Code for SGM-PINN, DAC 2024 on Modulus 22.09 This repository contains only additional files and files that were edited from the original. This project includes source code provided by Nvidia. This project also includes code from https://github.com/Feng-Research/SPADE/tree/main and https://github.com/Feng-Research/HyperEF

Interact within a docker environment based on the 22.09 container. Full requirements and install guide for the container is here with abbreviated instructions below.

After following the installation below, from within the container you can navigate to cd /sgm-examples/ and run any of the .py files within using python <file>.py

To view the results use tensorboard from outside the container. E.g. tensorboard --logdir=./sgm-examples/ldc --port=7007 and open a web browser at localhost:7007. Scripts labelled tb_...sh are provided.

1. Ensure the docker engine in installed.
2. Install the baseline image

• sudo apt-get install nvidia-docker2
• docker pull nvcr.io/nvidia/modulus/modulus:22.09

3. Clone this repository to a working directory of your choice.
4. Open a terminal in the working directory.
5. Create and enter a persistent container to install the remaining requirements, also mounts the folders sgm-examples and sgm-modulus from the current directory.

    docker run --shm-size=1g --ulimit memlock=-1 --ulimit stack=67108864 \
    --runtime nvidia -v ${PWD}/sgm-examples:/sgm-examples \
    -v ${PWD}/sgm-modulus:/sgm-modulus \
    -it nvcr.io/nvidia/modulus/modulus:22.09 bash

• If this container is stopped or closed you can find it with docker container ls -a, then start or stop based on its NAME (tab to auto-complete). Then re-enter with docker exec -w /sgm-examples/ -it <CONTAINER_ID> /bin/bash.

6. Inside the container, install Julia and required packages
   • curl -fsSL https://install.julialang.org | sh and follow a standard install
   • Run . /root/.bashrc as indicated
   • Enter the Julia REPL, and press ] to enter the pkg mode
   • Enter add SparseArrays, LinearAlgebra, Clustering, NearestNeighbors, MAT, Distances, Metis, Arpack, Statistics, DelimitedFiles, StatsBase, Random, Debugger, Laplacians#master, LinearMaps, PyCall and wait for install to complete
   • Ctrl+C and Ctrl+D to exit the REPL.
7. In the container, install the following via pip
   • pip install pyjulia
   • pip install pycall
   • pip install julia
   • pip install hnswlib
8. Copy the contents of sgm-modulus to /modulus/modulus cp -r /sgm-modulus/* /modulus/modulus/ or run . /sgm-modulus/load-SGM-PINN-SPADE.sh
9. Due to a bug with PyJulia and multiprocessing, sometimes errors will be thrown at the end of training. This will be updated in a future release by implementing all code in python, but does not affect results.

For convenience a script 'dEnter' is included to enter the most recently started container.