For a synopsis of the project and a presentation of the results, view the project's presentation poster.
If you have not yet installed conda, just run make, make sure *.sh are executable (i.e. chmod +x ...). The script will first try to install anaconda for your platform
(only MacOS and Linux supported at the moment) and then creates the conda environment and installs necessary dependencies.
If you already have anaconda installed, just run make environment.
Please note that it will try to detect if you have cuda installed and downloads the cpu-only version if not.
As our work was mostly separated into two parts, you can find further instructions on how to re-run our experiments in the respective folders, nri and graph.
To run a program on a specific GPU only, execute
CUDA_VISIBLE_DEVICES=2 python train.py or CUDA_VISIBLE_DEVICES=2,3 jupyter notebook