-    0 0 1 0 0 0 0 0 1 1 0 0 0 0 0 1
-    0 0 0 1 1 1 1 1 1 0 1 1 1 1 1 1
-  ]
-  for (i, f) in enumerate(foo_list), (j, meta) in enumerate(meta_list)
-    @test f(meta) == results[i, j]
-    @test f(DummyModel(meta)) == results[i, j]
-  end
-  for f in fieldnames(NLPModelMeta), (j, meta) in enumerate(meta_list)
-    @test eval(Meta.parse("get_" * string(f)))(meta) == getproperty(meta, f)
-    @test eval(Meta.parse("get_" * string(f)))(DummyModel(meta)) == getproperty(meta, f)
+  @testset "Analysis = $bool" for bool in (false, true)
+    foo_list = [
+      has_bounds,
+      bound_constrained,
+      unconstrained,
+      linearly_constrained,
+      equality_constrained,
+      inequality_constrained,
+      has_equalities,
+      has_inequalities,
+    ]
+    meta_list = [
+      NLPModelMeta(2, variable_bounds_analysis=bool, constraint_bounds_analysis=bool),
+      NLPModelMeta(2, lvar = zeros(2), uvar = ones(2), variable_bounds_analysis=bool, constraint_bounds_analysis=bool),
+      NLPModelMeta(2, ncon = 1, lcon = [0.0], ucon = [0.0], variable_bounds_analysis=bool, constraint_bounds_analysis=bool),
+      NLPModelMeta(2, ncon = 1, lcon = [0.0], ucon = [1.0], variable_bounds_analysis=bool, constraint_bounds_analysis=bool),
+      NLPModelMeta(2, ncon = 1, lcon = [0.0], ucon = [Inf], variable_bounds_analysis=bool, constraint_bounds_analysis=bool),
+      NLPModelMeta(2, ncon = 1, lcon = [-Inf], ucon = [0.0], variable_bounds_analysis=bool, constraint_bounds_analysis=bool),
+      NLPModelMeta(2, ncon = 1, lcon = [0.0], ucon = [1.0], lin = [1], variable_bounds_analysis=bool, constraint_bounds_analysis=bool),
+      NLPModelMeta(2, ncon = 2, lcon = [0.0, 0.0], ucon = [1.0, 1.0], lin = [1], variable_bounds_analysis=bool, constraint_bounds_analysis=bool),
+      NLPModelMeta(2, ncon = 2, lcon = [0.0, 0.0], ucon = [1.0, 0.0], lin = [1], variable_bounds_analysis=bool, constraint_bounds_analysis=bool),
+      NLPModelMeta(2, lvar = zeros(2), uvar = ones(2), ncon = 1, lcon = [0.0], ucon = [0.0], variable_bounds_analysis=bool, constraint_bounds_analysis=bool),
+      NLPModelMeta(2, lvar = zeros(2), uvar = ones(2), ncon = 1, lcon = [0.0], ucon = [1.0], variable_bounds_analysis=bool, constraint_bounds_analysis=bool),
+      NLPModelMeta(2, lvar = zeros(2), uvar = ones(2), ncon = 1, lcon = [0.0], ucon = [Inf], variable_bounds_analysis=bool, constraint_bounds_analysis=bool),
+      NLPModelMeta(2, lvar = zeros(2), uvar = ones(2), ncon = 1, lcon = [-Inf], ucon = [0.0], variable_bounds_analysis=bool, constraint_bounds_analysis=bool),
+      NLPModelMeta(
+        2,
+        lvar = zeros(2),
+        uvar = ones(2),
+        ncon = 1,
+        lcon = [0.0],
+        ucon = [1.0],
+        lin = [1],
+        variable_bounds_analysis=bool,
+        constraint_bounds_analysis=bool,
+      ),
+      NLPModelMeta(
+        2,
+        lvar = zeros(2),
+        uvar = ones(2),
+        ncon = 2,
+        lcon = [0.0, 0.0],
+        ucon = [1.0, 1.0],
+        lin = [1],
+        variable_bounds_analysis=bool,
+        constraint_bounds_analysis=bool,
+      ),
+      NLPModelMeta(
+        2,
+        lvar = zeros(2),
+        uvar = ones(2),
+        ncon = 2,
+        lcon = [0.0, 0.0],
+        ucon = [1.0, 0.0],
+        lin = [1],
+        variable_bounds_analysis=bool,
+        constraint_bounds_analysis=bool,
+      ),
+    ]
+    results = Bool[
+      0 1 0 0 0 0 0 0 0 1 1 1 1 1 1 1
+      0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
+      1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
+      0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0
+      0 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0
+      0 0 0 1 1 1 1 1 0 0 1 1 1 1 1 0
+      0 0 1 0 0 0 0 0 1 1 0 0 0 0 0 1
+      0 0 0 1 1 1 1 1 1 0 1 1 1 1 1 1
+    ]
+    for (i, f) in enumerate(foo_list), (j, meta) in enumerate(meta_list)
+      @test f(meta) == results[i, j]
+      @test f(DummyModel(meta)) == results[i, j]
+    end
+    for f in fieldnames(NLPModelMeta), (j, meta) in enumerate(meta_list)
+      @test eval(Meta.parse("get_" * string(f)))(meta) == getproperty(meta, f)
+      @test eval(Meta.parse("get_" * string(f)))(DummyModel(meta)) == getproperty(meta, f)
+    end
   end
 end