From e60211c3ccd065dbe45cbd16fbc4c140702af80f Mon Sep 17 00:00:00 2001
From: Michiel Cottaar <MichielCottaar@protonmail.com>
Date: Wed, 18 Sep 2024 13:10:39 +0100
Subject: [PATCH] ENH: allow for integer variables
---
src/build_sequences.jl | 50 +++++++++++++++++++++++++++++++++---------
1 file changed, 40 insertions(+), 10 deletions(-)
diff --git a/src/build_sequences.jl b/src/build_sequences.jl
index 563afaf..9d5ba49 100644
--- a/src/build_sequences.jl
+++ b/src/build_sequences.jl
@@ -73,6 +73,7 @@ function build_sequence(f::Function, scanner::Union{Nothing, Scanner}, model::Tu
sequence = f()
if optimise
jump_model = GLOBAL_MODEL[][1]
+ set_optimizer(jump_model, get_optimiser(jump_model))
if !iszero(num_variables(jump_model))
optimise_with_cost_func!(jump_model, total_cost_func(), n_attempts)
prev_cost_func = nothing
@@ -98,6 +99,37 @@ function number_equality_constraints(model::Model)
sum([num_constraints(model, expr, comp) for (expr, comp) in JuMP.list_of_constraint_types(model) if comp <: MOI.EqualTo])
end
+"""
+ model_has_integers(model)
+
+Returns true if the model contains integer variables.
+"""
+model_has_integers(model::Model) = (VariableRef, MOI.Integer) in list_of_constraint_types(model)
+
+"""
+ get_optimiser(model)
+
+Returns a JuMP solver (https://jump.dev/JuMP.jl/stable/installation/#Supported-solvers) appropriate for this model.
+"""
+function get_optimiser(model::Model)
+ base_ipopt = optimizer_with_attributes(
+ Ipopt.Optimizer,
+ "print_level"=>0,
+ "mu_strategy"=>"adaptive",
+ "max_iter"=>1000,
+ )
+ if model_has_integers(model)
+ optimizer_with_attributes(
+ Juniper.Optimizer,
+ "nl_solver" => base_ipopt,
+ "log_levels" => [],
+ )
+ else
+ return base_ipopt
+ end
+end
+
+
function optimise_with_cost_func!(jump_model::Model, cost_func, n_attempts)
@objective jump_model Min cost_func
min_objective = Inf
@@ -130,7 +162,7 @@ function optimise_with_cost_func!(jump_model::Model, cost_func, n_attempts)
nsuccess += 1
if objective_value(jump_model) < min_objective
min_objective = objective_value(jump_model)
- min_values = copy(backend(jump_model).optimizer.model.inner.x)
+ min_values = copy(get_inner_state(jump_model))
end
break
elseif sub_attempt == 3
@@ -150,20 +182,18 @@ function optimise_with_cost_func!(jump_model::Model, cost_func, n_attempts)
end
error("Optimisation failed to converge. The following errors were raised: $err_string. Example errors for each type are printed above.")
end
- backend(jump_model).optimizer.model.inner.x .= min_values
+ get_inner_state(jump_model) .= min_values
@assert value(cost_func) ≈ min_objective eps(min_objective)
end
-function build_sequence(f::Function, scanner::Union{Nothing, Scanner}=Default_Scanner; optimise=true, n_attempts=20, print_level=0, mu_strategy="adaptive", max_iter=1000, kwargs...)
+get_inner_state(backend::Juniper.JuniperProblem) = backend.solution
+get_inner_state(backend::Ipopt.IpoptProblem) = backend.x
+get_inner_state(model::Model) = get_inner_state(backend(model).optimizer.model.inner)
+
+function build_sequence(f::Function, scanner::Union{Nothing, Scanner}=Default_Scanner; optimise=true, n_attempts=20)
if optimise || GLOBAL_MODEL[] == IGNORE_MODEL
- full_kwargs = Dict(
- :print_level => print_level,
- :mu_strategy => mu_strategy,
- :max_iter => max_iter,
- kwargs...
- )
model = (
- Model(optimizer_with_attributes(Ipopt.Optimizer, [string(k) => v for (k, v) in full_kwargs]...)),
+ Model(),
Tuple{Float64, AbstractJuMPScalar}[]
)
else
--
GitLab