Python: Port WebPPL code to Python, preparing for later use with Pytorch and Pyro or Gen

[martinkunev]

See inline comments in WebPPL code header for details.

Please try to prepare for later multidimensional aspirations by storing all aspiration values not as intervals (as it is the case currently in WebPPL) but as d-dimensional boxes encoded by a 2-by-d tuple ((lo_1,lo_2,...,lo_d), (hi_1,...,hi_d)), even though at the moment we only support d=1.

Dependency:

• WebPPL code basically complete (WAIT for notification!)

It's still unclear whether we want to later use Pyro or Gen as a PPL framework. These resources might help to prepare for Pyro:

• https://forum.pyro.ai/t/is-there-an-equivalent-pyro-construct-for-webppls-infer-operator/612
• http://pyro.ai/examples/RSA-implicature.html
• https://probmods.org/

We would only use Gen if it is ported from Julia to Python in time. Watch this page: https://github.com/probcomp

[mensch72]

Hi Martin, do you want to reuse this empty repo for your python version? That's fine I think, but maybe then we should maybe rename the repo into something like "pysatisfia" or so, to make clear that it is the python version.