MIPLIBing

MIPLIBing is a library for downloading MIP benchmark instances on demand.

Usage

Install the library using python. Python 3 is required.

$ python setup.py install

You can now download MIP benchmark data and instances in Python scripts.

from MIPLIBing import MIPLIBing
from MIPLIBing import Libraries
from MIPLIBing import Status

mip = MIPLIBing(library = Libraries.MIPLIB2017_Collection, verbose = True)
instances = mip.get_instances(with_status = Status.open, max_var=200)
print()
for instance in instances:
    print(instance)

The script above may produce the following output:

Instance pb-market-split8-70-4: Already downloaded
Instance v150d30-2hopcds: Downloading from http://miplib.zib.de/WebData/instances/v150d30-2hopcds.mps.gz

Instance:              	pb-market-split8-70-4
Type:                  	MILP
Local path:            	MIPLIBing_cache/MIPLIB2017_Collection/pb-market-split8-70-4.mps
Feasible:              	Unknown
Primal Bound:          	None
Dual Bound:            	None
Status:                	open
Variables:             	71	(71 binary)	 (0 integer)	 (0 continuous)
Constraints:           	17
Non-zeroes:            	1113


Instance:              	v150d30-2hopcds
Type:                  	MILP
Local path:            	MIPLIBing_cache/MIPLIB2017_Collection/v150d30-2hopcds.mps
Feasible:              	Yes
Primal Bound:          	41.0
Dual Bound:            	None
Status:                	open
Variables:             	150	(150 binary)	 (0 integer)	 (0 continuous)
Constraints:           	7822
Non-zeroes:            	103991

The library parameter can be any of:

• Libraries.MIPLIB2017_Benchmark
• Libraries.MIPLIB2017_Collection
• Libraries.MINLPLIB
• Libraries.QPLIB

According to the library chosen (see observations below), the with_status (or without_status) parameter can be any of:

• Status.easy
• Status.hard
• Status.open
• Status.closed

When instance is printed, all relevant information is displayed for the type of library used. The location of the downloaded instance in your machine is given by instance.path.

Here is the complete list of arguments for the MIPLIBing constructor:

Argument	Description	Default value	Observations
library	Problem library to be queried	Libraries.MIPLIB2017_Benchmark	Value should be in Libraries.
update_csv	If CSV file summarizing problem data should be updated, in case one already exists	False
verbose	Prints the steps involved when getting instances	False	If the chosen file_extension is not available for an instance, you will see the warnings with verbose.
local_directory	Define a local (or global) directory for caching data about the problem library and the instances	"MIPLIBing_cache"
file_extension		None	Cannot be used if library is Libraries.MIPLIB2017_Benchmark or Libraries.MIPLIB2017_Collection (the file extension is "mps" in that case). The default value is set to "gms" if library is Libraries.MINLPLIB and "qplib" is library is Libraries.QPLIB. When the file format is not available for some instance, a warning is shown if verbose is True and the local path of the instance has value None.

For the get_instances method, every argument has default value None and is only applicable if changed to another value. Here is the complete list of arguments:

Argument	Description	Libraries supported	Observations
instance_name	Exact name of instance	All
min_var	Minimum number of decision variables	All
max_var	Maximum number of decision variables	All
min_bin	Minimum number of binary decision variables	All
max_bin	Maximum number of binary decision variables	All
min_int	Minimum number of integer and non-binary decision variables	All
max_int	Maximum number of integer and non-binary decision variables	All
min_cont	Minimum number of continuous variables	All
max_cont	Maximum number of continuous variables	All
min_cons	Minimum number of constraints	All
max_cons	Maximum number of constraints	All
min_nz	Minimum number of nonzeroes	All
max_nz	Maximum number of nonzeroes	All
with_status	Only a specific status is allowed	MIPLIB2017_Benchmark, MIPLIB2017_Collection, MINLPLIB	Value should be in Status. If library is Libraries.MIPLIB2017_Benchmark or Libraries.MIPLIB2017_Collection, with_status can be Status.easy, Status.hard, or Status.open. If library is Libraries.MINLPLIB, with_status can be Status.closed (if 3 solvers found optimal solution or proven infeasibility) or Status.open (othewise).
without_status	Only a specific status is forbidden	MIPLIB2017_Benchmark, MIPLIB2017_Collection, MINLPLIB	Value should be in Status. If library is Libraries.MIPLIB2017_Benchmark or Libraries.MIPLIB2017_Collection, without_status can be Status.easy, Status.hard, or Status.open. If library is Libraries.MINLPLIB, without_status can be Status.closed (if 3 solvers found optimal solution or proven infeasibility) or Status.open (othewise).
tags	A list of tags to query the MIPLIB2017 Collections	MIPLIB2017_Benchmark and MIPLIB2017_Collection	Must be a list of strings. Each entry forms a literal of an "and" query. An "or" query is performed using "
min_sos	Minimum number of SOS constraints	MINLPLIB
max_sos	Maximum number of SOS constraints	MINLPLIB
min_semi	Minimum number of semicontinuity / semiintegrality constraints	MINLPLIB
max_semi	Maximum number of semicontinuity / semiintegrality constraints	MINLPLIB
problem_type	Only one type of problem is allowed	MINLPLIB	The problem type should be given as a string such as "MBNLP", "QP", "MBQP", "NLP", etc.
min_obj_density	Minimum objective density (%) in the quadratic part of the objective matrix	QPLIB
max_obj_density	Maximum objective density (%) in the quadratic part of the objective matrix	QPLIB
min_problematic_ev_density	Minimum density (%) of problematic eigenvalues in the quadratic part of the objective matrix	QPLIB
max_problematic_ev_density	Maximum density (%) of problematic eigenvalues in the quadratic part of the objective matrix	QPLIB
min_quadratic_cons	Minimum number of quadratic constraints	QPLIB
max_quadratic_cons	Maximum number of quadratic constraints	QPLIB
objective_type	Only one type of objective is allowed	QPLIB	The types should be given as a string among "L", "D", "C", and "Q" as described in the documentation for QPLIB Problem Type.
variables_type	Only one type of variables is allowed	QPLIB	The types should be given as a string among "C", "B", "M", "I", and "G" as described in the documentation for QPLIB Problem Type.
constraints_type	Only one type of constraints is allowed	QPLIB	The types should be given as a string among "N", "B", "L", "D", "C", and "Q" as described in the documentation for QPLIB Problem Type.

Citation

A manuscript that describes and contextualizes MIPLIBing is currently under review:

@unpublished{MIPLIBing,
    author = {Thiago Serra and Ryan J. O'Neil},
    title = {MIPLIBing: Seamless benchmarking of mathematical optimization problems and metadata extensions},
    year = {2020}
}

License

MIPLIBing is distributed under the MIT license.