Benchmarks that may be used in the evaluation
This page lets you download a preliminary set of benchmarks
that may be used during the evaluation. Currently, only linear
pseudo-Boolean instances are available.
Each benchmark that was initially named
file has been normalized and
renamed to normalized-opb-file.
Comments in the original files were preserved as much as possible.
Comments indicating the number of variables and constraints, as well
as the initial file name and the variables renaming have been prepended.
We insist on the fact that these files are only a preliminary set
of benchmarks that is only made available so that you can test your
solvers. Some benchmarks will be removed from the evalution because
they are too easy, or duplicates of another benchmark, or incorrectly
translated, etc. Besides, the evaluation will use other benchmarks
which are not yet available.
These benchmarks come from different sources
- Submitted benchmarks
These are the instances which were directly submitted to the evaluation
- Benchmarks obtained on the web
The comments in the normalized file (as well as the path to the file)
should give you a good indication on where it comes from.
- Benchmarks translated from MPS files gathered on the web
These benchmarks are a translation of linear programming problems
available on the web and expressed in MPS format. These benchmarks are
named normalized-mps-vX-YY-ZZ-file where
X is the version of the translator, YY is the maximum number
of bits used for the representation of the integer part of a variable,
ZZ is the number of bits used for the representation of the
fractional part of a variable and file is the name of the initial
MPS file. Note that the translation process hasn't been thoroughly tested.
Therefore, the semantics of these files might be different from the
initial MPS file. Furthermore, the limitation on the number of
bits used to represent a real variable imposes constraints which
are not initially expressed in the MPS file. Therefore, the set of
solutions is likely to be different. At last, the translation process
tends to generate huge coefficients that can exceed the limits of a 32
bits integer.
For each instance file that does not use only small integers, a new
instance will be generated with reduced coefficients. These instances
are named normalized-reduced-file.
Reduced coefficients are obtained by dividing all numbers in a
constraint by a same number so that the resulting constraint fits in
category SMALLINT. This ensures that we will have more instances in the
SMALLINT category than in any other ones. Keep in mind however that a
reduced constraint is not equivalent to the initial one. Therefore
normalized-reduced-file does not have the same semantics as
file.
Linear instances used in the PB06 evaluation are available here.