Generative Multi-Agent Behavioral Cloning (
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.

Generative Multi-Agent Behavioral Cloning


5/20/2018 - a new version of the code will be available soon (with PyTorch v0.4), stay tuned!


Code is written with PyTorch v0.2 (Python 3.6.1).

Dataset is available from STATS. A pre-processed version is available here. Download the data into the bball_data/data/ folder.

Example Run

You can edit and run that script, or run directly with command-line parameters.

$ ./
$ python -t 105 -b 10 -n 10
$ python -t 105 --animate

Trained models for RNN_GAUSS (101), VRNN_SINGLE (102), VRNN_INDEP (103), and our model MACRO_VRNN (104) are included.

Files contains the models. MACRO_VRNN is our model with macro-goals. contains the training process, and can be called with is used to sample rollouts from a trained model. is used to plot the samples as well as animate them (with --animate flag). --showmacro will display macro-goals, where applicable. contains functions for sampling and calculating various losses.

bball_data/ contains the Dataset object.

bball_data/ contains constants for the data.

bball_data/ is the script used to extract macro-goals. Don't need to run again.

bball_data/ contains the functions for plotting and animating.