NLopt.jl is a wrapper for the NLopt library for nonlinear optimization.
NLopt provides a common interface for many different optimization algorithms, including:
- Both global and local optimization
- Algorithms using function values only (derivative-free) and also algorithms exploiting user-supplied gradients.
- Algorithms for unconstrained optimization, bound-constrained optimization, and general nonlinear inequality/equality constraints.
NLopt.jl
is licensed under the MIT License.
The underlying solver, stevengj/nlopt, is licensed under the LGPL v3.0 license.
Install NLopt.jl
using the Julia package manager:
import Pkg
Pkg.add("NLopt")
In addition to installing the NLopt.jl
package, this will also download and
install the NLopt binaries. You do not need to install NLopt separately.
The following example code solves the nonlinearly constrained minimization problem from the NLopt Tutorial.
using NLopt
function my_objective_fn(x::Vector, grad::Vector)
if length(grad) > 0
grad[1] = 0
grad[2] = 0.5 / sqrt(x[2])
end
return sqrt(x[2])
end
function my_constraint_fn(x::Vector, grad::Vector, a, b)
if length(grad) > 0
grad[1] = 3 * a * (a * x[1] + b)^2
grad[2] = -1
end
return (a * x[1] + b)^3 - x[2]
end
opt = NLopt.Opt(:LD_MMA, 2)
NLopt.lower_bounds!(opt, [-Inf, 0.0])
NLopt.xtol_rel!(opt, 1e-4)
NLopt.min_objective!(opt, my_objective_fn)
NLopt.inequality_constraint!(opt, (x, g) -> my_constraint_fn(x, g, 2, 0), 1e-8)
NLopt.inequality_constraint!(opt, (x, g) -> my_constraint_fn(x, g, -1, 1), 1e-8)
min_f, min_x, ret = NLopt.optimize(opt, [1.234, 5.678])
num_evals = NLopt.numevals(opt)
println(
"""
objective value : $min_f
solution : $min_x
solution status : $ret
# function evaluation : $num_evals
"""
)
The output is:
objective value : 0.5443310477213124
solution : [0.3333333342139688, 0.29629628951338166]
solution status : XTOL_REACHED
# function evaluation : 11
NLopt implements the MathOptInterface interface for nonlinear optimization, which means that it can be used interchangeably with other optimization packages from modeling packages like JuMP. Note that NLopt does not exploit sparsity of Jacobians.
You can use NLopt with JuMP as follows:
using JuMP, NLopt
model = Model(NLopt.Optimizer)
set_attribute(model, "algorithm", :LD_MMA)
set_attribute(model, "xtol_rel", 1e-4)
set_attribute(model, "constrtol_abs", 1e-8)
@variable(model, x[1:2])
set_lower_bound(x[2], 0.0)
set_start_value.(x, [1.234, 5.678])
@NLobjective(model, Min, sqrt(x[2]))
@NLconstraint(model, (2 * x[1] + 0)^3 - x[2] <= 0)
@NLconstraint(model, (-1 * x[1] + 1)^3 - x[2] <= 0)
optimize!(model)
min_f, min_x, ret = objective_value(model), value.(x), raw_status(model)
println(
"""
objective value : $min_f
solution : $min_x
solution status : $ret
"""
)
The output is:
objective value : 0.5443310477213124
solution : [0.3333333342139688, 0.29629628951338166]
solution status : XTOL_REACHED
The algorithm
attribute is required. The value must be one of the supported
NLopt algorithms.
Other parameters include stopval
, ftol_rel
, ftol_abs
, xtol_rel
,
xtol_abs
, constrtol_abs
, maxeval
, maxtime
, initial_step
, population
,
seed
, and vector_storage
.
The algorithm
parameter is required, and all others are optional. The
meaning and acceptable values of all parameters, except constrtol_abs
, match
the descriptions below from the specialized NLopt API.
The constrtol_abs
parameter is an absolute feasibility tolerance applied to
all constraints.
Some algorithms in NLopt require derivatives, which you must manually provide
in the if length(grad) > 0
branch of your objective and constraint functions.
To stay simple and lightweight, NLopt does not provide ways to automatically compute derivatives. If you do not have analytic expressions for the derivatives, use a package such as ForwardDiff.jl to compute automatic derivatives.
Here is an example of how to wrap a function f(x::Vector)
using ForwardDiff so
that it is compatible with NLopt:
using NLopt
import ForwardDiff
function autodiff(f::Function)
function nlopt_fn(x::Vector, grad::Vector)
if length(grad) > 0
# Use ForwardDiff to compute the gradient. Replace with your
# favorite Julia automatic differentiation package.
ForwardDiff.gradient!(grad, f, x)
end
return f(x)
end
end
# These functions do not implement `grad`:
my_objective_fn(x::Vector) = sqrt(x[2]);
my_constraint_fn(x::Vector, a, b) = (a * x[1] + b)^3 - x[2];
opt = NLopt.Opt(:LD_MMA, 2)
NLopt.lower_bounds!(opt, [-Inf, 0.0])
NLopt.xtol_rel!(opt, 1e-4)
# But we wrap them in autodiff before passing to NLopt:
NLopt.min_objective!(opt, autodiff(my_objective_fn))
NLopt.inequality_constraint!(opt, autodiff(x -> my_constraint_fn(x, 2, 0)), 1e-8)
NLopt.inequality_constraint!(opt, autodiff(x -> my_constraint_fn(x, -1, 1)), 1e-8)
min_f, min_x, ret = NLopt.optimize(opt, [1.234, 5.678])
# (0.5443310477213124, [0.3333333342139688, 0.29629628951338166], :XTOL_REACHED)
The main purpose of this section is to document the syntax and unique features of the Julia interface. For more detail on the underlying features, please refer to the C documentation in the NLopt Reference.
To use NLopt in Julia, your Julia program should include the line:
using NLopt
which imports the NLopt module and its symbols. Alternatively, you can use
import NLopt
if you want to keep all the NLopt symbols in their own namespace.
You would then prefix all functions below with NLopt.
, for example NLopt.Opt
and so
on.
The NLopt API revolves around an object of type Opt
.
The object should normally be created via the constructor:
opt = Opt(algorithm::Symbol, n::Int)
given an algorithm (see NLopt Algorithms
for possible values) and the dimensionality of the problem (n
, the number of
optimization parameters).
Whereas in C the algorithms are specified by nlopt_algorithm
constants of the
form like NLOPT_LD_MMA
, the Julia algorithm
values are symbols of the form
:LD_MMA
with the NLOPT_
prefix replaced by :
to create a Julia symbol.
There is also a copy(opt::Opt)
function to make a copy of a given object
(equivalent to nlopt_copy
in the C API).
If there is an error in these functions, an exception is thrown.
The algorithm and dimension parameters of the object are immutable (cannot be changed without constructing a new object). Query them using:
ndims(opt::Opt)
algorithm(opt::Opt)
Get a string description of the algorithm via:
algorithm_name(opt::Opt)
The objective function is specified by calling one of:
min_objective!(opt::Opt, f::Function)
max_objective!(opt::Opt, f::Function)
depending on whether one wishes to minimize or maximize the objective function
f
, respectively.
The function f
must be of the form:
function f(x::Vector{Float64}, grad::Vector{Float64})
if length(grad) > 0
...set grad to gradient, in-place...
end
return ...value of f(x)...
end
The return value must be the value of the function at the point x
, where x
is a Vector{Float64}
array of length n
of the optimization parameters.
In addition, if the argument grad
is not empty (that is, length(grad) > 0
),
then grad
is a Vector{Float64}
array of length n
which should (upon
return) be set to the gradient of the function with respect to the optimization
parameters at x
.
Not all of the optimization algorithms (below) use the gradient information: for
algorithms listed as "derivative-free," the grad
argument will always be empty
and need never be computed. For algorithms that do use gradient information,
grad
may still be empty for some calls.
Note that grad
must be modified in-place by your function f
. Generally,
this means using indexing operations grad[...] = ...
to overwrite the contents
of grad
. For example grad = 2x
will not work, because it points grad
to
a new array 2x
rather than overwriting the old contents; instead, use an
explicit loop or use grad[:] = 2x
.
Add bound constraints with:
lower_bounds!(opt::Opt, lb::Union{AbstractVector,Real})
upper_bounds!(opt::Opt, ub::Union{AbstractVector,Real})
where lb
and ub
are real arrays of length n
(the same as the dimension
passed to the Opt
constructor).
For convenience, you can instead use a single scalar for lb
or ub
in order
to set the lower/upper bounds for all optimization parameters to a single
constant.
To retrieve the values of the lower or upper bounds, use:
lower_bounds(opt::Opt)
upper_bounds(opt::Opt)
both of which return Vector{Float64}
arrays.
To specify an unbounded dimension, you can use Inf
or -Inf
.
Specify nonlinear inequality and equality constraints by the functions:
inequality_constraint!(opt::Opt, f::Function, tol::Real = 0.0)
equality_constraint!(opt::Opt, f::Function, tol::Real = 0.0)
where the arguments f
have the same form as the objective function above.
The optional tol
arguments specify a tolerance (which defaults to zero) that
is used to judge feasibility for the purposes of stopping the optimization.
Each call to these function adds a new constraint to the set of constraints, rather than replacing the constraints.
Remove all of the inequality and equality constraints from a given problem with:
remove_constraints!(opt::Opt)
Specify vector-valued nonlinear inequality and equality constraints by the functions:
inequality_constraint!(opt::Opt, f::Function, tol::AbstractVector)
equality_constraint!(opt::Opt, f::Function, tol::AbstractVector)
where tol
is an array of the tolerances in each constraint dimension; the
dimensionality m
of the constraint is determined by length(tol)
.
The constraint function f
must be of the form:
function f(result::Vector{Float64}, x::Vector{Float64}, grad::Matrix{Float64})
if length(grad) > 0
...set grad to gradient, in-place...
end
result[1] = ...value of c1(x)...
result[2] = ...value of c2(x)...
return
where result
is a Vector{Float64}
array whose length equals the
dimensionality m
of the constraint (same as the length of tol
above), which
upon return, should be set in-place to the constraint results at the point x
.
Any return value of the function is ignored.
In addition, if the argument grad
is not empty (that is, length(grad) > 0
),
then grad
is a matrix of size n
×m
which should (upon return) be
set in-place (see above) to the gradient of the function with respect to the
optimization parameters at x
. That is, grad[j,i]
should upon return contain
the partial derivative ∂fi
/∂xj
.
Not all of the optimization algorithms (below) use the gradient information: for
algorithms listed as "derivative-free," the grad
argument will always be empty
and need never be computed. For algorithms that do use gradient information,
grad
may still be empty for some calls.
You can add multiple vector-valued constraints and/or scalar constraints in the same problem.
As explained in the C API Reference and the Introduction, you have multiple options for different stopping criteria that you can specify. (Unspecified stopping criteria are disabled; that is, they have innocuous defaults.)
For each stopping criteria, there are two functions that you can use to get and set the value of the stopping criterion.
stopval(opt::Opt) # return the current value of `stopval`
stopval!(opt::Opt, value) # set stopval to `value`
Stop when an objective value of at least stopval
is found. (Defaults to -Inf
.)
ftol_rel(opt::Opt)
ftol_rel!(opt::Opt, value)
Relative tolerance on function value. (Defaults to 0
.)
ftol_abs(opt::Opt)
ftol_abs!(opt::Opt, value)
Absolute tolerance on function value. (Defaults to 0
.)
xtol_rel(opt::Opt)
xtol_rel!(opt::Opt, value)
Relative tolerances on the optimization parameters. (Defaults to 0
.)
xtol_abs(opt::Opt)
xtol_abs!(opt::Opt, value)
Absolute tolerances on the optimization parameters. (Defaults to 0
.)
In the case of xtol_abs
, you can either set it to a scalar (to use the same
tolerance for all inputs) or a vector of length n
(the dimension specified in
the Opt
constructor) to use a different tolerance for each parameter.
maxeval(opt::Opt)
maxeval!(opt::Opt, value)
Stop when the number of function evaluations exceeds mev
. (0 or negative for
no limit, which is the default.)
maxtime(opt::Opt)
maxtime!(opt::Opt, value)
Stop when the optimization time (in seconds) exceeds t
. (0 or negative for no
limit, which is the default.)
In certain cases, the caller may wish to force the optimization to halt, for
some reason unknown to NLopt. For example, if the user presses Ctrl-C, or there
is an error of some sort in the objective function. You can do this by throwing
any exception inside your objective/constraint functions: the optimization will
be halted gracefully, and the same exception will be thrown to the caller. The
Julia equivalent of nlopt_forced_stop
from the C API is to throw a ForcedStop
exception.
Once all of the desired optimization parameters have been specified in a given
object opt::Opt
, you can perform the optimization by calling:
optf, optx, ret = optimize(opt::Opt, x::AbstractVector)
On input, x
is an array of length n
(the dimension of the problem from the
Opt
constructor) giving an initial guess for the optimization parameters. The
return value optx
is a array containing the optimized values of the
optimization parameters. optf
contains the optimized value of the objective
function, and ret
contains a symbol indicating the NLopt return code (below).
Alternatively:
optf, optx, ret = optimize!(opt::Opt, x::Vector{Float64})
is the same but modifies x
in-place (as well as returning optx = x
).
The possible return values are the same as the return values in the C API,
except that the NLOPT_
prefix is replaced with :
. That is, the return
values are like :SUCCESS
instead NLOPT_SUCCESS
.
Some of the algorithms, especially MLSL
and AUGLAG
, use a different
optimization algorithm as a subroutine, typically for local optimization. You
can change the local search algorithm and its tolerances by setting:
local_optimizer!(opt::Opt, local_opt::Opt)
Here, local_opt
is another Opt
object whose parameters are used to determine
the local search algorithm, its stopping criteria, and other algorithm
parameters. (However, the objective function, bounds, and nonlinear-constraint
parameters of local_opt
are ignored.) The dimension n
of local_opt
must
match that of opt
.
This makes a copy of the local_opt
object, so you can freely change your
original local_opt
afterwards without affecting opt
.
Just as in the C API, you can set the initial step sizes for derivative-free optimization algorithms with:
initial_step!(opt::Opt, dx::Vector)
Here, dx
is an array of the (nonzero) initial steps for each dimension, or a
single number if you wish to use the same initial steps for all dimensions.
initial_step(opt::Opt, x::AbstractVector)
returns the initial step that will
be used for a starting guess of x
in optimize(opt, x)
.
Just as in the C API, you can get and set the initial population for stochastic optimization with:
population(opt::Opt)
population!(opt::Opt, value)
A population
of zero, the default, implies that the heuristic default will be
used as decided upon by individual algorithms.
For stochastic optimization algorithms, NLopt uses pseudorandom numbers generated by the Mersenne Twister algorithm, based on code from Makoto Matsumoto.
By default, the seed for the random numbers is generated from the system time, so that you will get a different sequence of pseudorandom numbers each time you run your program. If you want to use a "deterministic" sequence of pseudorandom numbers, that is, the same sequence from run to run, you can set the seed by calling:
NLopt.srand(seed::Integer)
To reset the seed based on the system time, you can call NLopt.srand_time()
.
Normally, you don't need to call this as it is called automatically. However, it
might be useful if you want to "re-randomize" the pseudorandom numbers after
calling nlopt.srand
to set a deterministic seed.
Just as in the C API, you can get and set the number M of stored vectors for limited-memory quasi-Newton algorithms, via integer-valued property
vector_storage(opt::Opt)
vector_storage!(opt::Opt, value)
The default is 0
, in which case NLopt uses a heuristic nonzero value as
determined by individual algorithms.
The version number of NLopt is given by the global variable:
NLOPT_VERSION::VersionNumber
where VersionNumber
is a built-in Julia type from the Julia standard library.
The underlying NLopt library is threadsafe; however, re-using the same Opt
object across multiple threads is not.
As an example, instead of:
using NLopt
opt = Opt(:LD_MMA, 2)
# Define problem
solutions = Vector{Any}(undef, 10)
Threads.@threads for i in 1:10
# Not thread-safe because `opt` is re-used
solutions[i] = optimize(opt, rand(2))
end
Do instead:
solutions = Vector{Any}(undef, 10)
Threads.@threads for i in 1:10
# Thread-safe because a new `opt` is created for each thread
opt = Opt(:LD_MMA, 2)
# Define problem
solutions[i] = optimize(opt, rand(2))
end
This module was initially written by Steven G. Johnson, with subsequent contributions by several other authors (see the git history).