Python print Values and NonlinearFactorGraph

[gchenfc]

This fixes #131 by:

1. creating a static gtdynamics::str(obj, s) function that calls obj.print(s, GTDKeyFormatter)
2. creating pure-python "wrapper" classes around gtsam.Values and gtsam.NonlinearFactorGraph that override repr. Based on this SO answer. Now use gtd.Values instead of gtsam.Values - it should behave identically in every way except it will print using GTDKeyFormatter instead.

fg = gtd.NonlinearFactorGraph()
fg.push_back(gtd.MinTorqueFactor(gtd.internal.TorqueKey(0, 0).key(),
                                 gtsam.noiseModel.Unit.Create(1)))
print(fg)

will output:

size: 1

Factor 0: min torque factor
  keys = { T(0)0 }
  noise model: unit (1) 

I think I have also come up with a way make the GTDKeyFormatter function available in python (i.e. kf = gtd.GetKeyFormatter; fg.print_('fg', kf) and also supports functionals, but it will take some changes to wrap. i'll put that (unit tests + one-liner) in a different PR after wrap PR.

[varunagrawal]

This is ingenious!! But very hacky :(

I would prefer to

1. Have NonlinearFactorGraph and Values overload operator<< so we can directly do print(fg).
2. Update the print methods in GTSAM to use the key formatters assigned to each factor. So unless the user passes in a KeyFormatter, it should use the factor's default KeyFormatter which would be gtsam::DefaultKeyFormatter or GTDKeyformatter depending on the factor type.
3. Add functional support to wrap so we can pass in functions and lambdas. This is easy for me to do but I am swamped by other stuff.

[dellaert]

There might be more substantial changes to gtsam that could make this better, I agree with varun on that, but I think it’s good for now. One thing I ask is that you add a README file in the python folder that explains this and (later) other peculiarities of the wrapper