Skip to content

add L1/Linf cone tests and fix bridge when VAF has nonzero constants #1045

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

Merged
merged 4 commits into from
May 11, 2020
Merged

add L1/Linf cone tests and fix bridge when VAF has nonzero constants #1045

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
103aede
add norminf and normone tests
chriscoey Mar 7, 2020
bdb9720
fix bug in get primal for norminf to lp bridge
chriscoey Mar 7, 2020
02b107c
add/fix bridge tests
chriscoey Mar 7, 2020
1cb4f50
Remove <:
blegat May 11, 2020
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
72 changes: 31 additions & 41 deletions src/Bridges/Constraint/norm_to_lp.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -70,69 +70,61 @@ The `NormOneCone` is representable with LP constraints, since
``t \\ge \\sum_i \\lvert x_i \\rvert`` if and only if there exists a vector y such that
``t \\ge \\sum_i y_i`` and ``y_i \\ge x_i``, ``y_i \\ge -x_i`` for all ``i``.
"""
struct NormOneBridge{T, F, G, H} <: AbstractBridge
struct NormOneBridge{T, F, G} <: AbstractBridge
y::Vector{MOI.VariableIndex}
ge_index::CI{F, MOI.GreaterThan{T}}
nn_index::CI{G, MOI.Nonnegatives}
nn_index::CI{F, MOI.Nonnegatives}
end
function bridge_constraint(::Type{NormOneBridge{T, F, G, H}}, model::MOI.ModelLike, f::H, s::MOI.NormOneCone) where {T, F, G, H}
function bridge_constraint(::Type{NormOneBridge{T, F, G}}, model::MOI.ModelLike, f::G, s::MOI.NormOneCone) where {T, F, G}
f_scalars = MOIU.eachscalar(f)
d = MOI.dimension(s)
y = MOI.add_variables(model, d - 1)
ge_index = MOIU.normalize_and_add_constraint(model, MOIU.operate(-, T, f_scalars[1], MOIU.operate(sum, T, y)), MOI.GreaterThan(zero(T)), allow_modify_function=true)
ge = MOIU.operate(-, T, f_scalars[1], MOIU.operate(sum, T, y))
lb = f_scalars[2:d]
ub = MOIU.operate(-, T, lb)
lb = MOIU.operate!(+, T, lb, MOI.VectorOfVariables(y))
ub = MOIU.operate!(+, T, ub, MOI.VectorOfVariables(y))
f_new = MOIU.operate(vcat, T, ub, lb)
nn_index = MOI.add_constraint(model, f_new, MOI.Nonnegatives(2d - 2))
return NormOneBridge{T, F, G, H}(y, ge_index, nn_index)
f_new = MOIU.operate(vcat, T, ge, ub, lb)
nn_index = MOI.add_constraint(model, f_new, MOI.Nonnegatives(2d - 1))
return NormOneBridge{T, F, G}(y, nn_index)
end

MOI.supports_constraint(::Type{NormOneBridge{T}}, ::Type{<:MOI.AbstractVectorFunction}, ::Type{MOI.NormOneCone}) where T = true
MOIB.added_constrained_variable_types(::Type{<:NormOneBridge}) = Tuple{DataType}[]
MOIB.added_constraint_types(::Type{NormOneBridge{T, F, G, H}}) where {T, F, G, H} = [(F, MOI.GreaterThan{T}), (G, MOI.Nonnegatives)]
function concrete_bridge_type(::Type{<:NormOneBridge{T}}, H::Type{<:MOI.AbstractVectorFunction}, ::Type{MOI.NormOneCone}) where T
S = MOIU.scalar_type(H)
F = MOIU.promote_operation(+, T, S, S)
G = MOIU.promote_operation(+, T, H, H)
return NormOneBridge{T, F, G, H}
MOIB.added_constraint_types(::Type{<:NormOneBridge{T, F}}) where {T, F} = [(F, MOI.Nonnegatives)]
function concrete_bridge_type(::Type{<:NormOneBridge{T}}, G::Type{<:MOI.AbstractVectorFunction}, ::Type{MOI.NormOneCone}) where T
S = MOIU.scalar_type(G)
F = MOIU.promote_operation(vcat, T, MOIU.promote_operation(+, T, S, S), MOIU.promote_operation(-, T, S, S))
return NormOneBridge{T, F, G}
end

# Attributes, Bridge acting as a model
MOI.get(b::NormOneBridge, ::MOI.NumberOfVariables) = length(b.y)
MOI.get(b::NormOneBridge, ::MOI.ListOfVariableIndices) = b.y
MOI.get(b::NormOneBridge{T, F, G, H}, ::MOI.NumberOfConstraints{F, MOI.GreaterThan{T}}) where {T, F, G, H} = 1
MOI.get(b::NormOneBridge{T, F, G, H}, ::MOI.NumberOfConstraints{G, MOI.Nonnegatives}) where {T, F, G, H} = 1
MOI.get(b::NormOneBridge{T, F, G, H}, ::MOI.ListOfConstraintIndices{F, MOI.GreaterThan{T}}) where {T, F, G, H} = [b.ge_index]
MOI.get(b::NormOneBridge{T, F, G, H}, ::MOI.ListOfConstraintIndices{G, MOI.Nonnegatives}) where {T, F, G, H} = [b.nn_index]
MOI.get(b::NormOneBridge{T, F}, ::MOI.NumberOfConstraints{F, MOI.Nonnegatives}) where {T, F} = 1
MOI.get(b::NormOneBridge{T, F}, ::MOI.ListOfConstraintIndices{F, MOI.Nonnegatives}) where {T, F} = [b.nn_index]

# References
function MOI.delete(model::MOI.ModelLike, c::NormOneBridge)
MOI.delete(model, c.nn_index)
MOI.delete(model, c.ge_index)
MOI.delete(model, c.y)
end

# Attributes, Bridge acting as a constraint
function MOI.get(model::MOI.ModelLike, ::MOI.ConstraintFunction, c::NormOneBridge{T, F, G, H}) where {T, F, G, H}
ge_func = MOI.get(model, MOI.ConstraintFunction(), c.ge_index)
function MOI.get(model::MOI.ModelLike, ::MOI.ConstraintFunction, c::NormOneBridge{T, F, G}) where {T, F, G}
nn_func = MOIU.eachscalar(MOI.get(model, MOI.ConstraintFunction(), c.nn_index))
t = MOIU.operate!(+, T, ge_func, MOIU.operate!(/, T, sum(nn_func), T(2)))
d = div(length(nn_func), 2)
x = MOIU.operate!(/, T, MOIU.operate!(-, T, nn_func[(d + 1):end], nn_func[1:d]), T(2))
return MOIU.convert_approx(H, MOIU.remove_variable(MOIU.operate(vcat, T, t, x), c.y))
t = MOIU.operate!(/, T, nn_func[1] + sum(nn_func), T(2))
d = div(length(nn_func) - 1, 2)
x = MOIU.operate!(/, T, MOIU.operate!(-, T, nn_func[(d + 2):end], nn_func[2:(d + 1)]), T(2))
return MOIU.convert_approx(G, MOIU.remove_variable(MOIU.operate(vcat, T, t, x), c.y))
end
function MOI.get(model::MOI.ModelLike, ::MOI.ConstraintSet, c::NormOneBridge)
dim = 1 + div(MOI.dimension(MOI.get(model, MOI.ConstraintSet(), c.nn_index)), 2)
dim = div(MOI.dimension(MOI.get(model, MOI.ConstraintSet(), c.nn_index)) + 1, 2)
return MOI.NormOneCone(dim)
end

function MOI.supports(
::MOI.ModelLike,
::Union{MOI.ConstraintPrimalStart, MOI.ConstraintDualStart},
::Type{<:NormOneBridge})

return true
end
function MOI.set(model::MOI.ModelLike, attr::MOI.ConstraintPrimalStart,
Expand All @@ -142,18 +134,17 @@ function MOI.set(model::MOI.ModelLike, attr::MOI.ConstraintPrimalStart,
for i in eachindex(bridge.y)
MOI.set(model, MOI.VariablePrimalStart(), bridge.y[i], y_value[i])
end
MOI.set(model, attr, bridge.nn_index, [y_value - x_value; y_value + x_value])
MOI.set(model, attr, bridge.ge_index, value[1] - reduce(+, y_value, init=zero(T)))
nn_value = vcat(value[1] - reduce(+, y_value, init=zero(T)), y_value - x_value, y_value + x_value)
MOI.set(model, attr, bridge.nn_index, nn_value)
return
end
function MOI.get(model::MOI.ModelLike,
attr::Union{MOI.ConstraintPrimal, MOI.ConstraintPrimalStart},
bridge::NormOneBridge)
ge_primal = MOI.get(model, attr, bridge.ge_index)
nn_primal = MOI.get(model, attr, bridge.nn_index)
t = ge_primal + sum(nn_primal) / 2
t = (nn_primal[1] + sum(nn_primal)) / 2
d = length(bridge.y)
x = (nn_primal[(d + 1):end] - nn_primal[1:d]) / 2
x = (nn_primal[(d + 2):end] - nn_primal[2:(d + 1)]) / 2
return vcat(t, x)
end
# Given a_i is dual on y_i - x_i >= 0 and b_i is dual on y_i + x_i >= 0 and c is dual on t - sum(y) >= 0,
Expand All @@ -162,27 +153,26 @@ end
function MOI.get(model::MOI.ModelLike,
attr::Union{MOI.ConstraintDual, MOI.ConstraintDualStart},
bridge::NormOneBridge)
t = MOI.get(model, attr, bridge.ge_index)
nn_dual = MOI.get(model, attr, bridge.nn_index)
d = length(bridge.y)
x = nn_dual[(d + 1):end] - nn_dual[1:d]
return vcat(t, x)
x = nn_dual[(d + 2):end] - nn_dual[2:(d + 1)]
return vcat(nn_dual[1], x)
end
# value[1 + i] = nn_dual[d + i] - nn_dual[i]
# value[1 + i] = nn_dual[1 + d + i] - nn_dual[1 + i]
# and `nn_dual` is nonnegative. By complementarity slackness, only one of each
# `nn_dual` can be nonzero (except if `x = 0`) so we can set
# depending on the sense of `value[1 + i]`.
function MOI.set(model::MOI.ModelLike, ::MOI.ConstraintDualStart,
bridge::NormOneBridge, value)
t = MOI.set(model, MOI.ConstraintDualStart(), bridge.ge_index, value[1])
d = length(bridge.y)
nn_dual = zeros(eltype(value), 2d)
nn_dual = zeros(eltype(value), 2d + 1)
nn_dual[1] = value[1]
for i in eachindex(bridge.y)
v = value[1 + i]
if v < 0
nn_dual[i] = -v
nn_dual[1 + i] = -v
else
nn_dual[d + i] = v
nn_dual[1 + d + i] = v
end
end
MOI.set(model, MOI.ConstraintDualStart(), bridge.nn_index, nn_dual)
Expand Down
140 changes: 138 additions & 2 deletions src/Test/contconic.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -432,9 +432,77 @@ function norminf2test(model::MOI.ModelLike, config::TestConfig)
end
end

function norminf3test(model::MOI.ModelLike, config::TestConfig)
atol = config.atol
rtol = config.rtol
# Problem NormInf3
# min x
# st (-1 + x, 2 .+ y) in NormInf(1 + n)
# (1 .+ y) in Nonnegatives(n)
# let n = 3. optimal solution: y .= -1, x = 2

@test MOIU.supports_default_copy_to(model, #=copy_names=# false)
@test MOI.supports(model, MOI.ObjectiveFunction{MOI.SingleVariable}())
@test MOI.supports(model, MOI.ObjectiveSense())
@test MOI.supports_constraint(model, MOI.VectorAffineFunction{Float64}, MOI.NormInfinityCone)

MOI.empty!(model)
@test MOI.is_empty(model)

x = MOI.add_variable(model)
y = MOI.add_variables(model, 3)

MOI.set(model, MOI.ObjectiveFunction{MOI.SingleVariable}(), MOI.SingleVariable(x))
MOI.set(model, MOI.ObjectiveSense(), MOI.MIN_SENSE)

norminf_vaf = MOI.VectorAffineFunction(MOI.VectorAffineTerm.(1:4,
MOI.ScalarAffineTerm.(1.0, vcat(x, y))), [-1.0, 2, 2, 2])
norminf = MOI.add_constraint(model, norminf_vaf, MOI.NormInfinityCone(4))
nonneg_vaf = MOI.VectorAffineFunction(MOI.VectorAffineTerm.(1:3,
MOI.ScalarAffineTerm.(1.0, y)), ones(3))
nonneg = MOI.add_constraint(model, nonneg_vaf, MOI.Nonnegatives(3))

@test MOI.get(model, MOI.NumberOfConstraints{MOI.VectorAffineFunction{Float64}, MOI.NormInfinityCone}()) == 1
@test MOI.get(model, MOI.NumberOfConstraints{MOI.VectorAffineFunction{Float64}, MOI.Nonnegatives}()) == 1
loc = MOI.get(model, MOI.ListOfConstraints())
@test length(loc) == 2
@test (MOI.VectorAffineFunction{Float64}, MOI.NormInfinityCone) in loc
@test (MOI.VectorAffineFunction{Float64}, MOI.Nonnegatives) in loc

if config.solve
@test MOI.get(model, MOI.TerminationStatus()) == MOI.OPTIMIZE_NOT_CALLED

MOI.optimize!(model)

@test MOI.get(model, MOI.TerminationStatus()) == config.optimal_status

@test MOI.get(model, MOI.PrimalStatus()) == MOI.FEASIBLE_POINT
if config.duals
@test MOI.get(model, MOI.DualStatus()) == MOI.FEASIBLE_POINT
end

@test MOI.get(model, MOI.ObjectiveValue()) ≈ 2 atol=atol rtol=rtol
if config.dual_objective_value
@test MOI.get(model, MOI.DualObjectiveValue()) ≈ 2 atol=atol rtol=rtol
end

@test MOI.get(model, MOI.VariablePrimal(), x) ≈ 2 atol=atol rtol=rtol
@test MOI.get(model, MOI.VariablePrimal(), y) ≈ fill(-1.0, 3) atol=atol rtol=rtol

@test MOI.get(model, MOI.ConstraintPrimal(), norminf) ≈ ones(4) atol=atol rtol=rtol
@test MOI.get(model, MOI.ConstraintPrimal(), nonneg) ≈ zeros(3) atol=atol rtol=rtol

if config.duals
@test MOI.get(model, MOI.ConstraintDual(), norminf) ≈ vcat(1, fill(-inv(3), 3)) atol=atol rtol=rtol
@test MOI.get(model, MOI.ConstraintDual(), nonneg) ≈ fill(inv(3), 3) atol=atol rtol=rtol
end
end
end

const norminftests = Dict("norminf1v" => norminf1vtest,
"norminf1f" => norminf1ftest,
"norminf2" => norminf2test)
"norminf2" => norminf2test,
"norminf3" => norminf3test)

@moitestset norminf

Expand Down Expand Up @@ -568,9 +636,77 @@ function normone2test(model::MOI.ModelLike, config::TestConfig)
end
end

function normone3test(model::MOI.ModelLike, config::TestConfig)
atol = config.atol
rtol = config.rtol
# Problem NormOne3
# min x
# st (-1 + x, 2 .+ y) in NormOne(1 + n)
# (1 .+ y) in Nonnegatives(n)
# let n = 3. optimal solution: y .= -1, x = 4

@test MOIU.supports_default_copy_to(model, #=copy_names=# false)
@test MOI.supports(model, MOI.ObjectiveFunction{MOI.SingleVariable}())
@test MOI.supports(model, MOI.ObjectiveSense())
@test MOI.supports_constraint(model, MOI.VectorAffineFunction{Float64}, MOI.NormOneCone)

MOI.empty!(model)
@test MOI.is_empty(model)

x = MOI.add_variable(model)
y = MOI.add_variables(model, 3)

MOI.set(model, MOI.ObjectiveFunction{MOI.SingleVariable}(), MOI.SingleVariable(x))
MOI.set(model, MOI.ObjectiveSense(), MOI.MIN_SENSE)

norminf_vaf = MOI.VectorAffineFunction(MOI.VectorAffineTerm.(1:4,
MOI.ScalarAffineTerm.(1.0, vcat(x, y))), [-1.0, 2, 2, 2])
norminf = MOI.add_constraint(model, norminf_vaf, MOI.NormOneCone(4))
nonneg_vaf = MOI.VectorAffineFunction(MOI.VectorAffineTerm.(1:3,
MOI.ScalarAffineTerm.(1.0, y)), ones(3))
nonneg = MOI.add_constraint(model, nonneg_vaf, MOI.Nonnegatives(3))

@test MOI.get(model, MOI.NumberOfConstraints{MOI.VectorAffineFunction{Float64}, MOI.NormOneCone}()) == 1
@test MOI.get(model, MOI.NumberOfConstraints{MOI.VectorAffineFunction{Float64}, MOI.Nonnegatives}()) == 1
loc = MOI.get(model, MOI.ListOfConstraints())
@test length(loc) == 2
@test (MOI.VectorAffineFunction{Float64}, MOI.NormOneCone) in loc
@test (MOI.VectorAffineFunction{Float64}, MOI.Nonnegatives) in loc

if config.solve
@test MOI.get(model, MOI.TerminationStatus()) == MOI.OPTIMIZE_NOT_CALLED

MOI.optimize!(model)

@test MOI.get(model, MOI.TerminationStatus()) == config.optimal_status

@test MOI.get(model, MOI.PrimalStatus()) == MOI.FEASIBLE_POINT
if config.duals
@test MOI.get(model, MOI.DualStatus()) == MOI.FEASIBLE_POINT
end

@test MOI.get(model, MOI.ObjectiveValue()) ≈ 4 atol=atol rtol=rtol
if config.dual_objective_value
@test MOI.get(model, MOI.DualObjectiveValue()) ≈ 4 atol=atol rtol=rtol
end

@test MOI.get(model, MOI.VariablePrimal(), x) ≈ 4 atol=atol rtol=rtol
@test MOI.get(model, MOI.VariablePrimal(), y) ≈ fill(-1.0, 3) atol=atol rtol=rtol

@test MOI.get(model, MOI.ConstraintPrimal(), norminf) ≈ vcat(3, ones(3)) atol=atol rtol=rtol
@test MOI.get(model, MOI.ConstraintPrimal(), nonneg) ≈ zeros(3) atol=atol rtol=rtol

if config.duals
@test MOI.get(model, MOI.ConstraintDual(), norminf) ≈ vcat(1, fill(-1, 3)) atol=atol rtol=rtol
@test MOI.get(model, MOI.ConstraintDual(), nonneg) ≈ ones(3) atol=atol rtol=rtol
end
end
end

const normonetests = Dict("normone1v" => normone1vtest,
"normone1f" => normone1ftest,
"normone2" => normone2test)
"normone2" => normone2test,
"normone3" => normone3test)

@moitestset normone

Expand Down
Loading