GTDKeyformatter is not working

[varunagrawal]

If I print out the graph in test_dynamics_graph.py, I get the following which doesn't have the keys correctly formatted.

Factor 0: pose factor
  keys = { 31526292608253952 31527392119881728 32087047833387008 }
isotropic dim=6 sigma=0.001

Factor 1: pose factor
  keys = { 31527392119881728 31528491631509504 32087052128354304 }
isotropic dim=6 sigma=0.001

Factor 2: twist factor
  keys = { 24207943213776896 24209042725404672 32087047833387008 33494422716940288 }
isotropic dim=6 sigma=0.001

Factor 3: twist factor
  keys = { 24209042725404672 24210142237032448 32087052128354304 33494427011907584 }
isotropic dim=6 sigma=0.001

Factor 4: twist acceleration factor
  keys = { 24209042725404672 18296968702853120 18298068214480896 32087047833387008 33494422716940288 27583448206016512 }
isotropic dim=6 sigma=0.001

Factor 5: twist acceleration factor
  keys = { 24210142237032448 18298068214480896 18299167726108672 32087052128354304 33494427011907584 27583452500983808 }
isotropic dim=6 sigma=0.001

Factor 6: wrench factor
  keys = { 24207943213776896 18296968702853120 31526292608253952 19703248369745920 }
isotropic dim=6 sigma=0.001

Factor 7: wrench factor
  keys = { 24209042725404672 18298068214480896 31527392119881728 19704347881373696 19704352176340992 }
isotropic dim=6 sigma=0.001

Factor 8: wrench factor
  keys = { 24210142237032448 18299167726108672 31528491631509504 19705451687968768 }
isotropic dim=6 sigma=0.001

Factor 9: wrench equivalence factor
  keys = { 19703248369745920 19704347881373696 32087047833387008 }
isotropic dim=6 sigma=0.001

Factor 10: torque factor
  keys = { 19704347881373696 23924273508777984 }
isotropic dim=1 sigma=0.001

Factor 11: wrench equivalence factor
  keys = { 19704352176340992 19705451687968768 32087052128354304 }
isotropic dim=6 sigma=0.001

Factor 12: torque factor
  keys = { 19705451687968768 23924277803745280 }
isotropic dim=1 sigma=0.001

[yetongumich]

which function are you using? There's a wrapped function gtd.DynamicsGraph.printGraph() which uses the GTDKeyFormatter. Try it?

[varunagrawal]

I see. It should also be compatible with NonlinearFactorGraph::print and Values::print so that it is useful for debugging.

[yetongumich]

I see. It should also be compatible with NonlinearFactorGraph::print and Values::print so that it is useful for debugging.

I'm wondering how it is directly bounded with the default print() function in Python. Is there a way that I can specify the keyformatter when calling the Python print()?

[varunagrawal]

I guess look at the way how GTSAM does it? I believe it takes the key formatter as a default argument so we need to figure out how to achieve that here.

[gchenfc]

python implicit print (__repr__) will call the default print function:
https://github.com/borglab/wrap/blob/5ddaff8bab6c05e8a943c94993bf496e13296dd6/gtwrap/pybind_wrapper.py#L127-L150
overriding print like in e.g. CableLengthFactor:

GTDynamics/gtdynamics/cablerobot/factors/CableLengthFactor.h

Lines 78 to 83 in 5d22ebd

	void print(const std::string &s = "",
	const gtsam::KeyFormatter &keyFormatter =
	GTDKeyFormatter) const override {
	std::cout << s << "cable length factor" << std::endl;
	Base::print("", keyFormatter);
	}

you can use the gtdynamics formatter by default, but i guess you couldn't override nonlinearfactorgraph's print. Another annoyance is that calling NonlinearFactorGraph::print will then pass on its KeyFormatter (gtsam default key formatter) on to all its children so printing the graph will use the wrong formatter but printing out a single factor works. i.e.

****** print(fg) ******
 size: 1

Factor 0: cable length factor
  keys = { 32087052128354306 31527392119881730 }
  noise model: constrained sigmas[0];
  noise model: constrained mu[1000];


****** print(fg.at(0)) ******
 cable length factor
  keys = { q(1)2 p[1]2 }
  noise model: constrained sigmas[0];
  noise model: constrained mu[1000];

A couple potential solutions would be to:
a) make a static function print(const NonlinearFactorGraph &graph) etc
b) wrap GTDKeyFormatter

[varunagrawal]

Ah thanks for the reminder @gchenfc. Yeah the wrapper defines the __repr__ method which captures the output of .print and returns that, thus forming a bridge. This is because python's print function calls __repr__ (or __str__) on the object to print.

This is when I wish C++ supported protocol based inheritance.