Skip to content

Test objective #7

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 2 commits into
base: master
Choose a base branch
from
Open

Test objective #7

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
1c40650
add/fix support for linear and quadratic objective; should work with …
gnowzil Feb 3, 2021
7faaa52
remove constraint updates for now
gnowzil Feb 3, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
183 changes: 109 additions & 74 deletions src/MOIWrapper.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -53,8 +53,8 @@ MOI.supports(::Optimizer, ::MOI.NLPBlock) = true
MOI.supports(::Optimizer, ::MOI.ObjectiveFunction{MOI.SingleVariable}) = true
MOI.supports(::Optimizer, ::MOI.ObjectiveFunction{MOI.ScalarAffineFunction{Float64}}) = true
MOI.supports(::Optimizer, ::MOI.ObjectiveFunction{MOI.ScalarQuadraticFunction{Float64}}) = true
MOI.supports(::Optimizer, ::MOI.ObjectiveSense) = true
MOI.supports(::Optimizer, ::MOI.VariablePrimalStart,::Type{MOI.VariableIndex}) = true #?
MOI.supports(::Optimizer, ::MOI.ObjectiveSense) = true
MOI.supports(::Optimizer, ::MOI.RawParameter) = true

MOI.supports_constraint(::Optimizer, ::Type{MOI.SingleVariable}, ::Type{MOI.LessThan{Float64}}) = true
Expand Down Expand Up @@ -86,6 +86,36 @@ function MOI.set(model::Optimizer, ::MOI.ObjectiveFunction,
return
end

function eval_function(v::MOI.SingleVariable, x)
vi = v.variable
return x[vi.value]
end

function eval_function(fun::MOI.ScalarAffineFunction, x)
obj_value = fun.constant
for vi in fun.terms
obj_value += x[vi.variable_index.value] * vi.coefficient
end
return obj_value
end

function eval_function(fun::MOI.ScalarQuadraticFunction, x)
obj_value = fun.constant
for vi in fun.affine_terms
obj_value += x[vi.variable_index.value] * vi.coefficient
end
for vi in fun.quadratic_terms
irow = vi.variable_index_1
jcol = vi.variable_index_2
value = vi.coefficient * x[irow.value] * x[jcol.value]
if irow == jcol
value = 0.5 * value
end
obj_value += value
end
return obj_value
end

# Variables
function MOI.set(model::Optimizer, ::MOI.VariablePrimalStart, vi::MOI.VariableIndex, value::Float64)
model.variables[vi.value].x = value
Expand All @@ -101,6 +131,7 @@ function MOI.add_variables(model::Optimizer, n::Int)
return [MOI.add_variable(model) for i in 1:n]
end

# Constraints
function MOI.add_constraint(model::Optimizer, v::MOI.SingleVariable, lt::MOI.LessThan{Float64})
# Upper bound on x
vi = v.variable
Expand Down Expand Up @@ -171,6 +202,10 @@ end
# Solve
function MOI.optimize!(model::Optimizer)

# Initialize SNOPT
model.workspace = SNOPT7.initialize("snopt.out","screen")

# Problem setup
n = length(model.variables)
nncon = model.nonlin !== nothing ? length(model.nonlin.constraint_bounds) : 0
nlcon = length(model.linear_constraints)
Expand Down Expand Up @@ -240,96 +275,96 @@ function MOI.optimize!(model::Optimizer)



# Options
for (name,value) in model.options
# Replace underscore with space
keyword = replace(String(name), "_" => " ")
SNOPT7.setOption!(model.workspace, "$keyword $value")
end
SNOPT7.setOption!(model.workspace, "Summary frequency 1")
SNOPT7.setOption!(model.workspace, "Print frequency 1")

if model.nonlin !== nothing
# Nonlinear constraints/objective
nnobj = n

# Callbacks
function eval_obj(mode_::Ptr{Cint}, n_::Ptr{Cint}, x_::Ptr{Float64},
f_::Ptr{Float64}, g_::Ptr{Float64}, status::Ptr{Cint})
mode = unsafe_load(mode_)
n = unsafe_load(n_)

x = unsafe_wrap(Array, x_, Int(n))
# Nonlinear constraints/objective
nnobj = n

if mode == 0 || mode == 2
if model.nonlin.has_objective
obj = MOI.eval_objective(model.nonlin.evaluator, x)
elseif model.objective !== nothing
obj = eval_function(model.objective, x)
else
obj = 0.0
end
unsafe_store!(f_,obj)
end
# Callbacks
function eval_obj(mode_::Ptr{Cint}, n_::Ptr{Cint}, x_::Ptr{Float64},
f_::Ptr{Float64}, g_::Ptr{Float64}, status::Ptr{Cint})
mode = unsafe_load(mode_)
n = unsafe_load(n_)

if mode == 1 || mode == 2
g = unsafe_wrap(Array,g_, Int(n))
if model.nonlin.has_objective
MOI.eval_objective_gradient(model.nonlin.evaluator, g, x)
end
end
return
end
x = unsafe_wrap(Array, x_, Int(n))

function eval_con(mode_::Ptr{Cint},
nncon_::Ptr{Cint}, nnjac_::Ptr{Cint}, negcon_::Ptr{Cint},
x_::Ptr{Float64}, c_::Ptr{Float64}, J_::Ptr{Float64},
status::Ptr{Cint})
mode = unsafe_load(mode_)
nncon = unsafe_load(nncon_)
nnjac = unsafe_load(nnjac_)
negcon = unsafe_load(negcon_)

x = unsafe_wrap(Array, x_, Int(nnjac))
if mode == 0 || mode == 2
c = unsafe_wrap(Array, c_, Int(nncon))
MOI.eval_constraint(model.nonlin.evaluator, c, x)
if mode == 0 || mode == 2
if model.nonlin.has_objective
obj = MOI.eval_objective(model.nonlin.evaluator, x)
elseif model.objective !== nothing
obj = eval_function(model.objective, x)
else
obj = 0.0
end
unsafe_store!(f_,obj)
end

if mode == 1 || mode == 2
J = unsafe_wrap(Array,J_, Int(negcon))

# MOI returns Jacobian in order specified by j_rows, j_cols
# SNOPT needs it in sparse-by-col format
JJ = zeros(negcon)
MOI.eval_constraint_jacobian(model.nonlin.evaluator, JJ, x)
vals = sparse(j_rows,j_cols,JJ).nzval
for k in 1:negcon
J[k] = vals[k]
end
if mode == 1 || mode == 2
g = unsafe_wrap(Array,g_, Int(n))
if model.nonlin.has_objective
MOI.eval_objective_gradient(model.nonlin.evaluator, g, x)
end
return
end
return
end

# SNOPT
fobj = 0.0
iobj = 0
hs = zeros(Int,n+m)
function eval_con(mode_::Ptr{Cint},
nncon_::Ptr{Cint}, nnjac_::Ptr{Cint}, negcon_::Ptr{Cint},
x_::Ptr{Float64}, c_::Ptr{Float64}, J_::Ptr{Float64},
status::Ptr{Cint})
mode = unsafe_load(mode_)
nncon = unsafe_load(nncon_)
nnjac = unsafe_load(nnjac_)
negcon = unsafe_load(negcon_)

x = unsafe_wrap(Array, x_, Int(nnjac))
if mode == 0 || mode == 2
c = unsafe_wrap(Array, c_, Int(nncon))
MOI.eval_constraint(model.nonlin.evaluator, c, x)
end

model.workspace = SNOPT7.initialize("snopt.out","screen")
if mode == 1 || mode == 2
J = unsafe_wrap(Array,J_, Int(negcon))

for (name,value) in model.options
# Replace underscore with space
keyword = replace(String(name), "_" => " ")
SNOPT7.setOption!(model.workspace, "$keyword $value")
# MOI returns Jacobian in order specified by j_rows, j_cols
# SNOPT needs it in sparse-by-col format
JJ = zeros(negcon)
MOI.eval_constraint_jacobian(model.nonlin.evaluator, JJ, x)
vals = sparse(j_rows,j_cols,JJ).nzval
for k in 1:negcon
J[k] = vals[k]
end
end
SNOPT7.setOption!(model.workspace, "Summary frequency 1")
SNOPT7.setOption!(model.workspace, "Print frequency 1")
return
end

start = "Cold"
name = "prob"
status = SNOPT7.snopt!(model.workspace, start, name,
m, n, nncon, nnobj, nnjac,
fobj, iobj, eval_con, eval_obj,
J, bl, bu, hs, x)
SNOPT7.freeWorkspace!(model.workspace)
# SNOPT
fobj = 0.0
iobj = 0
hs = zeros(Int,n+m)

else
# Linear/quadratic objective
start = "Cold"
name = "prob"
status = SNOPT7.snopt!(model.workspace, start, name,
m, n, nncon, nnobj, nnjac,
fobj, iobj, eval_con, eval_obj,
J, bl, bu, hs, x)

end
# Todo: call sqopt for LP/QPs
# if model.nonlin !== nothing
# else
# end

SNOPT7.freeWorkspace!(model.workspace)

end

Expand Down