Skip to content
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

Quadratic constraints #102

Closed
wants to merge 16 commits into from
Closed

Quadratic constraints #102

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
16 commits
Select commit Hold shift + click to select a range
7273d2f
[WIP] Quadratic constraints
amontoison Mar 10, 2022
6990308
add straightforward test
tmigot May 6, 2022
2b1574d
debug utils functions
tmigot May 6, 2022
3bec27b
add API for quadratic constraints
tmigot May 6, 2022
0f08e9a
simplify `nquad`
tmigot May 6, 2022
af04c2e
add hs100
tmigot May 10, 2022
2aaf353
store `vec`
tmigot May 10, 2022
90ea72a
implement hprod and jac_coord properly
tmigot May 10, 2022
836a7b9
add test problem and fix
tmigot May 11, 2022
494e3a0
only evaluate the NLPEval when there is nonlinear cons
tmigot May 11, 2022
01863f7
[MathOptNLPModel] Fix hprod!
amontoison May 12, 2022
86ccf44
[MathOptNLPModel] Fix cases with nlp.eval == nothing
amontoison May 12, 2022
ef95770
Update test/nlp_problems/hs100.jl
amontoison May 12, 2022
c64ac74
Remove hessian_structure function
amontoison May 12, 2022
4b24cab
Optimize hessian_structure function for quadratic constraints
amontoison May 12, 2022
e141a9c
[MathOptNLPModel] hotfix
amontoison May 13, 2022
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
105 changes: 80 additions & 25 deletions src/moi_nlp_model.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,6 +4,7 @@ mutable struct MathOptNLPModel <: AbstractNLPModel{Float64, Vector{Float64}}
meta::NLPModelMeta{Float64, Vector{Float64}}
eval::Union{MOI.AbstractNLPEvaluator, Nothing}
lincon::LinearConstraints
quadcon::QuadraticConstraints
obj::Objective
counters::Counters
end
Expand Down Expand Up @@ -33,19 +34,19 @@ function MathOptNLPModel(jmodel::JuMP.Model; hessian::Bool = true, name::String
(nnln == 0 ? 0 : sum(length(nl_con.hess_I) for nl_con in eval.constraints)) : 0

moimodel = backend(jmodel)
nlin, lincon, lin_lcon, lin_ucon = parser_MOI(moimodel)
nlin, lincon, lin_lcon, lin_ucon, quadcon, quad_lcon, quad_ucon = parser_MOI(moimodel, nvar)

if (eval ≠ nothing) && eval.has_nlobj
obj = Objective("NONLINEAR", 0.0, spzeros(Float64, nvar), COO(), 0)
else
obj = parser_objective_MOI(moimodel, nvar)
end

ncon = nlin + nnln
lcon = vcat(lin_lcon, nl_lcon)
ucon = vcat(lin_ucon, nl_ucon)
nnzj = lincon.nnzj + nl_nnzj
nnzh = obj.nnzh + nl_nnzh
ncon = nlin + quadcon.nquad + nnln
lcon = vcat(lin_lcon, quad_lcon, nl_lcon)
ucon = vcat(lin_ucon, quad_ucon, nl_ucon)
nnzj = lincon.nnzj + quadcon.nnzj + nl_nnzj
nnzh = obj.nnzh + quadcon.nnzh + nl_nnzh

meta = NLPModelMeta(
nvar,
Expand All @@ -60,13 +61,13 @@ function MathOptNLPModel(jmodel::JuMP.Model; hessian::Bool = true, name::String
nnzh = nnzh,
lin = collect(1:nlin),
lin_nnzj = lincon.nnzj,
nln_nnzj = nl_nnzj,
nln_nnzj = quadcon.nnzj + nl_nnzj,
minimize = objective_sense(jmodel) == MOI.MIN_SENSE,
islp = (obj.type == "LINEAR") && (nnln == 0),
islp = (obj.type == "LINEAR") && (nnln == 0) && (quadcon.nquad == 0),
name = name,
)

return MathOptNLPModel(meta, eval, lincon, obj, Counters())
return MathOptNLPModel(meta, eval, lincon, quadcon, obj, Counters())
end

function NLPModels.obj(nlp::MathOptNLPModel, x::AbstractVector)
Expand Down Expand Up @@ -115,7 +116,13 @@ end

function NLPModels.cons_nln!(nlp::MathOptNLPModel, x::AbstractVector, c::AbstractVector)
increment!(nlp, :neval_cons_nln)
MOI.eval_constraint(nlp.eval, c, x)
for i = 1:(nlp.quadcon.nquad)
qcon = nlp.quadcon[i]
c[i] = 0.5 * coo_sym_dot(qcon.hessian.rows, qcon.hessian.cols, qcon.hessian.vals, x, x) + dot(qcon.b, x)
end
if nlp.meta.nnln > nlp.quadcon.nquad
MOI.eval_constraint(nlp.eval, view(c, (nlp.quadcon.nquad + 1):(nlp.meta.nnln)), x)
end
return c
end

Expand All @@ -134,11 +141,16 @@ function NLPModels.jac_nln_structure!(
rows::AbstractVector{<:Integer},
cols::AbstractVector{<:Integer},
)
jac_struct = MOI.jacobian_structure(nlp.eval)
for index = 1:(nlp.meta.nln_nnzj)
row, col = jac_struct[index]
rows[index] = row
cols[index] = col
quad_nnzj, jrows, jcols = nlp.quadcon.nnzj, nlp.quadcon.jrows, nlp.quadcon.jcols
rows[1:quad_nnzj] .= jrows
cols[1:quad_nnzj] .= jcols
if nlp.meta.nnln > nlp.quadcon.nquad
jac_struct = MOI.jacobian_structure(nlp.eval)
for index = 1:(nlp.meta.nln_nnzj - quad_nnzj)
row, col = jac_struct[index]
rows[quad_nnzj + index] = row + nlp.quadcon.nquad
cols[quad_nnzj + index] = col
end
end
return rows, cols
end
Expand All @@ -151,7 +163,22 @@ end

function NLPModels.jac_nln_coord!(nlp::MathOptNLPModel, x::AbstractVector, vals::AbstractVector)
increment!(nlp, :neval_jac_nln)
MOI.eval_constraint_jacobian(nlp.eval, vals, x)
vals .= 0.0
k = 0
for i = 1:(nlp.quadcon.nquad)
qcon = nlp.quadcon[i]
for j=1:length(qcon.vec)
vals[k + j] = qcon.b[qcon.vec[j]]
end
nnzj = length(qcon.hessian.vals)
Copy link
Member Author

@amontoison amontoison May 12, 2022 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nnzj should be the number of nnz of Qᵢx + bᵢ (length(set)) and not Qᵢ.

for i=1:nnzj
vals[k + i] += qcon.hessian.vals[i] * x[qcon.hessian.cols[i]]
end
k += nnzj
end
if nlp.meta.nnln > nlp.quadcon.nquad
MOI.eval_constraint_jacobian(nlp.eval, view(vals, (nlp.quadcon.nnzj + 1):(nlp.meta.nln_nnzj)), x)
end
return vals
end

Expand Down Expand Up @@ -305,10 +332,16 @@ function NLPModels.hess_structure!(
cols[index] = nlp.obj.hessian.cols[index]
end
end
if (nlp.obj.type == "NONLINEAR") || (nlp.meta.nnln > 0)
if nlp.quadcon.nquad > 0
for (index, tuple) in enumerate(nlp.quadcon.set)
rows[nlp.obj.nnzh + index] = tuple[1]
cols[nlp.obj.nnzh + index] = tuple[2]
end
end
if (nlp.obj.type == "NONLINEAR") || (nlp.meta.nnln > nlp.quadcon.nquad)
hesslag_struct = MOI.hessian_lagrangian_structure(nlp.eval)
amontoison marked this conversation as resolved.
Show resolved Hide resolved
for index = (nlp.obj.nnzh + 1):(nlp.meta.nnzh)
shift_index = index - nlp.obj.nnzh
for index = (nlp.obj.nnzh + nlp.quadcon.nnzh + 1):(nlp.meta.nnzh)
shift_index = index - nlp.obj.nnzh - nlp.quadcon.nnzh
rows[index] = hesslag_struct[shift_index][1]
cols[index] = hesslag_struct[shift_index][2]
end
Expand All @@ -327,13 +360,23 @@ function NLPModels.hess_coord!(
if nlp.obj.type == "QUADRATIC"
vals[1:(nlp.obj.nnzh)] .= obj_weight .* nlp.obj.hessian.vals
end
if (nlp.obj.type == "NONLINEAR") || (nlp.meta.nnln > 0)
if nlp.quadcon.nquad > 0
quad_nnzh = nlp.quadcon.nnzh
k = 0
for i = 1:(nlp.quadcon.nquad)
qcon = nlp.quadcon[i]
nnzh = length(qcon.hessian.vals)
vals[(k + 1):(k + nnzh)] .= qcon.hessian.vals .* y[nlp.meta.nlin + i]
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We want to avoid that with the hessian_quad function!
If we have 10 quadratic constraints Q_i with same structure, you will store 10 times the required number of nnz for nothing.

k += nnzh
end
end
if (nlp.obj.type == "NONLINEAR") || (nlp.meta.nnln > nlp.quadcon.nquad)
MOI.eval_hessian_lagrangian(
nlp.eval,
view(vals, (nlp.obj.nnzh + 1):(nlp.meta.nnzh)),
view(vals, (nlp.obj.nnzh + nlp.quadcon.nnzh + 1):(nlp.meta.nnzh)),
x,
obj_weight,
view(y, nlp.meta.nln),
view(y, (nlp.meta.nlin + nlp.quadcon.nquad + 1):(nlp.meta.ncon))
)
end
return vals
Expand All @@ -354,6 +397,7 @@ function NLPModels.hess_coord!(
vals[(nlp.obj.nnzh + 1):(nlp.meta.nnzh)] .= 0.0
end
if nlp.obj.type == "NONLINEAR"
vals .= 0.0
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

MOI.eval_hessian_lagrangian doesn't overwrite vals?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

https://github.com/jump-dev/MathOptInterface.jl/blob/master/src/Nonlinear/ReverseAD/mathoptinterface_api.jl#L261-L273

Copy link
Member Author

@amontoison amontoison May 13, 2022 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If vals .= 0.0 is required, we should also add it in the other NLPModels.hess_coord! function.

MOI.eval_hessian_lagrangian(nlp.eval, vals, x, obj_weight, zeros(nlp.meta.nnln))
end

Expand All @@ -372,8 +416,18 @@ function NLPModels.hprod!(
if (nlp.obj.type == "LINEAR") && (nlp.meta.nnln == 0)
hv .= 0.0
end
if (nlp.obj.type == "NONLINEAR") || (nlp.meta.nnln > 0)
MOI.eval_hessian_lagrangian_product(nlp.eval, hv, x, v, obj_weight, view(y, nlp.meta.nln))
if nlp.quadcon.nquad > 0
for i = 1:(nlp.quadcon.nquad)
qcon = nlp.quadcon[i]
for k = 1:length(qcon.hessian.vals)
hv[qcon.hessian.rows[k]] += qcon.hessian.vals[k] * v[qcon.hessian.cols[k]]
end
hv[i] *= y[nlp.meta.nlin + i]
end
end
if (nlp.obj.type == "NONLINEAR") || (nlp.meta.nnln > nlp.quadcon.nquad)
ind_nln = (nlp.meta.nlin + nlp.quadcon.nquad + 1):(nlp.meta.ncon)
MOI.eval_hessian_lagrangian_product(nlp.eval, hv, x, v, obj_weight, view(y, ind_nln))
end
if nlp.obj.type == "QUADRATIC"
nlp.meta.nnln == 0 && (hv .= 0.0)
Expand Down Expand Up @@ -404,7 +458,8 @@ function NLPModels.hprod!(
hv .*= obj_weight
end
if nlp.obj.type == "NONLINEAR"
MOI.eval_hessian_lagrangian_product(nlp.eval, hv, x, v, obj_weight, zeros(nlp.meta.nnln))
nnln = nlp.meta.nnln - nlp.quadcon.nquad
MOI.eval_hessian_lagrangian_product(nlp.eval, hv, x, v, obj_weight, zeros(nnln))
end
return hv
end
2 changes: 1 addition & 1 deletion src/moi_nls_model.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,7 +40,7 @@ function MathOptNLSModel(cmodel::JuMP.Model, F; hessian::Bool = true, name::Stri
(nnln == 0 ? 0 : sum(length(con.hess_I) for con in ceval.constraints)) : 0

moimodel = backend(cmodel)
nlin, lincon, lin_lcon, lin_ucon = parser_MOI(moimodel)
nlin, lincon, lin_lcon, lin_ucon, quadcon, quad_lcon, quad_ucon = parser_MOI(moimodel, nvar)

nequ = nlinequ + nnlnequ
Fnnzj = linequ.nnzj + nl_Fnnzj
Expand Down
Loading