IALS

Source code for the paper Influence-Augmented Local Simulators: a Scalable Solution for Fast Deep RL in Large Networked Systems by Miguel Suau, Jinke He, Matthijs Spaan, and Frans Oliehoek.

Requirements

Singularity

Installation

sudo singularity build IALS.sif IALS.def

This will create a singularity container and install all the required packages. Alternatively, you can create a virtual environment and install the packages listed in IALS.def

Running an experiment

Launch the singularity shell:

singularity shell --writable-tmpfs IALS.sif

To run a new experiment do:

cd runners
python experiment.py with ./configs/warehouse/IALS.yaml

This will train a new policy for the warehouse environment on IALS. To train on the global simulator change the config file path to ./configs/warehouse/global.yaml.