A package to read linear optimization problems in MPS format and quadratic optimization problems in QPS format.
If you use QPSReader.jl in your work, please cite using the format given in CITATION.bib.
The problems represented by the QPS format have the form
optimize c₀ + cᵀ x + ½ xᵀ Q x subject to L ≤ Ax ≤ U and ℓ ≤ x ≤ u,
where:
- "optimize" means either "minimize" or "maximize"
c₀∈ ℝ is a constant term,c∈ ℝⁿ is the linear term,Q = Qᵀis the n×n quadratic term,Ais the m×n constraint matrix,L,Uare constraint lower and upper bounds, respectivelyℓ,uare variable lower and upper bounds, respectively
Mixed-integer problems are supported, but semi-continuous and semi-integer variables are not.
julia> ]
pkg> add QPSReaderThis package exports the QPSData data type and the readqps() function.
Because MPS is a subset of QPS, the readqps() function accepts both formats.
Because the SIF is a superset of QPS, QPS problems implemented as SIF files (such as those from the Maros-Meszaros collection) are also supported.
Both fixed and free format are supported (see below for format conventions). To read a problem from a file:
julia> qps = readqps("Q25FV47.QPS") # Free MPS is used by default
julia> qps = readqps("Q25FV47.QPS", mpsformat=:fixed) # uses fixed MPS format
julia> qps = readqps("Q25FV47.QPS", mpsformat=:free) # uses free MPS formatreadqps also accepts an IO object as the first argument.
By default, a number of messages may be logged while reading. Log output can be suppressed as follows:
using QPSReader
using Logging
qps = with_logger(Logging.NullLogger()) do
readqps("Q25FV47.QPS")
endThe QPSData data type is defined as follows:
mutable struct QPSData
nvar::Int # number of variables
ncon::Int # number of constraints
objsense::Symbol # :min, :max or :notset
c0::Float64 # constant term in objective
c::Vector{Float64} # linear term in objective
# Quadratic objective, in COO format
qrows::Vector{Int}
qcols::Vector{Int}
qvals::Vector{Float64}
# Constraint matrix, in COO format
arows::Vector{Int}
acols::Vector{Int}
avals::Vector{Float64}
lcon::Vector{Float64} # constraints lower bounds
ucon::Vector{Float64} # constraints upper bounds
lvar::Vector{Float64} # variables lower bounds
uvar::Vector{Float64} # variables upper bounds
name::Union{Nothing, String} # problem name
objname::Union{Nothing, String} # objective function name
rhsname::Union{Nothing, String} # Name of RHS field
bndname::Union{Nothing, String} # Name of BOUNDS field
rngname::Union{Nothing, String} # Name of RANGES field
varnames::Vector{String} # variable names, aka column names
connames::Vector{String} # constraint names, aka row names
# name => index mapping for variables
# Variables are indexed from 1 and onwards
varindices::Dict{String, Int}
# name => index mapping for constraints
# Constraints are indexed from 1 and onwards
# Recorded objective function has index 0
# Rim objective rows have index -1
conindices::Dict{String, Int}
# Variable types
# `VTYPE_Continuous` <--> continuous
# `VTYPE_Integer` <--> integer
# `VTYPE_Binary` <--> binary
# `VTYPE_SemiContinuous` <--> semi-continuous (not supported)
# `VTYPE_SemiInteger` <--> semi-integer (not supported)
vartypes::Vector{VariableType}
# Indicates the sense of each row:
# `RTYPE_Objective` <--> objective row (`'N'`)
# `RTYPE_EqualTo` <--> equality constraint (`'E'`)
# `RTYPE_LessThan` <--> less-than constraint (`'L'`)
# `RTYPE_GreaterThan` <--> greater-than constraint (`'G'`)
contypes::Vector{RowType}
endRows and variables are indexed in the order in which they are read. The matrix Q is zero when reading an MPS file.
The file formats supported by QPSReader are described here:
The following conventions are enforced:
-
Multiple objective rows
- The first
N-type row encountered in theROWSsection is recorded as the objective function, and its name is stored inobjname. - If an additional
N-type row is present, awarning-level log is displayed. SubsequentN-type rows are ignored. - Each time a rim objective row is encountered in the
COLUMNSorRHSsection, the corresponding coefficients are skipped, and anerror-level log is displayed.
- The first
-
Multiple RHS / Range / Bound fields
- The second field of the first card in the
RHSsection determines the name of the right-hand side, which is stored inrhsname. Same goes for theRANGESandBOUNDSsections, with the corresponding names being stored inrngnameandbndname, respectively. - Any card whose second field does not match the expected name is then ignored.
A
warning-level log is displayed at the first such occurence. - In addition, any line or individual coefficient that is ignored triggers an
error-level log.
- The second field of the first card in the
- Default bounds for variables are
[0, Inf), to exception of integer variables (see below). - If multiple bounds are specified for a given variable, only the most recent bound is recorded.
There are two ways of declaring integer variables:
-
Through markers in the
COLUMNSsection. -
By specifying
BV,LIorUIbounds in theBOUNDSsection -
The convention for integer variable bounds in as follows:
Marker? BOUNDSfieldsType Bounds reported Yes - Integer [0, 1]Yes BVBinary [0, 1]Yes ( LI,l)Integer [l, Inf]Yes ( UI,u) withu≥0Integer [0, u]Yes ( UI,u) withu<0Integer [-Inf, u]Yes ( LI,l) + (UI,u)Integer [l, u]No BVBinary [0, 1]No ( LI,l)Integer [l, Inf]No ( UI,u) withu≥0Integer [0, u]No ( UI,u) withu<0Integer [-Inf, u]No ( LI,l) + (UI,u)Integer [l, u]The
LI/UIcan be replaced byLO/UPin the table above, with no impact on bounds. Only the integrality of variables are affected. For continuous variables, follow the second half of the table, and replaceLI/UIbyLO/UP.
- A row (resp. column) name that was not declared in the
ROWS(resp.COLUMNS) section, appears elsewhere in the file. The only case where an error is not thrown is if said un-declared row or column appears in a rim line that is skipped. - An
N-type row appears in theRANGESsection
- The Netlib LPs: original Netlib site | in SIF format | as tar files (incl. preprocessed versions)
- the Kennington LPs: original Netlib site
- infeasible Netlib LPs: original Netlib site
- the Maros-Meszaros QPs: in QPS format | in SIF format
Both the Netlib LP and Maros-Meszaros QP collections are provided as Julia artifacts (requires Julia 1.3).
This package exports the fetch_netlib and fetch_mm functions that return the path to the Netlib and Maros-Meszaros collections, repectively
using QPSReader
netlib_path = fetch_netlib()
mm_path = fetch_mm()
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent