Skip to content

Commit

Permalink
Tweaks to the MOI wrapper (#272)
Browse files Browse the repository at this point in the history
  • Loading branch information
@odow
odow authored Nov 28, 2023
1 parent fbd9bfc commit 4218c32
Show file tree
Hide file tree
Showing 3 changed files with 55 additions and 81 deletions.
3 changes: 2 additions & 1 deletion examples/jump/jump_licensemanager.jl
View file
Original file line number Diff line number Diff line change
Expand Up @@ -20,7 +20,8 @@ lm = KNITRO.LMcontext()
nruns = 10

for i in 1:nruns
model = Model(() -> KNITRO.Optimizer(license_manager=lm, outlev=1))
model = Model(() -> KNITRO.Optimizer(license_manager=lm))
set_attribute(model, "outlev", 1)

initval = [1, 5, 5, 1]

Expand Down
4 changes: 3 additions & 1 deletion examples/moi_nlp1.jl
View file
Original file line number Diff line number Diff line change
Expand Up @@ -64,7 +64,9 @@ end
function example_moi_nlp1(; verbose=true)
options = joinpath(dirname(@__FILE__), "..", "examples", "knitro.opt")
# Define a Knitro solver instance through MOI.
solver = KNITRO.Optimizer(outlev=3, opttol=1e-8)
solver = KNITRO.Optimizer()
MOI.set(solver, MOI.RawOptimizerAttribute("outlev"), 3)
MOI.set(solver, MOI.RawOptimizerAttribute("opttol"), 1e-8)
MOI.set(solver, MOI.Silent(), !verbose)

lb = []
Expand Down
129 changes: 50 additions & 79 deletions src/MOI_wrapper.jl
View file
Original file line number Diff line number Diff line change
Expand Up @@ -58,15 +58,7 @@ function _canonical_vector_affine_reduction(f::MOI.VectorAffineFunction)
return I, J, V
end

# Convert Julia'Inf to KNITRO's Inf.
function _check_value(val::Float64)
if val > KN_INFINITY
return KN_INFINITY
elseif val < -KN_INFINITY
return -KN_INFINITY
end
return val
end
_clamp_inf(x::Float64) = clamp(x, -KN_INFINITY, KN_INFINITY)

mutable struct _VariableInfo
has_lower_bound::Bool
Expand Down Expand Up @@ -140,21 +132,16 @@ mutable struct Optimizer <: MOI.AbstractOptimizer
options::Dict{String,Any}
end

function _set_options(model::Optimizer, options)
for (name, value) in options
MOI.set(model, MOI.RawOptimizerAttribute(string(name)), value)
function Optimizer(; license_manager::Union{LMcontext,Nothing}=nothing, kwargs...)
if !isempty(kwargs)
error("Unsupported keyword arguments passed to `Optimizer`. Set attributes instead")
end
return
end

function Optimizer(; license_manager=nothing, options...)
# Create KNITRO context.
kc = if isa(license_manager, LMcontext)
KN_new_lm(license_manager)
else
KN_new()
end
model = Optimizer(
return Optimizer(
kc,
_VariableInfo[],
0,
Expand All @@ -171,8 +158,6 @@ function Optimizer(; license_manager=nothing, options...)
license_manager,
Dict{String,Any}(),
)
_set_options(model, options)
return model
end

function Base.show(io::IO, model::Optimizer)
Expand Down Expand Up @@ -207,7 +192,9 @@ function MOI.empty!(model::Optimizer)
model.complementarity_cache = _ComplementarityCache()
model.constraint_mapping = Dict()
model.license_manager = model.license_manager
_set_options(model, model.options)
for (name, value) in model.options
MOI.set(model, MOI.RawOptimizerAttribute(name), value)
end
return
end

Expand Down Expand Up @@ -248,12 +235,6 @@ function _throw_if_solved(model::Optimizer, ::Type{MOI.VariableIndex})
return
end

function _number_constraints(model::Optimizer)
p = Ref{Cint}(0)
@_checked KN_get_number_cons(model.inner, p)
return p[]
end

# MOI.SolverName

MOI.get(model::Optimizer, ::MOI.SolverName) = "Knitro"
Expand Down Expand Up @@ -478,7 +459,7 @@ function MOI.add_constraint(
if _is_fixed(model, x)
error("Variable $x is fixed. Cannot also set upper bound.")
end
ub = _check_value(set.upper)
ub = _clamp_inf(set.upper)
model.variable_info[x.value].has_upper_bound = true
@_checked KN_set_var_upbnd(model.inner, _c_column(x), ub)
ci = MOI.ConstraintIndex{MOI.VariableIndex,MOI.LessThan{Float64}}(x.value)
Expand Down Expand Up @@ -513,7 +494,7 @@ function MOI.add_constraint(
if _is_fixed(model, x)
error("Variable $x is fixed. Cannot also set lower bound.")
end
lb = _check_value(set.lower)
lb = _clamp_inf(set.lower)
model.variable_info[x.value].has_lower_bound = true
@_checked KN_set_var_lobnd(model.inner, _c_column(x), lb)
ci = MOI.ConstraintIndex{MOI.VariableIndex,MOI.GreaterThan{Float64}}(x.value)
Expand Down Expand Up @@ -549,8 +530,8 @@ function MOI.add_constraint(
if _is_fixed(model, x)
error("Variable $x is fixed. Cannot also set lower bound.")
end
lb = _check_value(set.lower)
ub = _check_value(set.upper)
lb = _clamp_inf(set.lower)
ub = _clamp_inf(set.upper)
model.variable_info[x.value].has_lower_bound = true
model.variable_info[x.value].has_upper_bound = true
@_checked KN_set_var_lobnd(model.inner, _c_column(x), lb)
Expand Down Expand Up @@ -590,7 +571,7 @@ function MOI.add_constraint(
if _is_fixed(model, x)
error("Variable $x is already fixed.")
end
eqv = _check_value(set.value)
eqv = _clamp_inf(set.value)
model.variable_info[x.value].is_fixed = true
@_checked KN_set_var_fxbnd(model.inner, _c_column(x), eqv)
ci = MOI.ConstraintIndex{MOI.VariableIndex,MOI.EqualTo{Float64}}(x.value)
Expand Down Expand Up @@ -690,18 +671,18 @@ function MOI.add_constraint(
num_cons = p[]
# Add bound to constraint.
if isa(set, MOI.LessThan{Float64})
val = _check_value(set.upper)
val = _clamp_inf(set.upper)
@_checked KN_set_con_upbnd(model.inner, num_cons, val - func.constant)
elseif isa(set, MOI.GreaterThan{Float64})
val = _check_value(set.lower)
val = _clamp_inf(set.lower)
@_checked KN_set_con_lobnd(model.inner, num_cons, val - func.constant)
elseif isa(set, MOI.EqualTo{Float64})
val = _check_value(set.value)
val = _clamp_inf(set.value)
@_checked KN_set_con_eqbnd(model.inner, num_cons, val - func.constant)
else
@assert set isa MOI.Interval{Float64}
# Add upper bound.
lb, ub = _check_value(set.lower), _check_value(set.upper)
lb, ub = _clamp_inf(set.lower), _clamp_inf(set.upper)
@_checked KN_set_con_lobnd(model.inner, num_cons, lb - func.constant)
@_checked KN_set_con_upbnd(model.inner, num_cons, ub - func.constant)
end
Expand Down Expand Up @@ -771,17 +752,17 @@ function MOI.add_constraint(
num_cons = p[]
# Add upper bound.
if isa(set, MOI.LessThan{Float64})
val = _check_value(set.upper)
val = _clamp_inf(set.upper)
@_checked KN_set_con_upbnd(model.inner, num_cons, val - func.constant)
elseif isa(set, MOI.GreaterThan{Float64})
val = _check_value(set.lower)
val = _clamp_inf(set.lower)
@_checked KN_set_con_lobnd(model.inner, num_cons, val - func.constant)
elseif isa(set, MOI.EqualTo{Float64})
val = _check_value(set.value)
val = _clamp_inf(set.value)
@_checked KN_set_con_eqbnd(model.inner, num_cons, val - func.constant)
elseif isa(set, MOI.Interval{Float64})
lb = _check_value(set.lower)
ub = _check_value(set.upper)
lb = _clamp_inf(set.lower)
ub = _clamp_inf(set.upper)
@_checked KN_set_con_lobnd(model.inner, num_cons, lb - func.constant)
@_checked KN_set_con_upbnd(model.inner, num_cons, ub - func.constant)
end
Expand Down Expand Up @@ -1114,42 +1095,17 @@ end

# MOI.optimize!

function _load_complementarity_constraint(model::Optimizer, cache::_ComplementarityCache)
return @_checked KN_set_compcons(
model.inner,
length(cache.cc_types),
cache.cc_types,
cache.index_comps_1,
cache.index_comps_2,
)
end

# Keep loading of NLP constraints apart to load all NLP model all in once
# inside Knitro.
function _load_nlp_constraints(model::Optimizer)
num_nlp_constraints = length(model.nlp_data.constraint_bounds)
if num_nlp_constraints == 0
return
end
num_cons = zeros(Cint, num_nlp_constraints)
@_checked KN_add_cons(model.inner, num_nlp_constraints, num_cons)
for (ib, pair) in enumerate(model.nlp_data.constraint_bounds)
if pair.upper == pair.lower
@_checked KN_set_con_eqbnd(model.inner, num_cons[ib], pair.upper)
else
@_checked KN_set_con_upbnd(model.inner, num_cons[ib], _check_value(pair.upper))
@_checked KN_set_con_lobnd(model.inner, num_cons[ib], _check_value(pair.lower))
end
end
model.nlp_index_cons = num_cons
return
end

function MOI.optimize!(model::Optimizer)
@_checked KN_set_int_param_by_name(model.inner, "datacheck", 0)
@_checked KN_set_int_param_by_name(model.inner, "hessian_no_f", 1)
if _has_complementarity(model.complementarity_cache)
_load_complementarity_constraint(model, model.complementarity_cache)
@_checked KN_set_compcons(
model.inner,
length(model.complementarity_cache.cc_types),
model.complementarity_cache.cc_types,
model.complementarity_cache.index_comps_1,
model.complementarity_cache.index_comps_2,
)
end
if !MOI.is_empty(model.nlp_model) && !model.nlp_loaded
vars = MOI.get(model, MOI.ListOfVariableIndices())
Expand Down Expand Up @@ -1177,9 +1133,20 @@ function MOI.optimize!(model::Optimizer)
end
MOI.initialize(model.nlp_data.evaluator, init_feat)
# Load NLP structure inside Knitro.
offset = _number_constraints(model)
_load_nlp_constraints(model)
num_cons = _number_constraints(model)
p = Ref{Cint}(0)
@_checked KN_get_number_cons(model.inner, p)
offset = p[]
num_nlp_constraints = length(model.nlp_data.constraint_bounds)
if num_nlp_constraints > 0
nlp_rows = model.nlp_index_cons = zeros(Cint, num_nlp_constraints)
@_checked KN_add_cons(model.inner, num_nlp_constraints, nlp_rows)
for (row, pair) in zip(nlp_rows, model.nlp_data.constraint_bounds)
@_checked KN_set_con_upbnd(model.inner, row, _clamp_inf(pair.upper))
@_checked KN_set_con_lobnd(model.inner, row, _clamp_inf(pair.lower))
end
end
@_checked KN_get_number_cons(model.inner, p)
num_cons = p[]
# 1/ Definition of the callbacks
# Objective callback (used both for objective and constraint evaluation).
function eval_f_cb(kc, cb, evalRequest, evalResult, userParams)
Expand Down Expand Up @@ -1508,7 +1475,10 @@ function _check_cons(model, ci, cp)
error("Solve problem before accessing solution.")
elseif cp.result_index > 1
throw(MOI.ResultIndexBoundsError{typeof(cp)}(cp, 1))
elseif !(0 <= ci.value <= _number_constraints(model) - 1)
end
p = Ref{Cint}()
@_checked KN_get_number_cons(model.inner, p)
if !(0 <= ci.value <= p[] - 1)
error("Invalid constraint index ", ci.value)
end
return
Expand Down Expand Up @@ -1609,8 +1579,9 @@ function _reduced_cost(
end
x = MOI.VariableIndex(ci.value)
MOI.throw_if_not_valid(model, x)
offset = _number_constraints(model)
return _sense_dual(model) * _get_dual(model.inner, x.value + offset)
p = Ref{Cint}()
@_checked KN_get_number_cons(model.inner, p)
return _sense_dual(model) * _get_dual(model.inner, x.value + p[])
end

function MOI.get(
Expand Down

0 comments on commit 4218c32

Please sign in to comment.