Parse datasets stored in the SMPS file format.
Particular emphasis has been put in maintaining sparse representation of the data, to allow processing of large instances. Scenario matrices are generated lazily.
Requires gurobipy. Academic users can obtain free licenses here.
>> git clone https://github.com/robin-vjc/smps
>> cd smps/
>> pip install -e .
Optionally, run tests suite
>> py.test
SMPS is a file format for distributing Stochastic Multistage Optimization Models. It is an extension of the MPS file format, which is used to store optimization models.
There is a parser written in FORTRAN, but I was unable to make it work in my Python environment. Pyomo has some code to write SMPS files, but not to parse them (see also my request on their forum). I decided to write a parser script from scratch.
This library can be used to parse most models from the following archives:
- Parse stochastic model from smps dataset:
from smps.read import StochasticModel
sm = StochasticModel('./data/invest')
Expects the files './data/invest.cor', './data/invest.tim' and './data/invest.sto' to exist and follow the SMPS format.
- Plot scenario tree:
sm.plot_scenario_tree()
- Generate deterministic equivalent as a single Gurobi model.
det_eq = sm.generate_deterministic_equivalent()
det_eq.optimize()
-
Attributes in StochasticModel of particular interest
- StochasticModel.A:
A dictionary of sparse matrices; each entry of A is a sub-block of the nominal model matrix A. E.g., A[3,2] corresponds to the sub-block of A corresponding to the variables of the second period as they enter in the constraints of the third period of the problem.
- b, c, lb, ub, vtype:
Also dictionaries, one entry in the dictionary per stage, and each entry a vector with the nominal problem data.
- StochasticModel.scenario['SCENARIO_LABEL'].A:
modified matrix A according to scenario 'SCENARIO_LABEL'. Same way b, c, lb, ub and vtype can be accessed. These are generated lazily.
- StochasticModel.generate_deterministic_equivalent():
Returns a gurobi model for the deterministic equivalent of the problem. Only available for models of type scenarios discrete.
The parser does support SCENARIO DISCRETE types (which are by far the most commonly found in datasets archives);
it does not currently support BLOCK-INDEP mode.
Also, while standards are set in the SMPS documentation, certain models archives do not always follow them, which may cause parsing problems. Please report them as bugs in this case.
@article{Vujanic2018,
title={Dual Decomposition of Stochastic Integer Programs: New Results and Experimental Comparison of Solution Methods},
author={Vujanic, Robin and Esfahani, Peyman Mohajerin},
journal={TO BE COMPLETED},
volume={TO BE COMPLETED},
number={TO BE COMPLETED},
pages={TO BE COMPLETED},
year={2018},
publisher={TO BE COMPLETED}
}