Example Datasets with Python Package

[varunagrawal]

Updated the setup.py to include the example data files and datasets. Resolves #500.

Note that we now install the data to a common "gtsam_example_data" directory, since we need CMake to know the path of the directory when building the .so file. This is the only way I've found to break this circular dependency (setup.py depends on .so and .so depends on where setup.py installs the data), so any better suggestions or recommendations are welcome!

[ProfFan]

I think this would not work for wheels. I propose we have a Python version of this function, which uses import site to get the folder

[varunagrawal]

I actually checked that and it works with the wheel format. I am double checking right now.

[varunagrawal]

Yup, on running python setup.py bdist_wheel, I see output for adding the data files to the binary. 🎉

[ProfFan]

I mean we should not touch anything outside of the site-packages directory, per the standard guidelines for python packages.

[dellaert]

site-packages directory

Indeed, @varunagrawal, does the code now refer to the site-packages? That would be powerful magic, and would be the only way it could work for people that use "pip install gtsam", right?

[varunagrawal]

It doesn't quite unfortunately, because the C++ function findExampleDataFile has no idea what the site-packages directory can be at compile time.

I have seen packages like nltk be a bit more relaxed when it comes to packaging data (they download the corpus data into the home directory) but I agree with @ProfFan that it's good to stick with standards to avoid code rot (amongst other things).

[varunagrawal]

I'll add the custom function in a bit.

[dellaert]

@varunagrawal Yeah let's not be relaxed. If there is no c++ way, is there a python way as Fan suggested?

[varunagrawal]

@dellaert there is no C++ way without some crazy magic due to the circular dependency (and even then it could break).

I am creating the python version. Building the wheel as of this comment.

[varunagrawal]

@dellaert @ProfFan all done!

[dellaert]

Cool!!