Enable multiple edges in the same direction between two nodes for pandas

[jsfreischuetz]

Enable multiple edges in the same direction between two nodes

Description

Currently the solver does not support multiple edges of between two nodes in the same direction. This draft PR shows a quick way of enabling this feature in Pandas. This requires changing some minor implementation details about the way that indices are used in the DataFrame when converting.

fixes #155

For networkx, it would require a type change to MultiDiGraph, and for scipy, it may not be trivially possible to implement these changes.

This PR draft is mostly intended as a jumping off point for showing it is possible to solve this, rather then a general solution.

Checklist

• Implementation:
  • Implementation of the Mod in the gurobi_optimods installable package
  • Tests for the Mod implementation in tests/
  • Docstrings for public API, correctly linked using sphinx-autodoc
• Documentation page:
  • Background and problem specification
  • Example of the input data format (use gurobi_optimods.datasets for loading data)
  • Runnable code example
  • Presentation of solutions
  • Included in the mod gallery and toctree

[CLAassistant]

All committers have signed the CLA.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[jsfreischuetz]

@simonbowly Would it be possible to trigger the test cases. They are working locally, but I want to make sure that everything is working properly.

[simonbowly]

@jsfreischuetz thanks for the work! I re-ran the tests. The issue I have is that this changes the expected input/output data format. To make sure the PR doesn't make backwards-incompatible changes, could you please revert the original tests and add some new tests for cases with duplicate edges (and their expected results)?

This should still be do-able, pandas can handle duplicate values in the index.

[jsfreischuetz]

@jsfreischuetz thanks for the work! I re-ran the tests. The issue I have is that this changes the expected input/output data format. To make sure the PR doesn't make backwards-incompatible changes, could you please revert the original tests and add some new tests for cases with duplicate edges (and their expected results)?

This should still be do-able, pandas can handle duplicate values in the index.

The current format returns only identified the edge with a tuple of source destination for each flow. This makes it ambiguous which edge was taken. It can be reconstructed by the user by making a few assumptions about the flows.

I will make the change leaving the format, but for ease of use for users, eventually the output format should likely be modified.

[simonbowly]

The current format returns only identified the edge with a tuple of source destination for each flow. This makes it ambiguous which edge was taken. It can be reconstructed by the user by making a few assumptions about the flows.

Good point, it might need a re-think.

@torressa do you have any thoughts on this one?