NOTICE: This repository is currently in pre-alpha testing and should not be used in production for any purpose, as it is subject to change rapidly and without backwards compatibility. There are additional dependencies that have not yet been made public as well.
This code is currently provided mainly for research purposes and ongoing work with collaborators.
This repository uses inverse reinforcement learning to rationalize the full day utility function for individual economic agents based on multiple observations of their stay sequence trajectories.
It is assumed that you are have a method to extract daily activity-travel sequences. Place traces for individual agents into a directory.
Compatible w/ Python 2 and 3.
The recommended and only currently supported installation method assumes you have the anaconda python scientific computing framework installed on your machine. Create a new environment for this project and then do:
pip install -r requirements.txt
To execute the test script
python scripts/run_atp_experiment.py --config=data/misc/IRL_multimodal_scenario_params.json --traces_dir=<TRACES> --seed=1
where <TRACES> is a directory of persona trace files each ending in .csv.
You may use the following BibTeX entry to cite this work in a publication (TO BE UPDATED):
@article{Feygin2017dairl,
author = {{Feygin}, Sid and {Pozdnukhov}, Alexei},
title = "{DA-IRL: Structural Estimation of Full Day Activity-Travel Behavior Models}",
journal = {arXiv preprint arXiv:XXXX.XXXXXXX},
year = 2017,
url={https://arxiv.org/pdf/XXXX.XXXXX.pdf}
}