Experiments for d3p paper
See the subdirectory README files for information for the individual experiments in the paper.
This repo relies on nivida-docker for providing clean working environment with GPU support.
Build and run the docker container:
docker build . -t d3p-experiments
nvidia-docker run -it --rm d3p-experiments
Follow potentially shown instructions to adjust available memory for the container.
By default, only the CPU backend for JAX is installed. To enable GPU support, inside the container run
./reinstall_jaxlib_cuda.sh 111
Note: We recommend using the docker container for running the code from this repository. If you
nevertheless want to run it without the container, please install the requirements listed in the
requirements.txt file using the pip install -r requirements.txt command. You will further
need pip, git, a working latex installation (for matplotlib plots) and CUDA if you plan to run on GPU.
You can consult the Dockerfile for the dependencies installed in the container, but keep in mind that
exact installation commands may vary with different OSes.
Code in this repository is licensed under the same CC-BY-NC-ND license as the associated paper (see LICENSE).