UPDATE: There is a newer and much faster version of this codebase implemented in PyTorch! Take a look here!
This repository contains the code for Generative Modeling of Multimodal Multi-Human Behavior by Boris Ivanovic, Edward Schmerling, Karen Leung, and Marco Pavone.
Note: We use Git LFS to version large files (such as model checkpoints and data).
First, we'll create a conda environment to hold the dependencies.
conda create --name modeling python=2.7 -y
source activate modeling
pip install -r requirements.txt
Then, since this project uses IPython notebooks, we'll install this conda environment as a kernel.
python -m ipykernel install --user --name modeling --display-name "Python 2.7 (NHumanModeling)"
Now, you can start a Jupyter session and view/run all the notebooks with
jupyter notebook
When you're done, don't forget to deactivate the conda environment with
source deactivate
The preprocessed datasets are available in this repository, under data/ folders (e.g. nba-dataset/data/).
If you want the original traffic weaving or NBA datasets, I obtained them from here: Traffic Weaving Dataset and NBA Dataset.