Convert Gambit bindings to Python 3, make conda package

[jstac]

Gambit is a useful piece of software for computing Nash equilibria in games, among other things.

http://www.gambit-project.org/gambit14/index.html#

The core routines are written in C. There are Python bindings but they are in Python 2 and the developer has no plans to convert to Python 3. If someone is interested it would valuable to convert to Python 3 and build a conda package. A start would be to run the bindings through the 2to3 conversion tool and see how much manual work is left. If not much I think this is worth doing.

[oyamad]

I have some code on game theory (mostly for simulation of learning/evolutionary dynamics models, and not about computation of Nash equilibria) to submit to QE.py (the base code is ready, while applications are not yet): https://github.com/oyamad/game_theory_models

(I also have a preliminary version in Julia.)

There I have a certain data structure to represent a game, and would be interested in building a wrapper/binding in Python (and Julia) that converts my data structure and passes it to Gambit to return Nash equilibria of the game.

(Just to remind ourselves, Gambit is GPL licensed.)

[jstac]

The game theory code looks very promising.

I didn't notice the GPL license. They might be willing to change the license if we are prepared to update the Python bindings to Python 3 and build a conda package. I know Andy McLennan well and he is very reasonable.

[sglyon]

The GPL wouldn't be an issue for us in quantecon though.

The wrapper would probably need to be gpl, but we would be able to call it from a BSD qe.py

[oyamad]

I opened a PR #220, just for the base classes this time.