The pp.inference API is a mess. All but two of the dozen or so functions available were used for experimentation/troubleshooting and are irrelevant for the paper, so it might be a good idea to scrap stuff for clarity.
For example, there's "gradient-descent to get MLE beta" code and an entire "hardmax/" folder, which is superfluous now that the "softmax" maxent inference has been removed from the repo.
API rework route:
Maybe: pp.inference.hardmax => pp.inference
Get rid of all the inference tools except those we actually use in Andrea's crazy fly repo.
Submit PRs to all the repos that depend on this to install from pypi instead so they can use the legacy interface (pip install pedestrian-prediction == 0.1)
Maybe write a note on the README.md for people who are referencing crazy fly