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.

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

[Bridges.Constraint] add bridges in alphabetical order #2596

Merged
merged 9 commits into from
Jan 5, 2025
Merged
Show file tree
Hide file tree
Changes from 5 commits
Commits
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
155 changes: 92 additions & 63 deletions src/Bridges/Constraint/Constraint.jl
Original file line number Diff line number Diff line change
Expand Up @@ -29,84 +29,113 @@ Add all bridges defined in the `Bridges.Constraint` submodule to `model`.
The coefficient type used is `T`.
"""
function add_all_bridges(model, ::Type{T}) where {T}
if T <: AbstractFloat
MOI.Bridges.add_bridge(model, AllDifferentToCountDistinctBridge{T})
MOI.Bridges.add_bridge(model, BinPackingToMILPBridge{T})
MOI.Bridges.add_bridge(model, CircuitToMILPBridge{T})
MOI.Bridges.add_bridge(model, ComplexNormInfinityToSecondOrderConeBridge{T})
MOI.Bridges.add_bridge(model, CountAtLeastToCountBelongsBridge{T})
MOI.Bridges.add_bridge(model, CountBelongsToMILPBridge{T})
MOI.Bridges.add_bridge(model, CountDistinctToMILPBridge{T})
MOI.Bridges.add_bridge(model, CountGreaterThanToMILPBridge{T})
# * ExponentialConeToScalarNonlinearFunctionBridge{T}
# This bridge is not added by default because it starts with a convex
# conic constraint and adds a nonlinear constraint that local NLP
# solvers like Ipopt can struggle with because of log(x) when x is 0.
# In addition, the bridge does not support ConstraintDual.
# * FunctionConversionBridge{T}
# This bridge is not added because, even though it is not abstract, it
# is highly parameterized, and parameterized versions such as
# ScalarFunctionizeBridge are added.
MOI.Bridges.add_bridge(model, GeoMeanBridge{T})
MOI.Bridges.add_bridge(model, GeoMeanToPowerBridge{T})
MOI.Bridges.add_bridge(model, GeoMeantoRelEntrBridge{T})
if T <: AbstractFloat # See note in docstring of AbstractToIntervalBridge
MOI.Bridges.add_bridge(model, GreaterToIntervalBridge{T})
MOI.Bridges.add_bridge(model, LessToIntervalBridge{T})
end
odow marked this conversation as resolved.
Show resolved Hide resolved
MOI.Bridges.add_bridge(model, GreaterToLessBridge{T})
MOI.Bridges.add_bridge(model, HermitianToSymmetricPSDBridge{T})
MOI.Bridges.add_bridge(model, IndicatorActiveOnFalseBridge{T})
MOI.Bridges.add_bridge(model, IndicatorGreaterToLessThanBridge{T})
MOI.Bridges.add_bridge(model, IndicatorLessToGreaterThanBridge{T})
# * IndicatorSetMapBridge{T}
# This bridge is not added because, even though it is not abstract, it
# is highly parameterized, and parameterized versions such as
# IndicatorGreaterToLessThanBridge are added.
MOI.Bridges.add_bridge(model, IndicatorSOS1Bridge{T})
MOI.Bridges.add_bridge(model, IndicatorToMILPBridge{T})
# * InequalityToComplementsBridge{T}
# This bridge is not added because of a bug in Convex.jl:
# https://github.com/jump-dev/Convex.jl/blob/ca5324217575af263bfeee20b3e0526bed051887/src/MOI_wrapper.jl#L119-L133
# It is also really useful only to PATHSolver.jl, which could add this
# to MOI.ListOfRequiredBridges.
MOI.Bridges.add_bridge(model, IntegerToZeroOneBridge{T})
MOI.Bridges.add_bridge(model, LessToGreaterBridge{T})
if T <: AbstractFloat # See note in docstring of AbstractToIntervalBridge
MOI.Bridges.add_bridge(model, LessToIntervalBridge{T})
end
MOI.Bridges.add_bridge(model, LogDetBridge{T})
MOI.Bridges.add_bridge(model, NonnegToNonposBridge{T})
MOI.Bridges.add_bridge(model, NonposToNonnegBridge{T})
MOI.Bridges.add_bridge(model, ScalarizeBridge{T})
MOI.Bridges.add_bridge(model, VectorizeBridge{T})
MOI.Bridges.add_bridge(model, ScalarSlackBridge{T})
MOI.Bridges.add_bridge(model, VectorSlackBridge{T})
MOI.Bridges.add_bridge(model, ScalarFunctionizeBridge{T})
MOI.Bridges.add_bridge(model, VectorFunctionizeBridge{T})
MOI.Bridges.add_bridge(model, ToScalarQuadraticBridge{T})
MOI.Bridges.add_bridge(model, ToVectorQuadraticBridge{T})
MOI.Bridges.add_bridge(model, ToScalarNonlinearBridge{T})
MOI.Bridges.add_bridge(model, SplitHyperRectangleBridge{T})
MOI.Bridges.add_bridge(model, SplitIntervalBridge{T})
MOI.Bridges.add_bridge(model, SplitComplexEqualToBridge{T})
MOI.Bridges.add_bridge(model, SplitComplexZerosBridge{T})
MOI.Bridges.add_bridge(model, QuadtoSOCBridge{T})
# We do not add `(R)SOCtoNonConvexQuad` because it starts with a convex
# conic constraint and generate a non-convex constraint (in the QCP
# interpretation).
MOI.Bridges.add_bridge(model, NormInfinityBridge{T})
MOI.Bridges.add_bridge(model, NormInfinityConeToNormConeBridge{T})
MOI.Bridges.add_bridge(model, NormNuclearBridge{T})
MOI.Bridges.add_bridge(model, NormOneBridge{T})
MOI.Bridges.add_bridge(model, GeoMeantoRelEntrBridge{T})
MOI.Bridges.add_bridge(model, GeoMeanBridge{T})
MOI.Bridges.add_bridge(model, GeoMeanToPowerBridge{T})
MOI.Bridges.add_bridge(model, NormToPowerBridge{T})
MOI.Bridges.add_bridge(model, NormOneConeToNormConeBridge{T})
MOI.Bridges.add_bridge(model, SecondOrderConeToNormConeBridge{T})
MOI.Bridges.add_bridge(model, NormInfinityConeToNormConeBridge{T})
MOI.Bridges.add_bridge(model, ComplexNormInfinityToSecondOrderConeBridge{T})
MOI.Bridges.add_bridge(model, RelativeEntropyBridge{T})
# * NormSpecialCaseBridge{T}
# This bridge is not added because, even though it is not abstract, it
# is highly parameterized, and parameterized versions such as
# NormOneConeToNormConeBridge are added.
MOI.Bridges.add_bridge(model, NormSpectralBridge{T})
MOI.Bridges.add_bridge(model, NormNuclearBridge{T})
MOI.Bridges.add_bridge(model, HermitianToSymmetricPSDBridge{T})
MOI.Bridges.add_bridge(model, SquareBridge{T})
MOI.Bridges.add_bridge(model, SetDotScalingBridge{T})
MOI.Bridges.add_bridge(model, SetDotInverseScalingBridge{T})
MOI.Bridges.add_bridge(model, LogDetBridge{T})
MOI.Bridges.add_bridge(model, NormToPowerBridge{T})
# * NumberConversionBridge{T}
# This bridge is not added by default because it would silently enable
# models with mixed precision. In most cases, this is a bug in the
# user's code, so we leave this bridge as opt-in.
MOI.Bridges.add_bridge(model, QuadtoSOCBridge{T})
MOI.Bridges.add_bridge(model, ReifiedAllDifferentToCountDistinctBridge{T})
MOI.Bridges.add_bridge(model, ReifiedCountDistinctToMILPBridge{T})
MOI.Bridges.add_bridge(model, RelativeEntropyBridge{T})
MOI.Bridges.add_bridge(model, RootDetBridge{T})
MOI.Bridges.add_bridge(model, RSOCtoSOCBridge{T})
MOI.Bridges.add_bridge(model, SOCtoRSOCBridge{T})
# We do not add `SOCtoPSDBridge` as transforming the `SOC` to `RSOC` and
# then to `PSD` produces a smaller SDP constraint.
# MOI.Bridges.add_bridge(model, SOCtoPSDBridge{T})
# * RSOCtoNonConvexQuadBridge{T}
# This bridge is not added by default because it starts with a convex
# conic constraint and generate a non-convex constraint (in the QCP
# interpretation).
MOI.Bridges.add_bridge(model, RSOCtoPSDBridge{T})
MOI.Bridges.add_bridge(model, IndicatorActiveOnFalseBridge{T})
MOI.Bridges.add_bridge(model, IndicatorSOS1Bridge{T})
MOI.Bridges.add_bridge(model, IndicatorLessToGreaterThanBridge{T})
MOI.Bridges.add_bridge(model, IndicatorGreaterToLessThanBridge{T})
MOI.Bridges.add_bridge(model, RSOCtoSOCBridge{T})
MOI.Bridges.add_bridge(model, ScalarFunctionizeBridge{T})
MOI.Bridges.add_bridge(model, ScalarizeBridge{T})
MOI.Bridges.add_bridge(model, ScalarSlackBridge{T})
MOI.Bridges.add_bridge(model, SecondOrderConeToNormConeBridge{T})
MOI.Bridges.add_bridge(model, SemiToBinaryBridge{T})
MOI.Bridges.add_bridge(model, ZeroOneBridge{T})
MOI.Bridges.add_bridge(model, IntegerToZeroOneBridge{T})
MOI.Bridges.add_bridge(model, InequalityToComplementsBridge{T})
# Do not add by default
# MOI.Bridges.add_bridge(model, NumberConversionBridge{T})
# Constraint programming bridges
MOI.Bridges.add_bridge(model, AllDifferentToCountDistinctBridge{T})
MOI.Bridges.add_bridge(model, ReifiedAllDifferentToCountDistinctBridge{T})
MOI.Bridges.add_bridge(model, BinPackingToMILPBridge{T})
MOI.Bridges.add_bridge(model, CircuitToMILPBridge{T})
MOI.Bridges.add_bridge(model, CountAtLeastToCountBelongsBridge{T})
MOI.Bridges.add_bridge(model, CountBelongsToMILPBridge{T})
MOI.Bridges.add_bridge(model, CountDistinctToMILPBridge{T})
MOI.Bridges.add_bridge(model, ReifiedCountDistinctToMILPBridge{T})
MOI.Bridges.add_bridge(model, CountGreaterThanToMILPBridge{T})
MOI.Bridges.add_bridge(model, TableToMILPBridge{T})
# * SetConversionBridge{T}
# This bridge is not added because, even though it is not abstract, it
# is highly parameterized, and it intended for use by MOI extensions.
MOI.Bridges.add_bridge(model, SetDotInverseScalingBridge{T})
MOI.Bridges.add_bridge(model, SetDotScalingBridge{T})
# * SOCtoNonConvexQuadBridge{T}
# This bridge is not added by default because it starts with a convex
# conic constraint and generate a non-convex constraint (in the QCP
# interpretation).
# * SOCtoPSDBridge{T}
# This bridge is not added because transforming the `SOC` to `RSOC` and
# then to `PSD` produces a smaller SDP constraint. `RSOCtoPSDBridge` is
# added by default.
MOI.Bridges.add_bridge(model, SOCtoRSOCBridge{T})
MOI.Bridges.add_bridge(model, SOS1ToMILPBridge{T})
MOI.Bridges.add_bridge(model, SOS2ToMILPBridge{T})
MOI.Bridges.add_bridge(model, IndicatorToMILPBridge{T})
MOI.Bridges.add_bridge(
model,
ExponentialConeToScalarNonlinearFunctionBridge{T},
)
MOI.Bridges.add_bridge(model, SplitComplexEqualToBridge{T})
MOI.Bridges.add_bridge(model, SplitComplexZerosBridge{T})
MOI.Bridges.add_bridge(model, SplitHyperRectangleBridge{T})
MOI.Bridges.add_bridge(model, SplitIntervalBridge{T})
MOI.Bridges.add_bridge(model, SquareBridge{T})
MOI.Bridges.add_bridge(model, TableToMILPBridge{T})
MOI.Bridges.add_bridge(model, ToScalarNonlinearBridge{T})
MOI.Bridges.add_bridge(model, ToScalarQuadraticBridge{T})
MOI.Bridges.add_bridge(model, ToVectorQuadraticBridge{T})
MOI.Bridges.add_bridge(model, VectorFunctionizeBridge{T})
MOI.Bridges.add_bridge(model, VectorizeBridge{T})
MOI.Bridges.add_bridge(model, VectorSlackBridge{T})
MOI.Bridges.add_bridge(model, ZeroOneBridge{T})
return
end

Expand Down
10 changes: 10 additions & 0 deletions src/Bridges/Constraint/bridges/SOCtoPSDBridge.jl
Original file line number Diff line number Diff line change
Expand Up @@ -77,6 +77,11 @@
const SOCtoPSD{T,OT<:MOI.ModelLike} =
SingleBridgeOptimizer{SOCtoPSDBridge{T},OT}

# This bridge destorys a lot of structure and adding PSD variables is almost
# always undesirable. We give this bridge an arbitrarily hight cost so that it
odow marked this conversation as resolved.
Show resolved Hide resolved
odow marked this conversation as resolved.
Show resolved Hide resolved
# is used only if necessary.
bridging_cost(::Type{<:SOCtoPSDBridge}) = 10.0

Check warning on line 83 in src/Bridges/Constraint/bridges/SOCtoPSDBridge.jl

View check run for this annotation

Codecov / codecov/patch

src/Bridges/Constraint/bridges/SOCtoPSDBridge.jl#L83

Added line #L83 was not covered by tests
odow marked this conversation as resolved.
Show resolved Hide resolved

function concrete_bridge_type(
::Type{<:SOCtoPSDBridge{T}},
G::Type{<:MOI.AbstractVectorFunction},
Expand Down Expand Up @@ -186,6 +191,11 @@
const RSOCtoPSD{T,OT<:MOI.ModelLike} =
SingleBridgeOptimizer{RSOCtoPSDBridge{T},OT}

# This bridge destorys a lot of structure and adding PSD variables is almost
# always undesirable. We give this bridge an arbitrarily hight cost so that it
odow marked this conversation as resolved.
Show resolved Hide resolved
# is used only if necessary.
bridging_cost(::Type{<:RSOCtoPSDBridge}) = 10.0

Check warning on line 197 in src/Bridges/Constraint/bridges/SOCtoPSDBridge.jl

View check run for this annotation

Codecov / codecov/patch

src/Bridges/Constraint/bridges/SOCtoPSDBridge.jl#L197

Added line #L197 was not covered by tests
odow marked this conversation as resolved.
Show resolved Hide resolved
odow marked this conversation as resolved.
Show resolved Hide resolved

function concrete_bridge_type(
::Type{<:RSOCtoPSDBridge{T}},
G::Type{<:MOI.AbstractVectorFunction},
Expand Down
14 changes: 14 additions & 0 deletions test/Bridges/bridge_optimizer.jl
Original file line number Diff line number Diff line change
Expand Up @@ -567,6 +567,20 @@ function test_double_deletion_scalar()
# careful not to delete the second one twice, see https://github.com/jump-dev/MathOptInterface.jl/issues/1231
model =
MOI.instantiate(AffineOnlyModel{Float64}, with_bridge_type = Float64)
# If LessToGreaterBridge exists, this goes from:
# VariableIndex(x) in LessThan(1.0)
# ScalarAffineFunction(-1.0x) in GreaterThan(-1.0) (LessToGreater)
# ScalarAffineFunction(-1.0x) in Interval(-1.0, Inf) (GreaterToInterval)
# we want instead
# VariableIndex(x) in LessThan(1.0)
# VariableIndex(x) in Interval(-Inf, 1.0) (LessToInterval)
# ScalarAffineFunction(1.0x) in Interval(-Inf, 1.0) (FunctionConversion)
# To check that we handle the LessThan and Interval sets on the same
# VariableIndex correctly.
MOI.Bridges.remove_bridge(
model,
MOI.Bridges.Constraint.LessToGreaterBridge{Float64},
)
x = MOI.add_variable(model)
c = MOI.add_constraint(model, x, MOI.LessThan(1.0))
# Need to test the bridging to make sure it's not functionized first as otherwise,
Expand Down
Loading