an example with algorithmic differentiation

[ev-br]

Ceres docs advocate for using algorithmic differentiation instead of finite differences. Let's make one more IPy notebook example with the jacobian calculated using algopy, https://pythonhosted.org/algopy/

[nmayorov]

It rather seems like example for algopy than scipy.optimize. I mean what will it explain in usage of least_squares? But if you think it's interesting and important, the simplest idea might be extending bundle adjustment example and compare results.

Unfortunately it doesn't support sparse Jacobians according to potential improvements: "support for sparse Jacobian and sparse Hessian computations using graph coloring as explained in http://portal.acm.org/citation.cfm?id=1096222"

So please suggest some problem where it would make sense to use automatic differentiation (preferably with bounds).

[ev-br]

Hmmm, yeah, that's too bad. It'd be a good enhancement in there. An ideal example would be something where the errors from finite differences do play a role. I'll try thinking about it.