Skip to content

Commit

Permalink
Tidy ext/NLoptMathOptInterfaceExt.jl (#226)
Browse files Browse the repository at this point in the history
  • Loading branch information
@odow
odow authored Aug 20, 2024
1 parent 66ee350 commit cbfb766
Showing 1 changed file with 69 additions and 104 deletions.
173 changes: 69 additions & 104 deletions ext/NLoptMathOptInterfaceExt.jl
View file
Original file line number Diff line number Diff line change
Expand Up @@ -5,17 +5,15 @@

module NLoptMathOptInterfaceExt

using NLopt

import MathOptInterface

const MOI = MathOptInterface
import MathOptInterface as MOI
import NLopt

function __init__()
# we need to add extension types back to the toplevel module
@static if VERSION >= v"1.9"
setglobal!(NLopt, :Optimizer, Optimizer)
end
return
end

mutable struct _ConstraintInfo{F,S}
Expand All @@ -29,7 +27,7 @@ end
Create a new Optimizer object.
"""
mutable struct Optimizer <: MOI.AbstractOptimizer
inner::Union{Opt,Nothing}
inner::Union{NLopt.Opt,Nothing}
# Problem data.
variables::MOI.Utilities.VariablesContainer{Float64}
starting_values::Vector{Union{Nothing,Float64}}
Expand Down Expand Up @@ -80,7 +78,7 @@ mutable struct Optimizer <: MOI.AbstractOptimizer
nothing,
MOI.Utilities.VariablesContainer{Float64}(),
Union{Nothing,Float64}[],
empty_nlp_data(),
_empty_nlp_data(),
nothing,
nothing,
_ConstraintInfo{
Expand All @@ -100,7 +98,7 @@ mutable struct Optimizer <: MOI.AbstractOptimizer
MOI.EqualTo{Float64},
}[],
false,
copy(DEFAULT_OPTIONS),
copy(_DEFAULT_OPTIONS),
NaN,
Float64[],
Float64[],
Expand All @@ -113,38 +111,38 @@ mutable struct Optimizer <: MOI.AbstractOptimizer
end
end

struct EmptyNLPEvaluator <: MOI.AbstractNLPEvaluator end
struct _EmptyNLPEvaluator <: MOI.AbstractNLPEvaluator end

MOI.features_available(::EmptyNLPEvaluator) = [:Grad, :Jac, :Hess]
MOI.features_available(::_EmptyNLPEvaluator) = [:Grad, :Jac, :Hess]

MOI.initialize(::EmptyNLPEvaluator, features) = nothing
MOI.initialize(::_EmptyNLPEvaluator, features) = nothing

MOI.eval_objective(::EmptyNLPEvaluator, x) = NaN
MOI.eval_objective(::_EmptyNLPEvaluator, x) = NaN

function MOI.eval_constraint(::EmptyNLPEvaluator, g, x)
function MOI.eval_constraint(::_EmptyNLPEvaluator, g, x)
@assert length(g) == 0
return
end
function MOI.eval_objective_gradient(::EmptyNLPEvaluator, g, x)
function MOI.eval_objective_gradient(::_EmptyNLPEvaluator, g, x)
fill!(g, 0.0)
return
end

MOI.jacobian_structure(::EmptyNLPEvaluator) = Tuple{Int64,Int64}[]
MOI.jacobian_structure(::_EmptyNLPEvaluator) = Tuple{Int64,Int64}[]

function MOI.eval_constraint_jacobian(::EmptyNLPEvaluator, J, x)
function MOI.eval_constraint_jacobian(::_EmptyNLPEvaluator, J, x)
return
end

empty_nlp_data() = MOI.NLPBlockData([], EmptyNLPEvaluator(), false)
_empty_nlp_data() = MOI.NLPBlockData([], _EmptyNLPEvaluator(), false)

# empty! and is_empty

function MOI.empty!(model::Optimizer)
model.inner = nothing
MOI.empty!(model.variables)
empty!(model.starting_values)
model.nlp_data = empty_nlp_data()
model.nlp_data = _empty_nlp_data()
model.sense = nothing
model.objective = nothing
empty!(model.linear_le_constraints)
Expand All @@ -158,7 +156,7 @@ end
function MOI.is_empty(model::Optimizer)
return MOI.is_empty(model.variables) &&
isempty(model.starting_values) &&
model.nlp_data.evaluator isa EmptyNLPEvaluator &&
model.nlp_data.evaluator isa _EmptyNLPEvaluator &&
model.sense == nothing &&
isempty(model.linear_le_constraints) &&
isempty(model.linear_eq_constraints) &&
Expand Down Expand Up @@ -323,7 +321,7 @@ end

# MOI.RawOptimizerAttribute

const DEFAULT_OPTIONS = Dict{String,Any}(
const _DEFAULT_OPTIONS = Dict{String,Any}(
"algorithm" => :none,
"stopval" => NaN,
"ftol_rel" => 1e-7,
Expand All @@ -341,7 +339,7 @@ const DEFAULT_OPTIONS = Dict{String,Any}(
)

function MOI.supports(::Optimizer, p::MOI.RawOptimizerAttribute)
return p.name in DEFAULT_OPTIONS
return p.name in _DEFAULT_OPTIONS
end

function MOI.set(model::Optimizer, p::MOI.RawOptimizerAttribute, value)
Expand Down Expand Up @@ -562,13 +560,12 @@ function MOI.add_constraint(
return MOI.ConstraintIndex{typeof(func),typeof(set)}(length(constraints))
end

function starting_value(optimizer::Optimizer, i)
function _starting_value(optimizer::Optimizer, i)
if optimizer.starting_values[i] !== nothing
return optimizer.starting_values[i]
else
v = optimizer.variables
return min(max(0.0, v.lower[i]), v.upper[i])
end
v = optimizer.variables
return min(max(0.0, v.lower[i]), v.upper[i])
end

function MOI.supports(
Expand Down Expand Up @@ -643,7 +640,7 @@ function MOI.set(
return
end

function eval_objective(model::Optimizer, x)
function _eval_objective(model::Optimizer, x)
# The order of the conditions is important. NLP objectives override regular
# objectives.
if model.nlp_data.has_objective
Expand All @@ -656,21 +653,21 @@ function eval_objective(model::Optimizer, x)
end
end

function fill_gradient!(grad, x, var::MOI.VariableIndex)
function _fill_gradient!(grad, x, var::MOI.VariableIndex)
fill!(grad, 0.0)
grad[var.value] = 1.0
return
end

function fill_gradient!(grad, x, aff::MOI.ScalarAffineFunction{Float64})
function _fill_gradient!(grad, x, aff::MOI.ScalarAffineFunction{Float64})
fill!(grad, 0.0)
for term in aff.terms
grad[term.variable.value] += term.coefficient
end
return
end

function fill_gradient!(grad, x, quad::MOI.ScalarQuadraticFunction{Float64})
function _fill_gradient!(grad, x, quad::MOI.ScalarQuadraticFunction{Float64})
fill!(grad, 0.0)
for term in quad.affine_terms
grad[term.variable.value] += term.coefficient
Expand All @@ -689,18 +686,18 @@ function fill_gradient!(grad, x, quad::MOI.ScalarQuadraticFunction{Float64})
return
end

function eval_objective_gradient(model::Optimizer, grad, x)
function _eval_objective_gradient(model::Optimizer, grad, x)
if model.nlp_data.has_objective
MOI.eval_objective_gradient(model.nlp_data.evaluator, grad, x)
elseif model.objective !== nothing
fill_gradient!(grad, x, model.objective)
_fill_gradient!(grad, x, model.objective)
else
fill!(grad, 0.0)
end
return
end

function eval_constraint(model::Optimizer, g, x)
function _eval_constraint(model::Optimizer, g, x)
row = 1
for info in model.linear_le_constraints
g[row] = eval_function(info.func, x)
Expand All @@ -723,7 +720,7 @@ function eval_constraint(model::Optimizer, g, x)
return
end

function fill_constraint_jacobian!(
function _fill_constraint_jacobian!(
values,
start_offset,
x,
Expand All @@ -736,7 +733,7 @@ function fill_constraint_jacobian!(
return num_coefficients
end

function fill_constraint_jacobian!(
function _fill_constraint_jacobian!(
values,
start_offset,
x,
Expand Down Expand Up @@ -767,96 +764,64 @@ function fill_constraint_jacobian!(
return num_affine_coefficients + num_quadratic_coefficients
end

# Refers to local variables in eval_constraint_jacobian() below.
macro fill_constraint_jacobian(array_name)
esc_offset = esc(:offset)
quote
for info in $(esc(array_name))
$esc_offset += fill_constraint_jacobian!(
$(esc(:values)),
$esc_offset,
$(esc(:x)),
info.func,
)
end
end
end

function eval_constraint_jacobian(model::Optimizer, values, x)
function _eval_constraint_jacobian(model::Optimizer, values, x)
offset = 0
@fill_constraint_jacobian model.linear_le_constraints
@fill_constraint_jacobian model.linear_eq_constraints
@fill_constraint_jacobian model.quadratic_le_constraints
@fill_constraint_jacobian model.quadratic_eq_constraints
nlp_values = view(values, 1+offset:length(values))
MOI.eval_constraint_jacobian(model.nlp_data.evaluator, nlp_values, x)
return
end

function constraint_bounds(model::Optimizer)
constraint_lb = Float64[]
constraint_ub = Float64[]
for info in model.linear_le_constraints
push!(constraint_lb, -Inf)
push!(constraint_ub, info.set.upper)
for info_1 in model.linear_le_constraints
offset += _fill_constraint_jacobian!(values, offset, x, info_1.func)
end
for info in model.linear_eq_constraints
push!(constraint_lb, info.set.value)
push!(constraint_ub, info.set.value)
for info_2 in model.linear_eq_constraints
offset += _fill_constraint_jacobian!(values, offset, x, info_2.func)
end
for info in model.quadratic_le_constraints
push!(constraint_lb, -Inf)
push!(constraint_ub, info.set.upper)
for info_3 in model.quadratic_le_constraints
offset += _fill_constraint_jacobian!(values, offset, x, info_3.func)
end
for info in model.quadratic_eq_constraints
push!(constraint_lb, info.set.value)
push!(constraint_ub, info.set.value)
end
for bound in model.nlp_data.constraint_bounds
push!(constraint_lb, bound.lower)
push!(constraint_ub, bound.upper)
for info_4 in model.quadratic_eq_constraints
offset += _fill_constraint_jacobian!(values, offset, x, info_4.func)
end
return constraint_lb, constraint_ub
nlp_values = view(values, 1+offset:length(values))
MOI.eval_constraint_jacobian(model.nlp_data.evaluator, nlp_values, x)
return
end

function MOI.optimize!(model::Optimizer)
num_variables = length(model.starting_values)
model.inner = Opt(model.options["algorithm"], num_variables)
model.inner = NLopt.Opt(model.options["algorithm"], num_variables)
local_optimizer = model.options["local_optimizer"]
if local_optimizer !== nothing
if local_optimizer isa Symbol
local_optimizer = Opt(local_optimizer, num_variables)
local_optimizer = NLopt.Opt(local_optimizer, num_variables)
else
local_optimizer = Opt(local_optimizer.algorithm, num_variables)
local_optimizer =
NLopt.Opt(local_optimizer.algorithm, num_variables)
end
local_optimizer!(model.inner, local_optimizer)
NLopt.local_optimizer!(model.inner, local_optimizer)
end
# load parameters
stopval!(model.inner, model.options["stopval"])
NLopt.stopval!(model.inner, model.options["stopval"])
if !isnan(model.options["ftol_rel"])
ftol_rel!(model.inner, model.options["ftol_rel"])
NLopt.ftol_rel!(model.inner, model.options["ftol_rel"])
end
if !isnan(model.options["ftol_abs"])
ftol_abs!(model.inner, model.options["ftol_abs"])
NLopt.ftol_abs!(model.inner, model.options["ftol_abs"])
end
if !isnan(model.options["xtol_rel"])
xtol_rel!(model.inner, model.options["xtol_rel"])
NLopt.xtol_rel!(model.inner, model.options["xtol_rel"])
end
if model.options["xtol_abs"] != nothing
xtol_abs!(model.inner, model.options["xtol_abs"])
NLopt.xtol_abs!(model.inner, model.options["xtol_abs"])
end
maxeval!(model.inner, model.options["maxeval"])
maxtime!(model.inner, model.options["maxtime"])
NLopt.maxeval!(model.inner, model.options["maxeval"])
NLopt.maxtime!(model.inner, model.options["maxtime"])
if model.options["initial_step"] != nothing
initial_step!(model.inner, model.options["initial_step"])
NLopt.initial_step!(model.inner, model.options["initial_step"])
end
population!(model.inner, model.options["population"])
NLopt.population!(model.inner, model.options["population"])
if isa(model.options["seed"], Integer)
NLopt.srand(model.options["seed"])
end
vector_storage!(model.inner, model.options["vector_storage"])
lower_bounds!(model.inner, model.variables.lower)
upper_bounds!(model.inner, model.variables.upper)
NLopt.vector_storage!(model.inner, model.options["vector_storage"])
NLopt.lower_bounds!(model.inner, model.variables.lower)
NLopt.upper_bounds!(model.inner, model.variables.upper)
nleqidx = findall(
bound -> bound.lower == bound.upper,
model.nlp_data.constraint_bounds,
Expand Down Expand Up @@ -898,18 +863,18 @@ function MOI.optimize!(model::Optimizer)
fill!(grad, 0.0)
return 0.0
end
min_objective!(model.inner, z)
NLopt.min_objective!(model.inner, z)
else
function f(x::Vector, grad::Vector)
if length(grad) > 0
eval_objective_gradient(model, grad, x)
_eval_objective_gradient(model, grad, x)
end
return eval_objective(model, x)
return _eval_objective(model, x)
end
if model.sense == MOI.MIN_SENSE
min_objective!(model.inner, f)
NLopt.min_objective!(model.inner, f)
else
max_objective!(model.inner, f)
NLopt.max_objective!(model.inner, f)
end
end
Jac_IJ =
Expand Down Expand Up @@ -959,7 +924,7 @@ function MOI.optimize!(model::Optimizer)
)
end
g_eq(zeros(num_eq), zeros(num_variables), zeros(num_variables, num_eq))
equality_constraint!(
NLopt.equality_constraint!(
model.inner,
g_eq,
fill(model.options["constrtol_abs"], num_eq),
Expand Down Expand Up @@ -1029,7 +994,7 @@ function MOI.optimize!(model::Optimizer)
zeros(num_variables),
zeros(num_variables, num_ineq),
)
inequality_constraint!(
NLopt.inequality_constraint!(
model.inner,
g_ineq,
fill(model.options["constrtol_abs"], num_ineq),
Expand All @@ -1038,11 +1003,11 @@ function MOI.optimize!(model::Optimizer)
# If nothing is provided, the default starting value is 0.0.
model.solution = zeros(num_variables)
for i in eachindex(model.starting_values)
model.solution[i] = starting_value(model, i)
model.solution[i] = _starting_value(model, i)
end
start_time = time()
model.objective_value, _, model.status =
optimize!(model.inner, model.solution)
NLopt.optimize!(model.inner, model.solution)
model.constraint_primal_linear_le =
_fill_primal(model.solution, model.linear_le_constraints)
model.constraint_primal_linear_eq =
Expand Down

0 comments on commit cbfb766

Please sign in to comment.