Add model implementations API for python

[eerolinna]

This depends on #127 and should not be merged before it

This PR is mostly to make the discussion in #125 more concrete. If we reach the conclusion that this functionality should not be added then that's how it will be. However at least currently I think that this would be a good addition.

Add method implementations to Model, which takes a framework name and returns a model implementation object for that framework.

Currently model implementation objects only have a load method. Calling it loads the underlying model of the framework being used.

Here's an example how this might be used. (running these needs PR #129)

from posteriordb import PosteriorDatabase
my_pdb = PosteriorDatabase("../posterior_database/") # replace path if needed
model = my_pdb.model("eight_schools_centered")
data = my_pdb.data("eight_schools")

We can load a Stan model and obtain posterior draws

stan_imp = model.implementation("stan")
stan_model = stan_imp.load()
stan_draws = stan_model.sampling(data=data.values())

We can do the same with a PyMC3 model

import pymc3
pymc_imp = model.implementation("pymc3")
get_pymc_model = pymc_imp.load()
pymc_model = get_pymc_model(data.values())
pymc_draws = pymc3.sampling.sample(model=pymc_model)

Further down the line we might want to also add other methods to the model implementation objects. Obtaining prior or predictive draws might be one of these potential new methods.

[eerolinna]

The travis build timed out. For some reason the R CI checks seem to work really slowly right now, the python checks passed.

[MansMeg]

Sure. Ill take a look at this asap. Im now focused on the ICML deadline so I have some other stuff higher up in my todolist.

[eerolinna]

That's fine, thanks for letting me know

[eerolinna]

Any comments on this now? The ICML deadline has passed, right?

[MansMeg]

Yes. Got some other admin to do. This is now high on my todo.