Remove dependence on matplotlib

[jstac]

At least some modules pull in matplotlib, such as this one:

https://github.com/QuantEcon/QuantEcon.py/blob/master/quantecon/arma.py

Let's get rid of this dependency if possible. One way to do this when we have functions that generate plots is to have them take an axes instance as an argument and write to it.

(This could be added to the contributions style guide.)

[davidrpugh]

@jstac I agree with the sentiment, but worry that writing functions and methods that depend on an axes instance leaves us with an implicit dependency on matplotlib rather than the explicit dependency that we have now.

How much code is actually using plotting functionality? There are so many good plotting libraries out there and given that preferences over optimal plotting libraries are quite idiosyncratic (i.e., there is no dominant visualization library) perhaps we should consider removing all plotting code.

Code we have now could be moved to examples that demonstrate usage of the core library. This would leave us with less code to maintain in the long run and leave the choice of data viz tooling entirely to the end user.

[jstac]

@davidrpugh That's probably a good idea. I just ran a search and only arma.py is importing Matplotlib. If there's no objection, then I'll pull it out and put it in QuantEcon.applications.

Thanks ;-)

[cc7768]

Was just thinking about this -- Long story which I'll generate another thread for. It doesn't look like we ever removed this dependency. Are people still in favor of removing this dep? (I am)

[jstac]

@cc7768 Yes, definitely, I just haven't had time...

[davidrpugh]

Agreed!

[oyamad]

Done by #330 and #331.

[natashawatkins]

Seems we can close this @mmcky? matplotlib only returns files in the docs and something related to sphinxext

[mmcky]

indeed. Thanks @natashawatkins