lp_cutter
This repo holds an object with methods for solving large instances of the Minimum Bisection LP relaxation by finding the optimal solution for a small subset of the constraints and then re-solving with incrementally more until all constraints are satisfied.
To run the code, clone the git repo and create the conda environment
in environment.yml. If you could use some reminders on how to create and start
conda environments, this
is a good resource. If that looks like gibberish to you, contact Sean directly,
and he'll help you get setup. Once you have the conda environment activated
(or your IDE pointed at it), the following files will be of interest to you.
If you would like to contribute, please make a branch off of dev and be sure
to augment the unit tests to account for your changes and that they all pass
before opening a pull request.
min_bisection.py
This module implements the above mentioned work. It has the following two public instance methods:
- MinBisect.solve_iteratively()
- MinBisect.solve_once()
The first method does as the opening sentence describes, re-solving the LP with the most violated subset of the constraints each iteration until all constraints are satisfied. The second method throws all constraints into the first solve of the model, returning the same solution. For more details on each, feel free to check out the doc strings. Other than constructing the object, there are no other methods a non-contributing user should need to interface with in this module.
runner.py
Solves once and iteratively the Min Bisection Problem for each combination of the provided values of parameters. This is used for kicking off batch solves and collecting their corresponding data.
graphs.ipynb
A jupyter notebook where run statistics are compiled and fed into graphs for comparison
batch_maker.py
Creates .pbs files that can be used to run multiple instances of runner.py
in parallel with torque on the coral server cluster. For more information on
torque, check out this resource.