charjon / geco Goto Github PK

should contain:

• Description of the project.
• Installation.
• Simple example.

And make it compliant.

This paper generates MIPs that share a common 'backbone' (substructure). Write a prototype CommonSubstructureGenerator in the generator.py to extend our generator framework.

There should be some testcases that ensure different seeds are creating different instances.

Or delete it all together, as the pypi package will make this obsolete?

gamsworld. Unfortunately the downloads from the side itself are in .gdx format, which need a plugin to be readable with scip. Not sure how to solve this problem yet, as sometime links to the original source in other formats are outdated, see e.g. this one.

The list should group by instance type on the top level and also mention all papers (we are aware of) doing experiments on instances of this type.

And make it compliant. Should be quit similar to set covering.

Shuffling here means shuffling rows and columns of the MIP.

A function to easily get all (easy, hard, open) instances that are (or are not) part of the benchmark set would be nice to have.

Add a small example to the README on how to save models to disk (maybe in different file formats).

The assertions should be of the form assert (seed1 == seed2 and param1 == param2) or (seed1 != seed2 and param1 != param2)

Add method to translate set-cover problems from or-lib to our parameters for set-cover formulations.

There should be some testcases that ensure different seeds are creating different instances.

Networkx decorator might be usefull. See e.g. here.

By origin I mean the paper.

Currently we are using the networkx decorator for randomness.
The design is tailored to not require numpy, but as we build on numpy anyways, we could write our own decorator.
Possible benefits:

• Easier to understand decorator, as we can keep the code way more compact than numpys implementation
• Better performance and more functions, as e.g. arrays of random elements can be created at once.

Link to numpy random generators.

I.a. take care of TODOs

Create input for #27 as in this paper.

Extended test cases needed.

For future reference: https://docs.conda.io/projects/conda/en/latest/user-guide/tasks/use-conda-with-travis-ci.html

Create input for #27 following VII.A of this paper. Check if networkx supports those graphs already.

For this the parameter generator needs to be split from the mip generator.

The example code in the README did not work:

from geco import knapsack
knapsack.yang_instance(n=5, seed=1)

For this we would have to change our init.py files to the same style networkx is using.

Shall we?

For solved problems the MIPLIB also offers solution files, that can be downloaded.
For our MIPLIB module we could add a function that downloads the (all) solution files, given an instance name.

And make it compliant.

https://miplib.zib.de/

The current test_heinz_formulation raises errors when counting the number of variables. This might be a time_step related problem. This needs to be fixed and the @pytest.mark.xfail should be removed afterwards.

Add a CONTRIBUTING.md file that contains:

• How to create a generator and where to function(s).
• Refer to development standards or merge this file's contents with [develop.md].
• Refer to some guide on contribution cycle on github (branch-implement-pull request-review-merge).

There should be some testcases that ensure different seeds are creating different instances.

Input is a nx.Graph. For MIP formulation see e.g. here.

Some instances did rotate out of the current miplib completely. To reproduce experiments of older papers it would be nice to be able do download instances from 2010 and 2003.

Add method to translate capacitated warehouse location problems from or-lib to our parameters for the existing formulation.

There should be some testcases that ensure different seeds are creating different instances.

http://people.brunel.ac.uk/~mastjjb/jeb/info.html

As in https://docs.python.org/3/library/typing.html.

quiet as in they don't output anything when optimize() is run on them.